tests: rename internal/livedns to livednstest and deny it outside tests (closes #164)

The live-DNS retry and concurrency limit is only for tests, but nothing
stopped program code from importing it and compiling it into the
binary. Its directory name now ends in test, and its import path is on
the test-support deny list in .golangci.yml, so make lint fails when
program code imports it. Every import and mention is updated to the new
name.

Model: opus-5-5

.golangci.yml

@@ -10,14 +10,20 @@ run:
 
 linters:
   default: all
+  enable:
+    # Successor to the deprecated gomodguard. Named explicitly, rather than
+    # left to `default: all`, because it carries the module policy below.
+    - gomodguard_v2
   disable:
     # Genuinely incompatible with project patterns
     - exhaustruct      # Requires all struct fields
-    - depguard         # Dependency allow/block lists
     - godot            # Requires comments to end with periods
-    - wsl              # Deprecated, replaced by wsl_v5
     - wrapcheck        # Too verbose for internal packages
     - varnamelen       # Short names like db, id are idiomatic Go
+    # Deprecated: the warning is attached to the old name, so it is
+    # silenced by disabling that name, not by enabling the successor.
+    - wsl              # Deprecated, replaced by wsl_v5
+    - gomodguard       # Deprecated, replaced by gomodguard_v2
   settings:
     lll:
       line-length: 88
@@ -28,6 +34,68 @@ linters:
       max-complexity: 15
     dupl:
       threshold: 100
+    depguard:
+      # Test-support code must not be compiled into the shipped binary. A
+      # test-support package exists to hand a test privileges the program
+      # itself must never have, so a file that is not a test must not import
+      # one. Test files, and the files inside a package whose directory name
+      # ends in `test`, are where that code belongs, and are exempt.
+      #
+      # The deny list below is the one part of this file a repository is
+      # expected to extend, and the only part it may. depguard matches an
+      # import path against a list of prefixes, so it cannot be told "any path
+      # whose last segment ends in test"; a repository's own test-support
+      # packages have to be named here one at a time, by full import path,
+      # under a module path that differs from repository to repository. Add
+      # them; change nothing else.
+      rules:
+        test-support:
+          list-mode: lax
+          files:
+            - "$all"
+            - "!$test"
+            - "!**/*test/**"
+          deny:
+            - pkg: net/http/httptest
+              desc: >-
+                Test-support code belongs in test files and in packages whose
+                directory name ends in test, not in the shipped binary.
+            - pkg: sneak.berlin/go/dnswatcher/internal/livednstest
+              desc: >-
+                Live-DNS test support belongs in test files and in packages
+                whose directory name ends in test, not in the shipped binary.
+    # Only decisions already recorded in the Go package defaults are
+    # listed here. Every entry matches the module path exactly.
+    gomodguard_v2:
+      blocked:
+        - module: github.com/rs/zerolog
+          recommendations:
+            - log/slog
+          reason: "Structured logging is stdlib log/slog."
+        # One entry per pre-fork module path, because the later releases
+        # are separate paths. A prefix match would be shorter but would
+        # also reach github.com/go-redis/redismock, the test double for
+        # the successor these entries recommend.
+        - module: github.com/go-redis/redis
+          recommendations:
+            - github.com/redis/go-redis/v9
+          reason: "Pre-fork module; use the maintained go-redis v9."
+        - module: github.com/go-redis/redis/v7
+          recommendations:
+            - github.com/redis/go-redis/v9
+          reason: "Pre-fork module; use the maintained go-redis v9."
+        - module: github.com/go-redis/redis/v8
+          recommendations:
+            - github.com/redis/go-redis/v9
+          reason: "Pre-fork module; use the maintained go-redis v9."
+        - module: github.com/sergi/go-diff
+          recommendations:
+            - github.com/aymanbagabas/go-udiff
+          reason: "No unified diff output; use go-udiff."
+        - module: github.com/hexops/gotextdiff
+          recommendations:
+            - github.com/aymanbagabas/go-udiff
+          reason: "Unmaintained fork; use go-udiff."
 
 issues:
   max-issues-per-linter: 0

Dockerfile

@@ -1,13 +1,10 @@
-# Build stage
-# golang 1.25-alpine, 2026-02-28
-FROM golang@sha256:f6751d823c26342f9506c03797d2527668d095b0a15f1862cddb4d927a7a4ced AS builder
-
-RUN apk add --no-cache git make gcc musl-dev binutils-gold
-
-# golangci-lint v2.12.2, 2026-08-07
-RUN go install github.com/golangci/golangci-lint/v2/cmd/golangci-lint@c0d3ddc9cf3faa61a4e378e879ece580256d76e5
-# goimports v0.42.0
-RUN go install golang.org/x/tools/cmd/goimports@009367f5c17a8d4c45a961a3a509277190a9a6f0
+# Lint stage - fast feedback on lint issues, before the build starts.
+# The linter is invoked directly rather than through `make lint`: that
+# target shells out to `docker build -f Dockerfile.lint`, and there is
+# no docker daemon inside a docker build.
+# script/cibuild and script/docker name this stage in --no-cache-filter.
+# golangci/golangci-lint:v2.12.2 (Debian-based), 2026-08-10
+FROM golangci/golangci-lint:v2.12.2@sha256:5cceeef04e53efe1470638d4b4b4f5ceefd574955ab3941b2d9a68a8c9ad5240 AS lint
 
 WORKDIR /src
 COPY go.mod go.sum ./
@@ -15,8 +12,27 @@ RUN go mod download
 
 COPY . .
 
-# Run all checks - build fails if any check fails
-RUN make check
+RUN make fmt-check
+RUN golangci-lint run --config .golangci.yml ./...
+
+# Build stage
+# script/cibuild and script/docker name this stage in --no-cache-filter.
+# golang 1.25-alpine, 2026-02-28
+FROM golang@sha256:f6751d823c26342f9506c03797d2527668d095b0a15f1862cddb4d927a7a4ced AS builder
+
+RUN apk add --no-cache git make gcc musl-dev binutils-gold
+
+# Force BuildKit to run the lint stage before proceeding
+COPY --from=lint /src/go.sum /dev/null
+
+WORKDIR /src
+COPY go.mod go.sum ./
+RUN go mod download
+
+COPY . .
+
+# Run the tests - build fails if any test fails
+RUN make test
 
 # Build the binary
 RUN make build
@@ -27,15 +43,33 @@ FROM alpine@sha256:c3f8e73fdb79deaebaa2037150150191b9dcbfba68b4a46d70103204c53f4
 
 RUN apk add --no-cache ca-certificates tzdata
 
-WORKDIR /app
+COPY --from=builder /src/bin/dnswatcher /usr/local/bin/dnswatcher
 
-COPY --from=builder /src/bin/dnswatcher /app/dnswatcher
-
-# Create data directory
-RUN mkdir -p /var/lib/dnswatcher
+# Run as an unprivileged user that owns the data directory. A fresh named
+# volume inherits this ownership; a bind-mounted host directory must be
+# owned by uid 10001 (see "Running under upaas" in README.md), or startup
+# fails.
+RUN addgroup -S -g 10001 dnswatcher \
+    && adduser -S -G dnswatcher -u 10001 dnswatcher \
+    && mkdir -p /var/lib/dnswatcher \
+    && chown dnswatcher:dnswatcher /var/lib/dnswatcher
 
 ENV DNSWATCHER_DATA_DIR=/var/lib/dnswatcher
 
+# Config loading also reads a `.env` file and a file named `dnswatcher`
+# (any config extension, or none) from the working directory. `/` holds
+# neither, so every setting comes from the environment. Do not make the
+# data directory, or the binary's directory, the working directory.
+WORKDIR /
+
+USER dnswatcher
+
 EXPOSE 8080
 
-ENTRYPOINT ["/app/dnswatcher"]
+# busybox wget (already in alpine) probes the health endpoint every 10
+# seconds, so the container is healthy well before upaas reads its health
+# 60 seconds after a deploy and fails the deploy unless it is healthy.
+HEALTHCHECK --interval=10s --timeout=5s --start-period=10s --retries=3 \
+    CMD wget -q -O /dev/null "http://127.0.0.1:${PORT:-8080}/.well-known/healthcheck" || exit 1
+
+ENTRYPOINT ["/usr/local/bin/dnswatcher"]

Dockerfile.lint

@@ -0,0 +1,29 @@
+# Lint-only image: used by script/lint. golangci-lint is never run on
+# the host — the repo is COPYed into the build context and the linter
+# runs as a build step, so a successful build IS a clean lint. This
+# also works where the docker daemon is remote and bind mounts are
+# impossible.
+#
+# `golangci-lint config verify` is deliberately NOT run here: it
+# fetches its JSON schema over a live, unpinned HTTPS call, which would
+# make linting network-dependent and defeat hash-pinning. The cost of
+# that: unknown top-level keys in .golangci.yml are silently ignored,
+# so a mistyped or wrong-schema key lints clean while applying nothing.
+#
+# golangci/golangci-lint:v2.12.2 (Debian-based), 2026-08-10
+FROM golangci/golangci-lint:v2.12.2@sha256:5cceeef04e53efe1470638d4b4b4f5ceefd574955ab3941b2d9a68a8c9ad5240 AS deps
+
+WORKDIR /src
+
+# Dependencies first, so this stage stays cached across lint runs.
+COPY go.mod go.sum ./
+RUN go mod download
+
+# Everything below is invalidated on every run by the
+# --no-cache-filter=lint that script/lint passes: caching is explicitly
+# waived for linting, and a cached build lints nothing.
+FROM deps AS lint
+
+COPY . .
+
+RUN golangci-lint run --config .golangci.yml ./...

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 sneak
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -1,6 +1,6 @@
 # dnswatcher
 
-dnswatcher is a pre-1.0 Go daemon by [@sneak](https://sneak.berlin) that monitors DNS records, TCP port availability, and TLS certificates, delivering real-time change notifications via Slack, Mattermost, and ntfy webhooks.
+dnswatcher is an MIT-licensed, pre-1.0 Go daemon by [@sneak](https://sneak.berlin) that monitors DNS records, TCP port availability, and TLS certificates, delivering real-time change notifications via Slack, Mattermost, and ntfy webhooks.
 
 > ⚠️ Pre-1.0 software. APIs, configuration, and behavior may change without notice.
 
@@ -61,6 +61,10 @@ rejected.
   record types: A, AAAA, CNAME, MX, TXT, SRV, CAA, NS.
 - Stores results **per nameserver**. The state for a hostname is not a
   merged view — it is a map from nameserver to record set.
+- DNS names inside record values (CNAME, MX, SRV and NS targets) are
+  stored in lower case, because names are case-insensitive and
+  nameservers may answer in any letter case. TXT and CAA values keep
+  their letter case; they are not lower-cased.
 - Any observable change in any nameserver's response triggers a
   notification. This includes:
   - **Record change**: A nameserver returns different records than it
@@ -70,8 +74,15 @@ rejected.
     This is distinct from "responded with no records."
   - **NS recovery**: A previously-unreachable nameserver starts
     responding again.
-  - **Inconsistency detected**: Two nameservers that previously agreed
-    now return different record sets for the same hostname.
+  - **Inconsistency detected**: Two nameservers return different record
+    sets for the same hostname and did not already differ on the previous
+    check. Every pair of nameservers is compared. The alert is sent once
+    for each such pair, on the check where they start to disagree, and not
+    again while they keep disagreeing, including after a restart. A
+    nameserver that was not in the previous check (newly added, or back
+    after dropping out) and answers differently is reported on the check
+    where it appears. If a pair agrees again and later disagrees, the
+    alert is sent again.
 
 ### TCP Port Monitoring
 
@@ -182,6 +193,67 @@ dnswatcher exposes a lightweight HTTP API for operational visibility:
 | `GET /api/v1/status`                  | Current monitoring state       |
 | `GET /metrics`                        | Prometheus metrics (optional)  |
 
+#### Server timeouts
+
+The HTTP server sets all four socket-level timeouts. These are compile-time
+constants in `internal/server/server.go`, not configurable via environment
+variables.
+
+| Timeout             | Value | Purpose                                       |
+|---------------------|-------|-----------------------------------------------|
+| `ReadHeaderTimeout` | 10s   | Bounds the request header read (slowloris)    |
+| `ReadTimeout`       | 15s   | Bounds the whole request read, headers + body |
+| `WriteTimeout`      | 75s   | Bounds handler execution plus response flush  |
+| `IdleTimeout`       | 120s  | Reaps idle keep-alive connections             |
+
+These are distinct from the 60s per-request handler budget applied by
+`chimw.Timeout` in `internal/server/routes.go`, which cancels the request
+context but does not touch the socket. `WriteTimeout` is deliberately
+larger than that budget: the write deadline is armed once request headers
+are read, so a smaller value would sever the connection before a handler
+using its full budget could respond. `IdleTimeout` exceeds common
+Prometheus scrape intervals so the scraper reuses its connection.
+
+### Security Headers
+
+Every response — the dashboard, the static assets under `/s/...`, the
+healthchecks, the JSON API, and `/metrics` — carries the following
+headers, set by a global middleware:
+
+| Header                      | Value                                 |
+|-----------------------------|---------------------------------------|
+| `Strict-Transport-Security` | `max-age=31536000; includeSubDomains` |
+| `Content-Security-Policy`   | see below                             |
+| `X-Frame-Options`           | `DENY`                                |
+| `X-Content-Type-Options`    | `nosniff`                             |
+| `Referrer-Policy`           | `no-referrer`                         |
+| `Permissions-Policy`        | all unused browser features denied    |
+
+The content security policy is:
+
+```
+default-src 'self'; script-src 'none'; style-src 'self'; img-src 'self';
+font-src 'none'; connect-src 'none'; object-src 'none'; base-uri 'none';
+form-action 'none'; frame-ancestors 'none'
+```
+
+The dashboard ships no JavaScript (the 30-second refresh is a
+`<meta http-equiv="refresh">`), no inline styles, no inline event
+handlers, and no images; its only subresource is the embedded stylesheet
+at `/s/css/tailwind.min.css`, which `style-src 'self'` permits. The
+policy therefore needs neither `unsafe-inline` nor `unsafe-eval`.
+`frame-ancestors 'none'` is the primary anti-framing control, with
+`X-Frame-Options: DENY` retained as the legacy fallback.
+
+HSTS is emitted unconditionally, including over plain HTTP. dnswatcher is
+expected to run behind a TLS-terminating reverse proxy, and the browser
+must still be told to enforce HTTPS end to end, so the header is never
+gated on whether the request itself arrived over TLS.
+
+`Referrer-Policy: no-referrer` is stricter than the
+`strict-origin-when-cross-origin` baseline: the dashboard has no
+cross-origin navigation needs, and its URL may name internal hosts.
+
 ---
 
 ## Architecture
@@ -194,7 +266,8 @@ internal/
   globals/globals.go           Build-time variables (version)
   logger/logger.go             slog structured logging (TTY detection)
   healthcheck/healthcheck.go   Health check service
-  middleware/middleware.go      HTTP middleware (logging, CORS, metrics auth)
+  middleware/middleware.go      HTTP middleware (logging, CORS, security
+                               headers, metrics auth)
   handlers/handlers.go         HTTP request handlers
   server/
     server.go                  HTTP server lifecycle
@@ -205,6 +278,8 @@ internal/
   tlscheck/tlscheck.go        TLS certificate inspector
   notify/notify.go             Notification service (Slack, Mattermost, ntfy)
   watcher/watcher.go           Main monitoring orchestrator and scheduler
+  livednstest/livednstest.go   Retry and concurrency limit for tests
+                               against live DNS (imported only by tests)
 ```
 
 ### Design Principles
@@ -218,7 +293,8 @@ internal/
 - **Structured logging**: All logs use `log/slog` with JSON output in
   production (TTY detection for development).
 - **Graceful shutdown**: All background goroutines respect context
-  cancellation and the fx lifecycle.
+  cancellation and the fx lifecycle. In-flight notification deliveries
+  are drained on shutdown, bounded by the shutdown timeout.
 
 ---
 
@@ -380,20 +456,35 @@ standard: normalized scripts in `script/` are the entrypoints for the
 development workflow, and the Makefile targets are thin shims that call
 them. We provide:
 
-- `script/bootstrap` — install all dependencies (go, pinned
-  golangci-lint and goimports, `go mod download`)
+- `script/bootstrap` — install all dependencies (go, pinned goimports,
+  `go mod download`). It does not install golangci-lint: see
+  `script/lint` below.
 - `script/setup` — make a fresh clone ready for development: bootstrap
   plus the git pre-commit hook
 - `script/projectname` — print the project name (used for the Docker
   image tag)
-- `script/test` — run the test suite (race detector, coverage)
-- `script/lint` — run golangci-lint
+- `script/test` — run the test suite (race detector, coverage). Caching
+  is waived for testing, exactly as it is for linting: `-count=1`
+  forces every invocation to execute, because the suite queries live
+  DNS and a cached pass queries nothing. Failures are rerun with `-v`
+  automatically, and the build fails even if that rerun passes.
+- `script/lint` — run golangci-lint, always inside Docker: it builds
+  `Dockerfile.lint`, which COPYs the repo into the digest-pinned
+  `golangci-lint` image and lints as a build step, so a successful
+  build is a clean lint. The linter is never installed or run on the
+  host, and Docker is the only prerequisite. Caching is waived for
+  linting: the lint stage is forced to execute on every run with
+  `--no-cache-filter`, because a cached build lints nothing.
 - `script/fmt` — format all code (gofmt -s, goimports)
 - `script/fmt-check` — check formatting (read-only)
 - `script/check` — run test, lint, and fmt-check
-- `script/docker` — build the Docker image tagged via
-  `script/projectname`
-- `script/cibuild` — CI entrypoint: plain `docker build .`
+- `script/docker` — build the Docker image tagged via `script/projectname`, with
+  `--no-cache-filter=lint,builder` so the lint stage and the builder stage,
+  which runs the tests, run on every invocation
+- `script/cibuild` — CI entrypoint: `docker build` with
+  `--no-cache-filter=lint,builder`, so the lint stage and the builder stage,
+  which runs the tests, run on every invocation, because a cached build lints
+  nothing and queries no DNS
 - `script/precommit` — run by the git pre-commit hook; `go mod tidy`
   guard, then `script/check`
 - `script/install-precommit` — install the git pre-commit hook
@@ -403,7 +494,7 @@ them. We provide:
 ```sh
 make build      # Build binary to bin/dnswatcher
 make test       # Run tests with race detector
-make lint       # Run golangci-lint
+make lint       # Run golangci-lint in Docker (requires docker)
 make fmt        # Format code
 make check      # Run all checks (test, lint, fmt-check)
 make clean      # Remove build artifacts
@@ -434,11 +525,57 @@ docker run -d \
 
 ---
 
+## Running under upaas
+
+[upaas](https://git.eeqj.de/sneak/upaas) builds the image from this
+repository's `Dockerfile` and runs it. The app needs:
+
+- **Branch:** `prod`. `prod` is cut from `main`, and merging a `main` to
+  `prod` pull request is a deploy.
+- **Volume:** one host directory mounted at `/var/lib/dnswatcher`, where
+  the state file lives. upaas bind-mounts the host path it is given and
+  does not create it. The container runs as uid 10001 and does not start
+  unless it can write there. Create the directory before the first
+  deploy:
+
+  ```sh
+  mkdir -p /path/to/data
+  chown 10001:10001 /path/to/data
+  chmod 700 /path/to/data
+  ```
+
+- **Network and port:** the dashboard is unauthenticated and shows every
+  watched name and recent alert, and upaas publishes every mapped port on
+  all interfaces of the host
+  ([upaas issue 113](https://git.eeqj.de/sneak/upaas/issues/113)). Add a
+  port mapping to container port `8080` only if the dashboard should be
+  public. Otherwise add none: set the app's Docker network in upaas to
+  your reverse proxy's Docker network, and the proxy reaches the app at
+  `upaas-` followed by the app name, port `8080`.
+- **Required environment:** `DNSWATCHER_TARGETS`, a comma-separated list
+  of the domains and hostnames to watch. dnswatcher refuses to start
+  without it.
+- **Recommended environment:** at least one notification endpoint
+  (`DNSWATCHER_SLACK_WEBHOOK`, `DNSWATCHER_MATTERMOST_WEBHOOK`,
+  `DNSWATCHER_NTFY_TOPIC`); without one, changes show only on the
+  dashboard. `DNSWATCHER_METRICS_USERNAME` and
+  `DNSWATCHER_METRICS_PASSWORD` serve `/metrics` behind basic auth.
+- **Leave unset:** `DNSWATCHER_DATA_DIR`, which the image sets to
+  `/var/lib/dnswatcher`, and `PORT`, which defaults to `8080`. Every
+  setting comes from the environment; the image holds no config file.
+- **Health check:** the image's own, which requests
+  `/.well-known/healthcheck` every 10 seconds. upaas reads the
+  container's health 60 seconds after a deploy and marks the deploy
+  failed unless it is `healthy`.
+
+---
+
 ## Monitoring Lifecycle
 
-1. **Startup**: Load state from disk. If no state file exists, start
-   with empty state (first check will establish baseline without
-   triggering change notifications).
+1. **Startup**: Check that the data directory can be written, and exit
+   with an error naming it if not. Load state from disk. If no state
+   file exists, start with empty state (first check will establish
+   baseline without triggering change notifications).
 2. **Initial check**: Immediately perform all DNS, port, and TLS checks
    on startup.
 3. **Periodic checks** (DNS always runs first):
@@ -452,8 +589,14 @@ docker run -d \
      from a previous cycle.
 4. **On change detection**: Send notifications to all configured
    endpoints, update in-memory state, persist to disk.
-5. **Shutdown**: Persist final state to disk, complete in-flight
-   notifications, stop gracefully.
+5. **Shutdown**: Persist final state to disk, wait for in-flight
+   notification deliveries to complete, stop gracefully. The wait is
+   bounded by the fx shutdown timeout (15s by default): deliveries still
+   retrying against an unreachable endpoint when that expires are
+   abandoned, and the number abandoned is logged at warn level rather
+   than dropped silently. Notifications generated after shutdown has
+   begun are refused and logged, so a late burst cannot extend the
+   shutdown.
 
 ---
 
@@ -475,8 +618,9 @@ Viper for configuration.
 
 ## License
 
-License has not yet been chosen for this project. Pending decision by the
-author (MIT, GPL, or WTFPL).
+dnswatcher is released under the MIT License, Copyright (c) 2026
+[@sneak](https://sneak.berlin). See the [`LICENSE`](./LICENSE) file in the
+repository root for the full text.
 
 ## Author
 

REPO_POLICIES.md

@@ -1,6 +1,6 @@
 ---
 title: Repository Policies
-last_modified: 2026-07-06
+last_modified: 2026-08-07
 ---
 
 This document covers repository structure, tooling, and workflow standards. Code
@@ -189,8 +189,13 @@ style conventions are in separate documents:
   module under test to verify it compiles/parses. There is no excuse for
   `make test` to be a no-op.
 
-- `make test` must complete in under 20 seconds. Add a 30-second timeout in the
-  Makefile.
+- `make test` must complete in under 60 seconds. That is the hard cap, and a
+  suite that exceeds it fails. Under 20 seconds is the target. A suite between
+  20 and 60 seconds is still green, but the overage must be filed as an
+  improvement bug against that repo. Add a 90-second timeout to the test
+  invocation in the Makefile (`go test -timeout 90s`). The backstop deliberately
+  sits above the hard cap so that it catches a genuinely hung test rather than a
+  merely slow one.
 
 - **`make test` should use the conditional verbose rerun pattern.** Run tests
   without `-v` (verbose) first. If tests fail, automatically rerun with `-v` to
@@ -209,9 +214,9 @@ style conventions are in separate documents:
 
     ```makefile
     test:
-    	@go test -timeout 30s -race -cover ./... || \
+    	@go test -timeout 90s -race -cover ./... || \
     		{ echo "--- Rerunning with -v for details ---"; \
-    		  go test -timeout 30s -race -v ./...; exit 1; }
+    		  go test -timeout 90s -race -v ./...; exit 1; }
     ```
 
     Python example:
@@ -260,7 +265,10 @@ style conventions are in separate documents:
 
 - `.golangci.yml` is standardized and must _NEVER_ be modified by an agent, only
   manually by the user. Fetch from
-  `https://git.eeqj.de/sneak/prompts/raw/branch/main/.golangci.yml`.
+  `https://git.eeqj.de/sneak/prompts/raw/branch/main/.golangci.yml`. The
+  canonical golangci-lint version is v2.12.2 (released 2026-05-06), installed
+  commit-pinned via
+  `go install github.com/golangci/golangci-lint/v2/cmd/golangci-lint@c0d3ddc9cf3faa61a4e378e879ece580256d76e5`.
 
 - When pinning images or packages by hash, add a comment above the reference
   with the version and date (YYYY-MM-DD).

TESTING.md

@@ -2,8 +2,12 @@
 
 ## DNS Resolution Tests
 
-All resolver tests **MUST** use live queries against real DNS servers.
-No mocking of the DNS client layer is permitted.
+DNS is never mocked in this project, not in tests and not anywhere
+else; see the README section "No DNS mocking. Ever." Every test that
+looks something up in DNS **MUST** query live DNS servers, never a
+stand-in. Logic that works on record data, such as comparing or
+formatting records, may be tested on that data directly with no
+lookup.
 
 ### Rationale
 
@@ -17,18 +21,25 @@ real servers ensures the resolver works correctly in production.
 
 - Tests hit real DNS infrastructure and require network access
 - Test duration depends on network conditions; timeout tuning keeps
-  the suite within the 30-second target
+  the suite within the 60-second target
 - Query timeout is calibrated to 3× maximum antipodal RTT (~300ms)
   plus processing margin
 - Root server fan-out is limited to reduce parallel query load
+- Live lookups that expect an answer go through `internal/livednstest`,
+  which limits how many run at once in a test binary and retries a
+  lookup that got none
 - Flaky failures from transient network issues are acceptable and
   should be investigated as potential resolver bugs, not papered over
   with mocks or skip flags
 
 ### What NOT to do
 
-- **Do not mock `DNSClient`** for resolver tests (the mock constructor
-  exists for unit-testing other packages that consume the resolver)
+- **Do not mock, fake or stub DNS** anywhere: no stand-in `DNSClient`,
+  no stand-in for the watcher's `DNSResolver`, no fake DNS server, no
+  canned responses
 - **Do not add `-short` flags** to skip slow tests
 - **Do not increase `-timeout`** to hide hanging queries
+- **Do not remove `-count=1` from `script/test`** — Go's test cache
+  replays a previous run's output without querying anything, so a
+  cached pass is not evidence that live resolution works
 - **Do not modify linter configuration** to suppress findings

TODO.md

@@ -10,28 +10,148 @@
 
 # Status
 
-pre-1.0. No git tags. Core resolver work in flight on feature/resolver
-(dirty: internal/resolver/resolver_test.go). Local checkout has diverged
-from origin: origin/main is 8 commits ahead (watcher orchestrator,
-unified TARGETS) and origin/feature/resolver already contains the full
-iterative resolver implementation with hermetic mocked tests.
+pre-1.0. No git tags.
 
 # Next Step
 
-Policy scaffold commit: add LICENSE, REPO_POLICIES.md, .editorconfig,
-.dockerignore, and .gitea/workflows/check.yml, and add the missing
-fmt-check, docker, and hooks targets to the Makefile. One commit, then
-confirm make check still passes.
+Add the README sections required by policy (Description, Getting Started,
+Rationale, Design, TODO, License, Author) if any are still missing.
 
 # Completed Steps
 
-- 2026-08-09: `script/bootstrap` now installs the pinned `golangci-lint`
-  and `goimports` unconditionally instead of only when the binary is
-  absent from `PATH`, so the commit pins actually take effect on
-  already-provisioned machines; it also warns when `PATH` resolves
-  either tool to a copy outside the directory `go install` writes to.
-  The `missing` presence check is retained for `git`, `make`, and `go`
-  (#117)
+- 2026-09-29: the live-DNS test package is renamed `internal/livednstest` and
+  added to the `test-support` `deny` list in `.golangci.yml`, so `make lint`
+  fails when program code imports it (closes #164).
+- 2026-09-29: `.golangci.yml` re-fetched unchanged from `sneak/prompts`. It
+  replaces the deprecated `gomodguard` with `gomodguard_v2`, so `make lint` no
+  longer warns about it, and turns on `depguard` with the org `test-support`
+  rule, which rejects `net/http/httptest` except in test files and in files
+  under a directory whose name ends in `test`. This repo had no `deny` entries
+  of its own to carry forward (closes #123).
+- 2026-09-29: nothing stands in for DNS any more. Watcher tests that look
+  something up in DNS use the real resolver against live DNS servers and test
+  record and nameserver changes by preparing the saved state a check starts
+  from; the resolver timeout test queries an address that never answers, and
+  `NewFromLoggerWithClient`, used only by its stand-in client, is gone. The
+  live-DNS retry and concurrency limit moved to `internal/livednstest`, which
+  both test packages use. `TESTING.md` states the README's rule (closes #159).
+- 2026-09-28: the inconsistency alert is sent once, on the check where two
+  nameservers start to disagree or where a nameserver that disagrees first
+  appears, instead of on every check while they disagree, and not again after
+  a restart. Every pair of nameservers is compared, not only neighbours in
+  sorted order of name (closes #158).
+- 2026-09-28: DNS names in record values (CNAME, MX, SRV and NS targets) are
+  lower-cased, so nameservers that answer in different letter case no longer
+  count as inconsistent or as a record change (closes #157).
+- 2026-09-28: `script/cibuild` and `script/docker` now pass
+  `--no-cache-filter=lint,builder` so lint and tests run every build (closes
+  #115).
+- 2026-09-28: the server timeout test now drives `Run` and checks the
+  `http.Server` it serves carries the timeouts; corrected the `ReadTimeout`
+  note in that test (closes #120).
+- 2026-09-28: upaas deploy readiness — runtime image runs as unprivileged
+  `dnswatcher`, Docker `HEALTHCHECK`, startup fails when the data directory is
+  not writable, README "Running under upaas" (closes #147).
+- 2026-09-21: added behavioural tests for `internal/globals`,
+  `internal/healthcheck`, and `internal/logger` (closes #110).
+- 2026-09-21: `go mod tidy` dropped the redundant `golang.org/x/sync`
+  `// indirect` line so `script/bootstrap` leaves a clean tree (#132)
+- 2026-08-10: comment-only corrections to `script/bootstrap`,
+  `script/cibuild`, and `Dockerfile.lint`. The `goimports` pin in
+  `script/bootstrap` was justified by a claim that `script/fmt-check`
+  runs it on the host; it does not (it runs `gofmt -l .` only), so the
+  header now credits `script/fmt` alone. `script/cibuild` still claimed
+  the `Dockerfile` runs `make check`, which stopped being true when
+  linting moved to its own stage; it now describes the lint stage
+  (`make fmt-check` plus `golangci-lint`) and the builder stage
+  (`make test`, `make build`). The `docker`-missing warning in
+  `script/bootstrap` reads as one sentence instead of three fragments
+  each re-prefixed with `bootstrap:`. `Dockerfile.lint` now records the
+  residual risk of omitting `golangci-lint config verify`: unknown
+  top-level keys in `.golangci.yml` are silently ignored, so a mistyped
+  key lints clean while applying nothing. No behaviour changed
+- 2026-08-10: MIT `LICENSE` added at the repository root, closing the
+  last gap in `REPO_POLICIES.md`'s required-minimum file list and
+  removing the all-rights-reserved default that would otherwise have
+  shipped with a 1.0 tag. The licence choice is the standing org policy
+  (any public repo lacking a licence gets MIT; a private repo with no
+  licence is already all-rights-reserved), and this repo is public. The
+  file holds the canonical MIT text byte-for-byte with only the
+  copyright line filled in (`Copyright (c) 2026 sneak`); no clauses were
+  added, removed, or reflowed. `README.md`'s first line now names the
+  licence, as the Description requirement demands, and the License
+  section states MIT and points at the file instead of saying the choice
+  is pending. `make fmt` covers only Go sources (`gofmt -s`,
+  `goimports`), so it cannot reflow `LICENSE`
+- 2026-08-10: the policy scaffold (`REPO_POLICIES.md`, `.editorconfig`,
+  `.dockerignore`, `.gitea/workflows/check.yml`, and the `fmt-check`,
+  `docker`, and hooks Makefile targets) is present; it landed piecemeal
+  across the scripts-to-rule-them-all and policy commits rather than as
+  the single commit this file once planned
+- 2026-08-10: Go's test cache disabled for `script/test` via `-count=1`,
+  so every invocation actually executes. A cached pass replays an
+  earlier run's output without querying DNS at all, which in this repo
+  means the suite's entire premise goes unexercised while the run
+  reports green in under a second. The conditional verbose rerun that
+  `REPO_POLICIES.md` mandates was added at the same time (the primary
+  run had been unconditionally `-v`): quiet first, `-v` only on
+  failure, `-count=1` on both, and exit 1 forced regardless of the
+  rerun's result so a flake passing the second time cannot turn the
+  build green. `-timeout 90s` left alone as the deliberate backstop
+  above the 60s hard cap. Uncached suite runs ~4s, well inside the 20s
+  target
+- 2026-08-10: live-DNS test flakiness addressed by robustness rather
+  than gating, per the owner's ruling on #93: new
+  `internal/resolver/livedns_test.go` adds a package-wide concurrency
+  gate (so parallel tests stop bursting at the first root server),
+  retry with exponential backoff on transport failures only, and
+  quorum instead of unanimity for multi-nameserver assertions. Quorum
+  tolerates silence only: every per-nameserver status must be in a
+  closed allowlist (`ok`/`timeout`/`error`, or
+  `nxdomain`/`timeout`/`error`), so a wrong answer from a minority —
+  `nodata` today, any status added later — fails the test instead of
+  sliding through under the majority. The
+  `make test` cap moved to the new org-wide 60s hard cap / 20s target
+  with a 90s `-timeout` backstop; `REPO_POLICIES.md` re-vendored
+  byte-identical from `sneak/prompts`. No mocks, no `-short`, no build
+  tags, no skips, and no change to production resolver behaviour
+- 2026-08-10: all linting moved into Docker: new root `Dockerfile.lint`
+  on the digest-pinned `golangci/golangci-lint:v2.12.2` image,
+  `script/lint` reduced to a thin wrapper that builds it with
+  `--no-cache-filter=lint` so the linter actually executes every run,
+  golangci-lint install dropped from `script/bootstrap` (goimports
+  stays, `script/fmt` needs it on the host), and the root `Dockerfile`
+  given its own lint stage so its build no longer recurses through
+  `make check` into `script/lint`. `golangci-lint config verify` is
+  deliberately omitted: it fetches its schema over an unpinned live
+  HTTPS call
+- 2026-08-09: in-flight notification deliveries are now drained at
+  shutdown (#106): `notify.New` registers an fx `OnStop` hook that waits
+  on a `sync.WaitGroup` of tracked delivery goroutines, bounded by the
+  `OnStop` context; on expiry the outstanding count is logged at warn
+  level and parked retry backoffs are released instead of being dropped
+  silently, and deliveries submitted after the drain begins are refused
+  so shutdown cannot be extended indefinitely; an `OnStop` context that
+  is already expired on entry with nothing outstanding drains quietly
+  rather than warning about deliveries that were never abandoned
+- 2026-08-09: `http.Server` now sets all four socket-level timeouts
+  (`ReadTimeout` 15s, `ReadHeaderTimeout` 10s, `WriteTimeout` 75s,
+  `IdleTimeout` 120s) as named constants in `internal/server/server.go`,
+  closing the slowloris / unreaped-keep-alive exposure required by
+  `REPO_POLICIES.md` before 1.0; `WriteTimeout` is deliberately greater
+  than the 60s `chimw.Timeout` handler budget so that budget stays
+  reachable, and tests in `internal/server` pin both the non-zero
+  values and that relationship (#99)
+- 2026-08-09: security response headers middleware
+  (`SecurityHeaders()` in `internal/middleware/middleware.go`)
+  registered globally in `internal/server/routes.go`, so HSTS, CSP,
+  `X-Frame-Options`, `X-Content-Type-Options`, `Referrer-Policy`, and
+  `Permissions-Policy` are set on every response including `/s/...` and
+  `/metrics`; the CSP needs no `unsafe-inline`/`unsafe-eval` because the
+  dashboard ships no JavaScript and no inline styles; HSTS is emitted
+  unconditionally per policy (TLS-terminating proxy in front). Remaining
+  1.0 hardening items — `http.Server` timeouts, request body limits,
+  rate limiting, CORS scoping — are tracked separately
 - 2026-08-07: golangci-lint bumped to v2.12.2 (commit-pinned installs
   in `Dockerfile` and `script/bootstrap`); `.golangci.yml` set to the
   org-standard v2-schema config used across the org's repos
@@ -60,8 +180,6 @@ confirm make check still passes.
 
 Compliance:
 
-- Add README sections required by policy (Description, Getting Started,
-  Rationale, Design, TODO, License, Author) if any are missing
 - Pin Dockerfile base images by sha256 and ensure the Docker build runs
   make check
 
@@ -155,6 +273,5 @@ Infrastructure notes (from untracked TODO.md):
 - Module path sneak.berlin/go/dnswatcher differs from the git.eeqj.de
   remote intentionally; do not "fix" it
 - Dependencies: github.com/miekg/dns, golang.org/x/net/publicsuffix
-- Resolver tests originally used live DNS against *.dns.sneak.cloud
-  (required records documented in the test file header); origin now has
-  mocked hermetic tests, keep them hermetic
+- DNS is never mocked; tests that look something up in DNS query live DNS
+  servers (README, "No DNS mocking. Ever.")

go.mod

@@ -40,7 +40,6 @@ require (
 	go.yaml.in/yaml/v2 v2.4.2 // indirect
 	go.yaml.in/yaml/v3 v3.0.4 // indirect
 	golang.org/x/mod v0.32.0 // indirect
-	golang.org/x/sync v0.19.0 // indirect
 	golang.org/x/sys v0.41.0 // indirect
 	golang.org/x/text v0.34.0 // indirect
 	golang.org/x/tools v0.41.0 // indirect

internal/globals/globals_test.go

@@ -0,0 +1,50 @@
+package globals_test
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"sneak.berlin/go/dnswatcher/internal/globals"
+)
+
+// TestGlobals exercises the package-level version and appname
+// variables through their setters and read-back via New. These are
+// shared package state, so the test mutates a global and must run
+// sequentially; it cannot use t.Parallel().
+//
+//nolint:paralleltest // mutates shared package-level globals, must run sequentially
+func TestGlobals(t *testing.T) {
+	versions := []string{"v1.2.3", "dev", "", "v1.2.3-4-gabcdef"}
+	for _, want := range versions {
+		globals.SetVersion(want)
+
+		g, err := globals.New(nil)
+		require.NoError(t, err)
+		assert.Equal(t, want, g.Version,
+			"New must surface the version set by SetVersion")
+	}
+
+	names := []string{"dnswatcher", "other", ""}
+	for _, want := range names {
+		globals.SetAppname(want)
+
+		g, err := globals.New(nil)
+		require.NoError(t, err)
+		assert.Equal(t, want, g.Appname,
+			"New must surface the appname set by SetAppname")
+	}
+
+	// New returns a snapshot: a later SetVersion must not mutate a
+	// Globals handed out earlier.
+	globals.SetVersion("first")
+
+	g, err := globals.New(nil)
+	require.NoError(t, err)
+
+	globals.SetVersion("second")
+	assert.Equal(t, "first", g.Version,
+		"a Globals returned by New must not change when the "+
+			"package variable is set again")
+}

internal/healthcheck/healthcheck_test.go

@@ -0,0 +1,134 @@
+package healthcheck_test
+
+import (
+	"context"
+	"encoding/json"
+	"testing"
+	"time"
+
+	"go.uber.org/fx"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"sneak.berlin/go/dnswatcher/internal/config"
+	"sneak.berlin/go/dnswatcher/internal/globals"
+	"sneak.berlin/go/dnswatcher/internal/healthcheck"
+	"sneak.berlin/go/dnswatcher/internal/logger"
+)
+
+// recordingLifecycle is a minimal fx.Lifecycle that records the hooks
+// appended to it, so healthcheck.New can be exercised through its real
+// constructor without standing up a whole fx application.
+type recordingLifecycle struct {
+	hooks []fx.Hook
+}
+
+func (l *recordingLifecycle) Append(hook fx.Hook) {
+	l.hooks = append(l.hooks, hook)
+}
+
+// newHealthcheck builds a Healthcheck through the real constructor and
+// runs the registered OnStart hook so StartupTime is set the same way
+// the fx lifecycle would set it.
+func newHealthcheck(
+	t *testing.T,
+	maintenance bool,
+	version string,
+) *healthcheck.Healthcheck {
+	t.Helper()
+
+	g := &globals.Globals{Appname: "dnswatcher", Version: version}
+
+	log, err := logger.New(nil, logger.Params{Globals: g})
+	require.NoError(t, err)
+
+	lifecycle := &recordingLifecycle{}
+
+	hc, err := healthcheck.New(lifecycle, healthcheck.Params{
+		Globals: g,
+		Config:  &config.Config{MaintenanceMode: maintenance},
+		Logger:  log,
+	})
+	require.NoError(t, err)
+
+	require.Len(t, lifecycle.hooks, 1,
+		"New must register exactly one lifecycle hook")
+	require.NotNil(t, lifecycle.hooks[0].OnStart)
+	require.NoError(t, lifecycle.hooks[0].OnStart(context.Background()))
+
+	return hc
+}
+
+func TestCheckStatusAndPayloadShape(t *testing.T) {
+	t.Parallel()
+
+	hc := newHealthcheck(t, false, "v9.9.9")
+	resp := hc.Check()
+
+	assert.Equal(t, "ok", resp.Status)
+
+	// The JSON shape and field names are part of the contract for the
+	// /health and /.well-known/healthcheck routes, so assert on the
+	// exact set of keys the response marshals to.
+	raw, err := json.Marshal(resp)
+	require.NoError(t, err)
+
+	var fields map[string]json.RawMessage
+	require.NoError(t, json.Unmarshal(raw, &fields))
+
+	wantKeys := []string{
+		"status",
+		"now",
+		"uptimeSeconds",
+		"uptimeHuman",
+		"version",
+		"appname",
+		"maintenanceMode",
+	}
+	assert.Len(t, fields, len(wantKeys),
+		"response must marshal to exactly the documented fields")
+
+	for _, key := range wantKeys {
+		assert.Contains(t, fields, key, "missing JSON field %q", key)
+	}
+
+	// The Now field is documented as RFC3339Nano; a change to the
+	// format constant should turn this red.
+	_, err = time.Parse(time.RFC3339Nano, resp.Now)
+	assert.NoError(t, err, "Now must be RFC3339Nano")
+}
+
+func TestCheckMaintenanceModeReflectsConfig(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name        string
+		maintenance bool
+	}{
+		{"maintenance off", false},
+		{"maintenance on", true},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			hc := newHealthcheck(t, tt.maintenance, "test")
+			resp := hc.Check()
+			assert.Equal(t, tt.maintenance, resp.Maintenance,
+				"maintenanceMode must mirror Config.MaintenanceMode")
+		})
+	}
+}
+
+func TestCheckSurfacesVersionAndAppname(t *testing.T) {
+	t.Parallel()
+
+	hc := newHealthcheck(t, false, "surfaced-version-123")
+	resp := hc.Check()
+
+	assert.Equal(t, "surfaced-version-123", resp.Version,
+		"version from globals must appear in the payload")
+	assert.Equal(t, "dnswatcher", resp.Appname)
+}

internal/livednstest/livednstest.go

@@ -0,0 +1,121 @@
+// Package livednstest runs the live DNS operations of tests. Tests that
+// look something up in DNS query live DNS servers, never a stand-in —
+// see TESTING.md. Nothing here mocks, fakes, stubs, records or replays
+// DNS, and nothing here skips a test: it only changes *how* the live
+// queries are issued, so that a single dropped UDP packet or one slow
+// authoritative server does not turn correct code into a red build.
+//
+// Two mechanisms:
+//
+//  1. Bounded concurrency. Tests run in parallel and the build hosts
+//     have many cores, so without a limit every test starts its own
+//     iterative resolution at the same instant and they all hit the
+//     first root server within a few milliseconds of each other. Root
+//     servers rate-limit that, which shows up as a different arbitrary
+//     subset of tests failing on each run. Run caps how many live
+//     operations are in flight at once in one test binary.
+//
+//  2. Retry with exponential backoff. Each live operation gets several
+//     attempts with its own timeout. An attempt is retried when it
+//     obtained nothing to check, never because of what the test
+//     asserts about the result, so a wrong result still fails on the
+//     first attempt. A fault in the code under test that leaves
+//     nothing to check looks the same as live DNS not answering, and
+//     fails only after the last attempt.
+package livednstest
+
+import (
+	"context"
+	"errors"
+	"testing"
+	"time"
+)
+
+const (
+	// attempts is how many times a live DNS operation is attempted
+	// before the test fails.
+	attempts = 3
+
+	// AttemptTimeout bounds one attempt. Worst case for an operation
+	// is attempts * AttemptTimeout plus the backoff — about 26
+	// seconds, well inside the 90-second `go test -timeout` backstop
+	// even when several operations exhaust their attempts.
+	AttemptTimeout = 8 * time.Second
+
+	// backoffBase is the delay after the first failed attempt; it is
+	// multiplied by backoffFactor each time.
+	backoffBase = 500 * time.Millisecond
+
+	// backoffFactor is the exponential backoff multiplier.
+	backoffFactor = 2
+
+	// Concurrency caps how many live operations may be in flight
+	// across one test binary at once.
+	Concurrency = 6
+)
+
+// gate bounds concurrent live operations. It has to be package scoped:
+// the whole point is that it is shared by every parallel test in the
+// test binary.
+//
+//nolint:gochecknoglobals // package-wide live query rate limit
+var gate = make(chan struct{}, Concurrency)
+
+// ErrNoAnswer reports that a live operation produced no usable answer,
+// which is retried rather than asserted on.
+var ErrNoAnswer = errors.New("no answer from live DNS")
+
+// Run executes one attempt of a live operation, holding a slot in gate
+// for its duration and bounding it with its own timeout.
+func Run(op func(ctx context.Context) error) error {
+	gate <- struct{}{}
+	defer func() { <-gate }()
+
+	ctx, cancel := context.WithTimeout(
+		context.Background(), AttemptTimeout,
+	)
+	defer cancel()
+
+	return op(ctx)
+}
+
+// Retry runs op until it reports success, retrying failures with
+// exponential backoff, and fails the test if every attempt fails. op
+// returns an error only for a failure to obtain an answer — never for
+// an answer the test disagrees with, which belongs in an assertion so
+// that it fails immediately. op stores whatever it obtained where its
+// caller can find it.
+func Retry(
+	t *testing.T,
+	what string,
+	op func(ctx context.Context) error,
+) {
+	t.Helper()
+
+	var last error
+
+	backoff := backoffBase
+
+	for attempt := range attempts {
+		if attempt > 0 {
+			t.Logf(
+				"%s: attempt %d of %d failed (%v), "+
+					"retrying in %s",
+				what, attempt, attempts, last, backoff,
+			)
+			time.Sleep(backoff)
+
+			backoff *= backoffFactor
+		}
+
+		last = Run(op)
+		if last == nil {
+			return
+		}
+	}
+
+	t.Fatalf(
+		"%s: all %d live attempts failed: %v",
+		what, attempts, last,
+	)
+}

internal/livednstest/livednstest_test.go

@@ -0,0 +1,103 @@
+package livednstest_test
+
+import (
+	"context"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+
+	"sneak.berlin/go/dnswatcher/internal/livednstest"
+)
+
+// Tests for the retry and the concurrency limit themselves. They
+// perform no DNS resolution of any kind.
+
+func TestRetryRecoversFromTransientFailure(t *testing.T) {
+	t.Parallel()
+
+	const wantAttempts = 2
+
+	attempts := 0
+
+	livednstest.Retry(t, "transient", func(_ context.Context) error {
+		attempts++
+
+		if attempts < wantAttempts {
+			return livednstest.ErrNoAnswer
+		}
+
+		return nil
+	})
+
+	assert.Equal(t, wantAttempts, attempts)
+}
+
+func TestRetryGivesEachAttemptADeadline(t *testing.T) {
+	t.Parallel()
+
+	livednstest.Retry(t, "deadline", func(ctx context.Context) error {
+		deadline, ok := ctx.Deadline()
+		assert.True(t, ok, "attempt should carry a deadline")
+
+		remaining := time.Until(deadline)
+
+		assert.LessOrEqual(t, remaining, livednstest.AttemptTimeout)
+
+		// Lower bound too: without one this passes for a
+		// deadline far shorter than intended, which would
+		// silently turn every live attempt into an instant
+		// timeout.
+		assert.Greater(t, remaining, livednstest.AttemptTimeout/2)
+
+		return nil
+	})
+}
+
+func TestRunBoundsConcurrency(t *testing.T) {
+	t.Parallel()
+
+	const workers = 24
+
+	var (
+		mu       sync.Mutex
+		wg       sync.WaitGroup
+		inFlight int
+		maxSeen  int
+	)
+
+	wg.Add(workers)
+
+	for range workers {
+		go func() {
+			defer wg.Done()
+
+			_ = livednstest.Run(func(_ context.Context) error {
+				mu.Lock()
+				inFlight++
+
+				if inFlight > maxSeen {
+					maxSeen = inFlight
+				}
+				mu.Unlock()
+
+				time.Sleep(time.Millisecond)
+
+				mu.Lock()
+				inFlight--
+				mu.Unlock()
+
+				return nil
+			})
+		}()
+	}
+
+	wg.Wait()
+
+	assert.Positive(t, maxSeen)
+	assert.LessOrEqual(
+		t, maxSeen, livednstest.Concurrency,
+		"live queries must stay under the package-wide gate",
+	)
+}

internal/logger/logger_test.go

@@ -0,0 +1,66 @@
+package logger_test
+
+import (
+	"context"
+	"log/slog"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"sneak.berlin/go/dnswatcher/internal/globals"
+	"sneak.berlin/go/dnswatcher/internal/logger"
+)
+
+func newTestLogger(t *testing.T) *logger.Logger {
+	t.Helper()
+
+	g := &globals.Globals{Appname: "dnswatcher", Version: "test"}
+
+	l, err := logger.New(nil, logger.Params{Globals: g})
+	require.NoError(t, err)
+
+	return l
+}
+
+// TestNewReturnsUsableLogger checks that the constructor yields a
+// working *slog.Logger.
+func TestNewReturnsUsableLogger(t *testing.T) {
+	t.Parallel()
+
+	l := newTestLogger(t)
+	require.NotNil(t, l.Get(), "Get must return a non-nil logger")
+}
+
+// TestDefaultLevelExcludesDebug verifies the default configuration
+// logs at info: debug records are suppressed, info records pass.
+func TestDefaultLevelExcludesDebug(t *testing.T) {
+	t.Parallel()
+
+	log := newTestLogger(t).Get()
+	ctx := context.Background()
+
+	assert.False(t, log.Enabled(ctx, slog.LevelDebug),
+		"debug must be suppressed at the default level")
+	assert.True(t, log.Enabled(ctx, slog.LevelInfo),
+		"info must be enabled at the default level")
+}
+
+// TestEnableDebugLoggingChangesLevel verifies the debug and non-debug
+// configurations differ as intended: enabling debug makes debug
+// records pass where they previously did not.
+func TestEnableDebugLoggingChangesLevel(t *testing.T) {
+	t.Parallel()
+
+	l := newTestLogger(t)
+	log := l.Get()
+	ctx := context.Background()
+
+	require.False(t, log.Enabled(ctx, slog.LevelDebug),
+		"debug must start disabled")
+
+	l.EnableDebugLogging()
+
+	assert.True(t, log.Enabled(ctx, slog.LevelDebug),
+		"debug must be enabled after EnableDebugLogging")
+}

internal/middleware/middleware.go

@@ -21,6 +21,60 @@ import (
 // corsMaxAge is the maximum age for CORS preflight responses.
 const corsMaxAge = 300
 
+// Security response header values applied to every response.
+//
+// The CSP is as strict as the dashboard allows: the template ships no
+// JavaScript, no inline styles, no inline event handlers and no images,
+// and its only subresource is the embedded stylesheet at
+// /s/css/tailwind.min.css, which style-src 'self' permits. Neither
+// unsafe-inline nor unsafe-eval is used. frame-ancestors 'none' is the
+// primary anti-framing control; X-Frame-Options is the legacy fallback.
+const (
+	// hstsValue is emitted unconditionally, including over plain HTTP,
+	// because the service runs behind a TLS-terminating proxy and the
+	// browser must still enforce HTTPS end to end.
+	hstsValue = "max-age=31536000; includeSubDomains"
+
+	cspValue = "default-src 'self'; " +
+		"script-src 'none'; " +
+		"style-src 'self'; " +
+		"img-src 'self'; " +
+		"font-src 'none'; " +
+		"connect-src 'none'; " +
+		"object-src 'none'; " +
+		"base-uri 'none'; " +
+		"form-action 'none'; " +
+		"frame-ancestors 'none'"
+
+	frameOptionsValue = "DENY"
+
+	contentTypeOptionsValue = "nosniff"
+
+	// referrerPolicyValue is stricter than the policy minimum of
+	// strict-origin-when-cross-origin: the dashboard has no
+	// cross-origin navigation needs and its URL may name internal
+	// hosts.
+	referrerPolicyValue = "no-referrer"
+
+	permissionsPolicyValue = "accelerometer=(), " +
+		"autoplay=(), " +
+		"camera=(), " +
+		"display-capture=(), " +
+		"encrypted-media=(), " +
+		"fullscreen=(), " +
+		"geolocation=(), " +
+		"gyroscope=(), " +
+		"magnetometer=(), " +
+		"microphone=(), " +
+		"midi=(), " +
+		"payment=(), " +
+		"picture-in-picture=(), " +
+		"publickey-credentials-get=(), " +
+		"screen-wake-lock=(), " +
+		"usb=(), " +
+		"xr-spatial-tracking=()"
+)
+
 // Params contains dependencies for Middleware.
 type Params struct {
 	fx.In
@@ -186,6 +240,37 @@ func (m *Middleware) CORS() func(http.Handler) http.Handler {
 	})
 }
 
+// SecurityHeaders returns middleware that sets the security response
+// headers required for production internet exposure on every response.
+//
+// The headers are set before the request reaches the next handler so
+// that they are present on every response, including panics recovered
+// by chi's Recoverer and timeouts produced by chi's Timeout.
+func (m *Middleware) SecurityHeaders() func(http.Handler) http.Handler {
+	return func(next http.Handler) http.Handler {
+		return http.HandlerFunc(func(
+			writer http.ResponseWriter,
+			request *http.Request,
+		) {
+			header := writer.Header()
+			header.Set("Strict-Transport-Security", hstsValue)
+			header.Set("Content-Security-Policy", cspValue)
+			header.Set("X-Frame-Options", frameOptionsValue)
+			header.Set(
+				"X-Content-Type-Options",
+				contentTypeOptionsValue,
+			)
+			header.Set("Referrer-Policy", referrerPolicyValue)
+			header.Set(
+				"Permissions-Policy",
+				permissionsPolicyValue,
+			)
+
+			next.ServeHTTP(writer, request)
+		})
+	}
+}
+
 // MetricsAuth returns basic auth middleware for /metrics.
 func (m *Middleware) MetricsAuth() func(http.Handler) http.Handler {
 	if m.params.Config.MetricsUsername == "" {

internal/middleware/middleware_test.go

@@ -0,0 +1,334 @@
+package middleware_test
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"github.com/go-chi/chi/v5"
+	"go.uber.org/fx/fxtest"
+
+	"sneak.berlin/go/dnswatcher/internal/config"
+	"sneak.berlin/go/dnswatcher/internal/globals"
+	"sneak.berlin/go/dnswatcher/internal/handlers"
+	"sneak.berlin/go/dnswatcher/internal/logger"
+	"sneak.berlin/go/dnswatcher/internal/middleware"
+	"sneak.berlin/go/dnswatcher/internal/notify"
+	"sneak.berlin/go/dnswatcher/internal/state"
+)
+
+// Expected security header values, spelled out literally so that any
+// change to the middleware has to be made deliberately here as well.
+const (
+	wantHSTS = "max-age=31536000; includeSubDomains"
+
+	wantCSP = "default-src 'self'; " +
+		"script-src 'none'; " +
+		"style-src 'self'; " +
+		"img-src 'self'; " +
+		"font-src 'none'; " +
+		"connect-src 'none'; " +
+		"object-src 'none'; " +
+		"base-uri 'none'; " +
+		"form-action 'none'; " +
+		"frame-ancestors 'none'"
+
+	wantFrameOptions = "DENY"
+
+	wantContentTypeOptions = "nosniff"
+
+	wantReferrerPolicy = "no-referrer"
+
+	wantPermissionsPolicy = "accelerometer=(), " +
+		"autoplay=(), " +
+		"camera=(), " +
+		"display-capture=(), " +
+		"encrypted-media=(), " +
+		"fullscreen=(), " +
+		"geolocation=(), " +
+		"gyroscope=(), " +
+		"magnetometer=(), " +
+		"microphone=(), " +
+		"midi=(), " +
+		"payment=(), " +
+		"picture-in-picture=(), " +
+		"publickey-credentials-get=(), " +
+		"screen-wake-lock=(), " +
+		"usb=(), " +
+		"xr-spatial-tracking=()"
+)
+
+// stylesheetPath is the only subresource the dashboard loads.
+const stylesheetPath = "/s/css/tailwind.min.css"
+
+// newTestLogger builds a logger for direct component construction.
+func newTestLogger(t *testing.T) *logger.Logger {
+	t.Helper()
+
+	glob, err := globals.New(nil)
+	if err != nil {
+		t.Fatalf("globals.New: %v", err)
+	}
+
+	log, err := logger.New(nil, logger.Params{Globals: glob})
+	if err != nil {
+		t.Fatalf("logger.New: %v", err)
+	}
+
+	return log
+}
+
+// newTestMiddleware builds a Middleware without an fx application.
+func newTestMiddleware(t *testing.T) *middleware.Middleware {
+	t.Helper()
+
+	glob, err := globals.New(nil)
+	if err != nil {
+		t.Fatalf("globals.New: %v", err)
+	}
+
+	mw, err := middleware.New(nil, middleware.Params{
+		Logger:  newTestLogger(t),
+		Globals: glob,
+		Config:  &config.Config{},
+	})
+	if err != nil {
+		t.Fatalf("middleware.New: %v", err)
+	}
+
+	return mw
+}
+
+// serveWithSecurityHeaders runs a GET through SecurityHeaders and
+// returns the recorded response.
+func serveWithSecurityHeaders(
+	t *testing.T,
+	target string,
+	handler http.Handler,
+) *httptest.ResponseRecorder {
+	t.Helper()
+
+	mw := newTestMiddleware(t)
+	rec := httptest.NewRecorder()
+	req := httptest.NewRequestWithContext(
+		t.Context(), http.MethodGet, target, nil,
+	)
+
+	mw.SecurityHeaders()(handler).ServeHTTP(rec, req)
+
+	return rec
+}
+
+// okHandler writes a trivial 200 response.
+func okHandler() http.Handler {
+	return http.HandlerFunc(func(
+		writer http.ResponseWriter,
+		_ *http.Request,
+	) {
+		writer.WriteHeader(http.StatusOK)
+	})
+}
+
+func TestSecurityHeaders(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name   string
+		header string
+		want   string
+	}{
+		{
+			"hsts",
+			"Strict-Transport-Security",
+			wantHSTS,
+		},
+		{
+			"csp",
+			"Content-Security-Policy",
+			wantCSP,
+		},
+		{
+			"frame options",
+			"X-Frame-Options",
+			wantFrameOptions,
+		},
+		{
+			"content type options",
+			"X-Content-Type-Options",
+			wantContentTypeOptions,
+		},
+		{
+			"referrer policy",
+			"Referrer-Policy",
+			wantReferrerPolicy,
+		},
+		{
+			"permissions policy",
+			"Permissions-Policy",
+			wantPermissionsPolicy,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			rec := serveWithSecurityHeaders(t, "/", okHandler())
+
+			got := rec.Header().Get(tt.header)
+			if got != tt.want {
+				t.Errorf(
+					"%s = %q, want %q",
+					tt.header, got, tt.want,
+				)
+			}
+		})
+	}
+}
+
+// TestSecurityHeadersCSPDirectives guards the properties the repo
+// policy requires of the content security policy itself.
+func TestSecurityHeadersCSPDirectives(t *testing.T) {
+	t.Parallel()
+
+	rec := serveWithSecurityHeaders(t, "/", okHandler())
+	csp := rec.Header().Get("Content-Security-Policy")
+
+	forbidden := []string{"unsafe-inline", "unsafe-eval"}
+	for _, directive := range forbidden {
+		if strings.Contains(csp, directive) {
+			t.Errorf("CSP must not contain %q: %q", directive, csp)
+		}
+	}
+
+	required := []string{
+		"default-src 'self'",
+		"script-src 'none'",
+		"style-src 'self'",
+		"frame-ancestors 'none'",
+	}
+	for _, directive := range required {
+		if !strings.Contains(csp, directive) {
+			t.Errorf("CSP must contain %q: %q", directive, csp)
+		}
+	}
+}
+
+// TestSecurityHeadersOnErrorResponse verifies the headers are emitted
+// even when the wrapped handler fails, since they are set before the
+// handler runs.
+func TestSecurityHeadersOnErrorResponse(t *testing.T) {
+	t.Parallel()
+
+	failing := http.HandlerFunc(func(
+		writer http.ResponseWriter,
+		_ *http.Request,
+	) {
+		http.Error(
+			writer,
+			"boom",
+			http.StatusInternalServerError,
+		)
+	})
+
+	rec := serveWithSecurityHeaders(t, "/api/v1/status", failing)
+
+	if rec.Code != http.StatusInternalServerError {
+		t.Fatalf("status = %d, want 500", rec.Code)
+	}
+
+	if got := rec.Header().Get(
+		"X-Content-Type-Options",
+	); got != wantContentTypeOptions {
+		t.Errorf(
+			"X-Content-Type-Options = %q, want %q",
+			got, wantContentTypeOptions,
+		)
+	}
+
+	if got := rec.Header().Get(
+		"Strict-Transport-Security",
+	); got != wantHSTS {
+		t.Errorf(
+			"Strict-Transport-Security = %q, want %q",
+			got, wantHSTS,
+		)
+	}
+}
+
+// newTestHandlers builds real Handlers with empty monitoring state.
+func newTestHandlers(t *testing.T) *handlers.Handlers {
+	t.Helper()
+
+	glob, err := globals.New(nil)
+	if err != nil {
+		t.Fatalf("globals.New: %v", err)
+	}
+
+	log := newTestLogger(t)
+
+	notifier, err := notify.New(fxtest.NewLifecycle(t), notify.Params{
+		Logger: log,
+		Config: &config.Config{},
+	})
+	if err != nil {
+		t.Fatalf("notify.New: %v", err)
+	}
+
+	hnd, err := handlers.New(nil, handlers.Params{
+		Logger:  log,
+		Globals: glob,
+		State:   state.NewForTest(),
+		Notify:  notifier,
+	})
+	if err != nil {
+		t.Fatalf("handlers.New: %v", err)
+	}
+
+	return hnd
+}
+
+// TestDashboardRendersWithSecurityHeaders renders the real dashboard
+// through the middleware and checks that the policy still permits the
+// one stylesheet the page loads.
+func TestDashboardRendersWithSecurityHeaders(t *testing.T) {
+	t.Parallel()
+
+	mw := newTestMiddleware(t)
+	hnd := newTestHandlers(t)
+
+	router := chi.NewRouter()
+	router.Use(mw.SecurityHeaders())
+	router.Get("/", hnd.HandleDashboard())
+
+	rec := httptest.NewRecorder()
+	req := httptest.NewRequestWithContext(
+		t.Context(), http.MethodGet, "/", nil,
+	)
+
+	router.ServeHTTP(rec, req)
+
+	if rec.Code != http.StatusOK {
+		t.Fatalf("status = %d, want 200", rec.Code)
+	}
+
+	body := rec.Body.String()
+	if !strings.Contains(body, stylesheetPath) {
+		t.Errorf("dashboard does not reference %q", stylesheetPath)
+	}
+
+	if !strings.Contains(body, "dnswatcher") {
+		t.Errorf("dashboard body looks empty: %d bytes", len(body))
+	}
+
+	csp := rec.Header().Get("Content-Security-Policy")
+	if csp != wantCSP {
+		t.Errorf("CSP = %q, want %q", csp, wantCSP)
+	}
+
+	// The stylesheet is same-origin, so style-src 'self' allows it.
+	if !strings.Contains(csp, "style-src 'self'") {
+		t.Errorf("CSP would block %q: %q", stylesheetPath, csp)
+	}
+}

internal/notify/export_test.go

@@ -32,11 +32,27 @@ func NewRequestForTest(
 // NewTestService creates a Service suitable for unit testing.
 // It discards log output and uses the given transport.
 func NewTestService(transport http.RoundTripper) *Service {
-	return &Service{
-		log:       slog.New(slog.DiscardHandler),
-		transport: transport,
-		history:   NewAlertHistory(),
+	return newService(slog.New(slog.DiscardHandler), transport)
 }
+
+// NewTestServiceWithLogger creates a Service that writes to the
+// given handler, so tests can assert on emitted log records.
+func NewTestServiceWithLogger(
+	transport http.RoundTripper,
+	handler slog.Handler,
+) *Service {
+	return newService(slog.New(handler), transport)
+}
+
+// Drain exports drain for testing.
+func (svc *Service) Drain(ctx context.Context) {
+	svc.drain(ctx)
+}
+
+// OutstandingDeliveries reports how many delivery goroutines
+// are currently tracked as in flight.
+func (svc *Service) OutstandingDeliveries() int64 {
+	return svc.outstanding.Load()
 }
 
 // SetNtfyURL sets the ntfy URL on a Service for testing.

internal/notify/notify.go

@@ -12,6 +12,8 @@ import (
 	"log/slog"
 	"net/http"
 	"net/url"
+	"sync"
+	"sync/atomic"
 	"time"
 
 	"go.uber.org/fx"
@@ -115,19 +117,41 @@ type Service struct {
 	history              *AlertHistory
 	retryConfig          RetryConfig
 	sleepFn              func(time.Duration) <-chan time.Time
+
+	// Shutdown draining state.  drainMu guards draining and
+	// serialises it against the counter increment in
+	// startDelivery; inFlight tracks the delivery goroutines
+	// themselves and outstanding mirrors its count so a timed
+	// out drain can report how many were abandoned.
+	drainMu     sync.Mutex
+	draining    bool
+	inFlight    sync.WaitGroup
+	outstanding atomic.Int64
+	abandon     chan struct{}
+	abandonOnce sync.Once
+}
+
+// newService builds a Service with the fields every Service
+// needs regardless of how it was constructed.
+func newService(
+	log *slog.Logger,
+	transport http.RoundTripper,
+) *Service {
+	return &Service{
+		log:       log,
+		transport: transport,
+		history:   NewAlertHistory(),
+		abandon:   make(chan struct{}),
+	}
 }
 
 // New creates a new notify Service.
 func New(
-	_ fx.Lifecycle,
+	lifecycle fx.Lifecycle,
 	params Params,
 ) (*Service, error) {
-	svc := &Service{
-		log:       params.Logger.Get(),
-		transport: http.DefaultTransport,
-		config:    params.Config,
-		history:   NewAlertHistory(),
-	}
+	svc := newService(params.Logger.Get(), http.DefaultTransport)
+	svc.config = params.Config
 
 	if params.Config.NtfyTopic != "" {
 		u, err := ValidateWebhookURL(
@@ -168,6 +192,14 @@ func New(
 		svc.mattermostWebhookURL = u
 	}
 
+	lifecycle.Append(fx.Hook{
+		OnStop: func(ctx context.Context) error {
+			svc.drain(ctx)
+
+			return nil
+		},
+	})
+
 	return svc, nil
 }
 
@@ -194,6 +226,32 @@ func (svc *Service) SendNotification(
 	svc.dispatchMattermost(ctx, title, message, priority)
 }
 
+// dispatch delivers a notification to one endpoint on a
+// tracked background goroutine.
+//
+// The delivery context is detached from ctx with
+// context.WithoutCancel so that a cancelled caller does not
+// kill a delivery already under way; the shutdown drain, not
+// the caller, decides how long deliveries may keep running.
+func (svc *Service) dispatch(
+	ctx context.Context,
+	endpoint string,
+	send func(context.Context) error,
+) {
+	notifyCtx := context.WithoutCancel(ctx)
+
+	svc.startDelivery(endpoint, func() {
+		err := svc.deliverWithRetry(notifyCtx, endpoint, send)
+		if err != nil {
+			svc.log.Error(
+				"failed to send notification after retries",
+				"endpoint", endpoint,
+				"error", err,
+			)
+		}
+	})
+}
+
 func (svc *Service) dispatchNtfy(
 	ctx context.Context,
 	title, message, priority string,
@@ -202,26 +260,11 @@ func (svc *Service) dispatchNtfy(
 		return
 	}
 
-	go func() {
-		notifyCtx := context.WithoutCancel(ctx)
-
-		err := svc.deliverWithRetry(
-			notifyCtx, "ntfy",
-			func(c context.Context) error {
+	svc.dispatch(ctx, "ntfy", func(c context.Context) error {
 		return svc.sendNtfy(
-					c, svc.ntfyURL,
-					title, message, priority,
+			c, svc.ntfyURL, title, message, priority,
 		)
-			},
-		)
-		if err != nil {
-			svc.log.Error(
-				"failed to send ntfy notification "+
-					"after retries",
-				"error", err,
-			)
-		}
-	}()
+	})
 }
 
 func (svc *Service) dispatchSlack(
@@ -232,26 +275,11 @@ func (svc *Service) dispatchSlack(
 		return
 	}
 
-	go func() {
-		notifyCtx := context.WithoutCancel(ctx)
-
-		err := svc.deliverWithRetry(
-			notifyCtx, "slack",
-			func(c context.Context) error {
+	svc.dispatch(ctx, "slack", func(c context.Context) error {
 		return svc.sendSlack(
-					c, svc.slackWebhookURL,
-					title, message, priority,
+			c, svc.slackWebhookURL, title, message, priority,
 		)
-			},
-		)
-		if err != nil {
-			svc.log.Error(
-				"failed to send slack notification "+
-					"after retries",
-				"error", err,
-			)
-		}
-	}()
+	})
 }
 
 func (svc *Service) dispatchMattermost(
@@ -262,11 +290,8 @@ func (svc *Service) dispatchMattermost(
 		return
 	}
 
-	go func() {
-		notifyCtx := context.WithoutCancel(ctx)
-
-		err := svc.deliverWithRetry(
-			notifyCtx, "mattermost",
+	svc.dispatch(
+		ctx, "mattermost",
 		func(c context.Context) error {
 			return svc.sendSlack(
 				c, svc.mattermostWebhookURL,
@@ -274,14 +299,6 @@ func (svc *Service) dispatchMattermost(
 			)
 		},
 	)
-		if err != nil {
-			svc.log.Error(
-				"failed to send mattermost notification "+
-					"after retries",
-				"error", err,
-			)
-		}
-	}()
 }
 
 func (svc *Service) sendNtfy(

internal/notify/retry.go

@@ -2,6 +2,7 @@ package notify
 
 import (
 	"context"
+	"fmt"
 	"math"
 	"math/rand/v2"
 	"time"
@@ -121,6 +122,14 @@ func (svc *Service) deliverWithRetry(
 		select {
 		case <-ctx.Done():
 			return ctx.Err()
+		case <-svc.abandon:
+			// Shutdown drained past its deadline; stop
+			// sleeping rather than outlive the process.
+			// A nil channel (Service built without a
+			// constructor) simply never fires.
+			return fmt.Errorf(
+				"%w: %s", ErrDeliveryAbandoned, endpoint,
+			)
 		case <-svc.sleepFunc(delay):
 		}
 	}

internal/notify/shutdown.go

@@ -0,0 +1,119 @@
+package notify
+
+import (
+	"context"
+	"errors"
+)
+
+// ErrDeliveryAbandoned is returned by a retry loop that was
+// cut short because shutdown drained past its deadline.
+var ErrDeliveryAbandoned = errors.New(
+	"notification delivery abandoned at shutdown",
+)
+
+// startDelivery runs fn on its own goroutine while tracking it,
+// so that drain can wait for it during shutdown.
+//
+// The WaitGroup counter is incremented here, on the caller's
+// goroutine, before the worker exists: incrementing it inside
+// the worker would race with drain's Wait and could let
+// shutdown sail past a delivery that had not started yet.
+//
+// Once draining has begun the delivery is refused outright
+// rather than queued, so a steady stream of newly submitted
+// notifications cannot keep extending the drain.
+func (svc *Service) startDelivery(endpoint string, fn func()) {
+	svc.drainMu.Lock()
+
+	if svc.draining {
+		svc.drainMu.Unlock()
+
+		svc.log.Warn(
+			"notification not dispatched: shutdown in progress",
+			"endpoint", endpoint,
+		)
+
+		return
+	}
+
+	svc.outstanding.Add(1)
+
+	// WaitGroup.Go increments the counter synchronously, here,
+	// and only then starts the goroutine.
+	svc.inFlight.Go(func() {
+		// Runs before the WaitGroup counter is decremented, so
+		// a drain that times out reports an accurate count.
+		defer svc.outstanding.Add(-1)
+
+		fn()
+	})
+
+	svc.drainMu.Unlock()
+}
+
+// drain waits for in-flight notification deliveries to finish.
+//
+// It first stops accepting new deliveries, then waits until
+// either every outstanding delivery has completed or ctx
+// expires — whichever comes first.  ctx is the context fx
+// passes to the OnStop hook, so a permanently dead webhook
+// cannot hang shutdown indefinitely.
+//
+// When the deadline arrives with deliveries still outstanding,
+// the count is logged at warn level and the abandon channel is
+// closed, which releases any retry loop sleeping in backoff.
+// Deliveries already inside an HTTP round trip are bounded by
+// the existing httpClientTimeout instead.
+//
+// A ctx that is already expired on entry is not by itself cause
+// for alarm: if nothing is outstanding there is nothing to
+// abandon, and the drain says so at debug level rather than
+// warning about deliveries that do not exist.
+func (svc *Service) drain(ctx context.Context) {
+	svc.drainMu.Lock()
+	svc.draining = true
+	svc.drainMu.Unlock()
+
+	done := make(chan struct{})
+
+	go func() {
+		svc.inFlight.Wait()
+		close(done)
+	}()
+
+	select {
+	case <-done:
+		svc.log.Debug(
+			"all in-flight notifications completed",
+		)
+	case <-ctx.Done():
+		// outstanding is decremented before the WaitGroup
+		// counter, and startDelivery can no longer add to it
+		// now that draining is set, so a zero here means every
+		// delivery really did finish.  ctx expiring in that
+		// state (an OnStop context that was already cancelled
+		// on entry is the usual way) abandons nothing, so it
+		// must not close abandon or warn about it.
+		abandoned := svc.outstanding.Load()
+		if abandoned == 0 {
+			svc.log.Debug(
+				"all in-flight notifications completed",
+			)
+
+			return
+		}
+
+		svc.abandonOnce.Do(func() {
+			if svc.abandon != nil {
+				close(svc.abandon)
+			}
+		})
+
+		svc.log.Warn(
+			"shutdown deadline reached with notifications "+
+				"still in flight; abandoning them",
+			"abandoned", abandoned,
+			"error", ctx.Err(),
+		)
+	}
+}

internal/notify/shutdown_test.go

@@ -0,0 +1,531 @@
+package notify_test
+
+import (
+	"bytes"
+	"context"
+	"log/slog"
+	"net/http"
+	"net/http/httptest"
+	"net/url"
+	"strings"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"go.uber.org/fx"
+
+	"sneak.berlin/go/dnswatcher/internal/config"
+	"sneak.berlin/go/dnswatcher/internal/globals"
+	"sneak.berlin/go/dnswatcher/internal/logger"
+	"sneak.berlin/go/dnswatcher/internal/notify"
+)
+
+// Timings used by the drain tests.  They stay in the same
+// 10-100ms band as the retry tests so the suite never waits on
+// a real backoff delay.
+const (
+	// inFlightHold is how long a delivery is kept mid-request
+	// before the handler is released.
+	inFlightHold = 30 * time.Millisecond
+
+	// drainDeadline bounds a drain that is expected to time
+	// out.
+	drainDeadline = 50 * time.Millisecond
+
+	// drainSlack is the upper bound on how long a bounded
+	// drain may take; generous enough for a loaded CI box,
+	// still far below the 20s test ceiling.
+	drainSlack = 2 * time.Second
+
+	// settleDelay is how long to wait before asserting that
+	// something did *not* happen.
+	settleDelay = 50 * time.Millisecond
+
+	// idleDrainBound is the upper bound on a drain that has
+	// nothing in flight.  It is deliberately far above the cost
+	// of the goroutine hop through inFlight.Wait() — which
+	// reached 57ms on a loaded box under -race with the package's
+	// parallel tests — and far below drainSlack, the deadline
+	// such a drain is given.  A drain that blocked until its
+	// deadline instead of returning on the WaitGroup therefore
+	// still fails this bound, but scheduling delay alone cannot.
+	idleDrainBound = 500 * time.Millisecond
+)
+
+// syncBuffer is an io.Writer safe for concurrent use, so log
+// output written from delivery goroutines can be inspected.
+type syncBuffer struct {
+	mu  sync.Mutex
+	buf bytes.Buffer
+}
+
+func (sb *syncBuffer) Write(p []byte) (int, error) {
+	sb.mu.Lock()
+	defer sb.mu.Unlock()
+
+	return sb.buf.Write(p) //nolint:wrapcheck // test helper
+}
+
+func (sb *syncBuffer) String() string {
+	sb.mu.Lock()
+	defer sb.mu.Unlock()
+
+	return sb.buf.String()
+}
+
+// newLoggingService returns a Service writing JSON logs into
+// the returned buffer.
+func newLoggingService(
+	transport http.RoundTripper,
+) (*notify.Service, *syncBuffer) {
+	logs := &syncBuffer{}
+	handler := slog.NewJSONHandler(logs, nil)
+
+	return notify.NewTestServiceWithLogger(transport, handler),
+		logs
+}
+
+// blockingNtfyServer returns a server whose handler signals on
+// entered, waits for release, and then responds 200.
+func blockingNtfyServer(
+	entered chan<- struct{},
+	release <-chan struct{},
+	served *atomic.Bool,
+) *httptest.Server {
+	var once sync.Once
+
+	return httptest.NewServer(
+		http.HandlerFunc(
+			func(w http.ResponseWriter, _ *http.Request) {
+				once.Do(func() { close(entered) })
+				<-release
+
+				served.Store(true)
+
+				w.WriteHeader(http.StatusOK)
+			}),
+	)
+}
+
+// TestDrainWaitsForInFlightDelivery verifies that a delivery
+// already under way when shutdown starts is allowed to finish.
+func TestDrainWaitsForInFlightDelivery(t *testing.T) {
+	t.Parallel()
+
+	var served atomic.Bool
+
+	entered := make(chan struct{})
+	release := make(chan struct{})
+
+	srv := blockingNtfyServer(entered, release, &served)
+	defer srv.Close()
+
+	topicURL, _ := url.Parse(srv.URL)
+
+	svc := notify.NewTestService(http.DefaultTransport)
+	svc.SetNtfyURL(topicURL)
+
+	svc.SendNotification(
+		context.Background(), "t", "m", prioInfo,
+	)
+
+	// Make sure the delivery really is mid-request before the
+	// drain begins.
+	select {
+	case <-entered:
+	case <-time.After(drainSlack):
+		t.Fatal("delivery never reached the endpoint")
+	}
+
+	// As in TestDrainBoundedByContextDeadline: start is captured
+	// before the clock it is compared against, here the timer
+	// holding the delivery open, so elapsed covers the whole hold
+	// and the lower bound cannot come out short from scheduling
+	// delay alone.
+	start := time.Now()
+
+	timer := time.AfterFunc(inFlightHold, func() {
+		close(release)
+	})
+	defer timer.Stop()
+
+	ctx, cancel := context.WithTimeout(
+		context.Background(), drainSlack,
+	)
+	defer cancel()
+
+	svc.Drain(ctx)
+
+	elapsed := time.Since(start)
+
+	if !served.Load() {
+		t.Error(
+			"drain returned before the in-flight delivery " +
+				"completed",
+		)
+	}
+
+	if elapsed < inFlightHold {
+		t.Errorf(
+			"drain took %v, want at least %v",
+			elapsed, inFlightHold,
+		)
+	}
+
+	if got := svc.OutstandingDeliveries(); got != 0 {
+		t.Errorf("outstanding deliveries = %d, want 0", got)
+	}
+}
+
+// neverFires returns a channel that never delivers, standing in
+// for a long backoff sleep without actually sleeping.
+func neverFires(_ time.Duration) <-chan time.Time {
+	return make(chan time.Time)
+}
+
+// TestDrainBoundedByContextDeadline verifies that a delivery
+// stuck retrying against a dead endpoint does not hold shutdown
+// past the OnStop context deadline, and that the abandoned
+// deliveries are logged at warn level rather than dropped
+// silently.
+func TestDrainBoundedByContextDeadline(t *testing.T) {
+	t.Parallel()
+
+	var requests atomic.Int64
+
+	srv := httptest.NewServer(
+		http.HandlerFunc(
+			func(w http.ResponseWriter, _ *http.Request) {
+				requests.Add(1)
+
+				w.WriteHeader(http.StatusInternalServerError)
+			}),
+	)
+	defer srv.Close()
+
+	topicURL, _ := url.Parse(srv.URL)
+
+	svc, logs := newLoggingService(http.DefaultTransport)
+	svc.SetNtfyURL(topicURL)
+	// Never let the backoff sleep complete: the delivery is
+	// parked in its retry wait until shutdown releases it.
+	svc.SetSleepFunc(neverFires)
+	svc.SetRetryConfig(notify.RetryConfig{
+		MaxRetries: 5,
+		BaseDelay:  time.Hour,
+		MaxDelay:   time.Hour,
+	})
+
+	svc.SendNotification(
+		context.Background(), "t", "m", prioError,
+	)
+
+	waitForCondition(t, func() bool {
+		return requests.Load() >= 1 &&
+			svc.OutstandingDeliveries() == 1
+	})
+
+	// start must be captured *before* the deadline clock starts,
+	// so that the measured interval is a superset of the deadline
+	// interval.  Capturing it after context.WithTimeout would
+	// make elapsed structurally smaller than drainDeadline and
+	// the lower bound below unfalsifiable-by-luck: it would fail
+	// whenever the two statements were separated by any
+	// scheduling delay, and pass otherwise, regardless of what
+	// the drain did.
+	start := time.Now()
+
+	ctx, cancel := context.WithTimeout(
+		context.Background(), drainDeadline,
+	)
+	defer cancel()
+
+	// The upper bound is enforced by a watchdog rather than by
+	// measuring after the fact: a drain that is not bounded at
+	// all never returns here (the delivery is parked in a backoff
+	// that never fires), so an unbounded drain must fail this
+	// test promptly instead of hanging the package until the test
+	// binary's 30s timeout.
+	returned := make(chan struct{})
+
+	go func() {
+		defer close(returned)
+
+		svc.Drain(ctx)
+	}()
+
+	select {
+	case <-returned:
+	case <-time.After(drainSlack):
+		t.Fatalf(
+			"drain did not return within %v; its %v deadline "+
+				"did not bound it",
+			drainSlack, drainDeadline,
+		)
+	}
+
+	// The lower bound is the real assertion: the drain must have
+	// waited for its whole deadline rather than giving up on the
+	// outstanding delivery early.  With start captured above, an
+	// early return is the only thing that can make it fail.
+	if elapsed := time.Since(start); elapsed < drainDeadline {
+		t.Errorf(
+			"drain returned after %v, before its %v deadline",
+			elapsed, drainDeadline,
+		)
+	}
+
+	assertAbandonLogged(t, logs.String())
+
+	// The abandoned delivery must stop retrying rather than
+	// outlive the drain.
+	waitForCondition(t, func() bool {
+		return svc.OutstandingDeliveries() == 0
+	})
+}
+
+// assertAbandonLogged checks that the drain logged the
+// abandoned deliveries at warn level with a count.
+func assertAbandonLogged(t *testing.T, output string) {
+	t.Helper()
+
+	if !strings.Contains(output, `"level":"WARN"`) {
+		t.Errorf(
+			"abandoned deliveries not logged at warn level; "+
+				"log output: %s",
+			output,
+		)
+	}
+
+	if !strings.Contains(output, `"abandoned":1`) {
+		t.Errorf(
+			"abandoned delivery count not logged; "+
+				"log output: %s",
+			output,
+		)
+	}
+}
+
+// TestDrainRefusesNewDeliveries verifies that notifications
+// submitted after the drain has begun are refused and logged,
+// so a stream of new work cannot extend shutdown indefinitely.
+func TestDrainRefusesNewDeliveries(t *testing.T) {
+	t.Parallel()
+
+	var requests atomic.Int64
+
+	srv := httptest.NewServer(
+		http.HandlerFunc(
+			func(w http.ResponseWriter, _ *http.Request) {
+				requests.Add(1)
+
+				w.WriteHeader(http.StatusOK)
+			}),
+	)
+	defer srv.Close()
+
+	target, _ := url.Parse(srv.URL)
+
+	svc, logs := newLoggingService(http.DefaultTransport)
+	svc.SetNtfyURL(target)
+	svc.SetSlackWebhookURL(target)
+	svc.SetMattermostWebhookURL(target)
+
+	ctx, cancel := context.WithTimeout(
+		context.Background(), drainSlack,
+	)
+	defer cancel()
+
+	// Nothing is in flight, so this returns immediately and
+	// leaves the service refusing further deliveries.
+	svc.Drain(ctx)
+
+	for range 3 {
+		svc.SendNotification(
+			context.Background(), "t", "m", prioInfo,
+		)
+	}
+
+	time.Sleep(settleDelay)
+
+	if got := requests.Load(); got != 0 {
+		t.Errorf(
+			"%d requests reached the endpoint after drain, "+
+				"want 0",
+			got,
+		)
+	}
+
+	if got := svc.OutstandingDeliveries(); got != 0 {
+		t.Errorf("outstanding deliveries = %d, want 0", got)
+	}
+
+	output := logs.String()
+	if !strings.Contains(output, "shutdown in progress") {
+		t.Errorf(
+			"refused deliveries not logged; log output: %s",
+			output,
+		)
+	}
+}
+
+// recordingLifecycle is a minimal fx.Lifecycle that records the
+// hooks appended to it, so the wiring done by notify.New can be
+// inspected without standing up a whole fx application.
+type recordingLifecycle struct {
+	hooks []fx.Hook
+}
+
+func (l *recordingLifecycle) Append(hook fx.Hook) {
+	l.hooks = append(l.hooks, hook)
+}
+
+// newNotifyService builds a Service through the real
+// constructor, wired to the given lifecycle.
+func newNotifyService(
+	t *testing.T,
+	lifecycle fx.Lifecycle,
+	ntfyTopic string,
+) *notify.Service {
+	t.Helper()
+
+	g, err := globals.New(nil)
+	if err != nil {
+		t.Fatalf("globals.New: %v", err)
+	}
+
+	log, err := logger.New(nil, logger.Params{Globals: g})
+	if err != nil {
+		t.Fatalf("logger.New: %v", err)
+	}
+
+	svc, err := notify.New(lifecycle, notify.Params{
+		Logger: log,
+		Config: &config.Config{NtfyTopic: ntfyTopic},
+	})
+	if err != nil {
+		t.Fatalf("notify.New: %v", err)
+	}
+
+	return svc
+}
+
+// TestNewRegistersDrainingStopHook verifies that notify.New
+// wires an OnStop hook into the fx lifecycle and that the hook
+// waits for in-flight deliveries.
+func TestNewRegistersDrainingStopHook(t *testing.T) {
+	t.Parallel()
+
+	var served atomic.Bool
+
+	entered := make(chan struct{})
+	release := make(chan struct{})
+
+	srv := blockingNtfyServer(entered, release, &served)
+	defer srv.Close()
+
+	lifecycle := &recordingLifecycle{}
+	svc := newNotifyService(t, lifecycle, srv.URL)
+
+	if len(lifecycle.hooks) != 1 {
+		t.Fatalf(
+			"appended %d lifecycle hooks, want 1",
+			len(lifecycle.hooks),
+		)
+	}
+
+	stop := lifecycle.hooks[0].OnStop
+	if stop == nil {
+		t.Fatal("lifecycle hook has no OnStop function")
+	}
+
+	svc.SendNotification(
+		context.Background(), "t", "m", prioInfo,
+	)
+
+	select {
+	case <-entered:
+	case <-time.After(drainSlack):
+		t.Fatal("delivery never reached the endpoint")
+	}
+
+	timer := time.AfterFunc(inFlightHold, func() {
+		close(release)
+	})
+	defer timer.Stop()
+
+	ctx, cancel := context.WithTimeout(
+		context.Background(), drainSlack,
+	)
+	defer cancel()
+
+	err := stop(ctx)
+	if err != nil {
+		t.Fatalf("OnStop returned error: %v", err)
+	}
+
+	if !served.Load() {
+		t.Error(
+			"OnStop returned before the in-flight delivery " +
+				"completed",
+		)
+	}
+}
+
+// TestDrainWithoutDeliveriesReturnsImmediately verifies the
+// common case: nothing in flight, shutdown is not delayed.
+func TestDrainWithoutDeliveriesReturnsImmediately(t *testing.T) {
+	t.Parallel()
+
+	svc := notify.NewTestService(http.DefaultTransport)
+
+	// Captured before the deadline clock, as elsewhere in this
+	// file; for an upper bound that is the conservative
+	// direction, since the measured interval can then only be
+	// longer than the drain itself.
+	start := time.Now()
+
+	ctx, cancel := context.WithTimeout(
+		context.Background(), drainSlack,
+	)
+	defer cancel()
+
+	svc.Drain(ctx)
+
+	if elapsed := time.Since(start); elapsed > idleDrainBound {
+		t.Errorf(
+			"drain of an idle service took %v, want well "+
+				"under its %v deadline",
+			elapsed, drainSlack,
+		)
+	}
+}
+
+// TestDrainWithCancelledContextDoesNotWarn verifies that an
+// OnStop context that is already dead on entry does not produce
+// an "abandoning them" warning when there was nothing in flight
+// to abandon.  The expired context wins the select immediately,
+// so only the outstanding count can tell the difference between
+// a genuine timeout and a shutdown that had simply already run
+// out of time with no work left.
+func TestDrainWithCancelledContextDoesNotWarn(t *testing.T) {
+	t.Parallel()
+
+	svc, logs := newLoggingService(http.DefaultTransport)
+
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel()
+
+	svc.Drain(ctx)
+
+	if output := logs.String(); strings.Contains(
+		output, `"level":"WARN"`,
+	) {
+		t.Errorf(
+			"drain with nothing in flight warned about "+
+				"abandoned deliveries; log output: %s",
+			output,
+		)
+	}
+}

internal/resolver/dns_client.go

@@ -7,8 +7,8 @@ import (
 	"github.com/miekg/dns"
 )
 
-// DNSClient abstracts DNS wire-protocol exchanges so the resolver
-// can be tested without hitting real nameservers.
+// DNSClient sends one DNS message to a nameserver and returns the
+// reply. The resolver holds one for UDP and one for TCP.
 type DNSClient interface {
 	ExchangeContext(
 		ctx context.Context,

internal/resolver/export_test.go

@@ -0,0 +1,8 @@
+package resolver
+
+import "github.com/miekg/dns"
+
+// ExtractRecordValue exports extractRecordValue for testing.
+func ExtractRecordValue(rr dns.RR) string {
+	return extractRecordValue(rr)
+}

internal/resolver/iterative.go

@@ -608,7 +608,10 @@ func classifyResponse(resp *NameserverResponse, state queryState) {
 	}
 }
 
-// extractRecordValue formats a DNS RR value as a string.
+// extractRecordValue formats a DNS RR value as a string. DNS names
+// are case-insensitive and nameservers may answer in any letter case,
+// so names are lower-cased to compare equal. TXT and CAA values keep
+// their letter case.
 func extractRecordValue(rr dns.RR) string {
 	switch r := rr.(type) {
 	case *dns.A:
@@ -616,22 +619,22 @@ func extractRecordValue(rr dns.RR) string {
 	case *dns.AAAA:
 		return r.AAAA.String()
 	case *dns.CNAME:
-		return r.Target
+		return strings.ToLower(r.Target)
 	case *dns.MX:
-		return fmt.Sprintf("%d %s", r.Preference, r.Mx)
+		return fmt.Sprintf("%d %s", r.Preference, strings.ToLower(r.Mx))
 	case *dns.TXT:
 		return strings.Join(r.Txt, "")
 	case *dns.SRV:
 		return fmt.Sprintf(
 			"%d %d %d %s",
-			r.Priority, r.Weight, r.Port, r.Target,
+			r.Priority, r.Weight, r.Port, strings.ToLower(r.Target),
 		)
 	case *dns.CAA:
 		return fmt.Sprintf(
 			"%d %s \"%s\"", r.Flag, r.Tag, r.Value,
 		)
 	case *dns.NS:
-		return r.Ns
+		return strings.ToLower(r.Ns)
 	default:
 		return ""
 	}

internal/resolver/iterative_test.go

@@ -0,0 +1,62 @@
+package resolver_test
+
+import (
+	"testing"
+
+	"github.com/miekg/dns"
+	"github.com/stretchr/testify/assert"
+
+	"sneak.berlin/go/dnswatcher/internal/resolver"
+)
+
+func TestExtractRecordValue_LetterCase(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name string
+		rr   dns.RR
+		want string
+	}{
+		{
+			name: "MX target lower-cased",
+			rr:   &dns.MX{Preference: 1, Mx: "ASPMX.L.GOOGLE.COM."},
+			want: "1 aspmx.l.google.com.",
+		},
+		{
+			name: "NS target lower-cased",
+			rr:   &dns.NS{Ns: "x.ns.joker.COM."},
+			want: "x.ns.joker.com.",
+		},
+		{
+			name: "CNAME target lower-cased",
+			rr:   &dns.CNAME{Target: "WWW.Example.Com."},
+			want: "www.example.com.",
+		},
+		{
+			name: "SRV target lower-cased",
+			rr: &dns.SRV{
+				Priority: 10, Weight: 5, Port: 443,
+				Target: "SIP.Example.Com.",
+			},
+			want: "10 5 443 sip.example.com.",
+		},
+		{
+			name: "TXT value keeps its case",
+			rr:   &dns.TXT{Txt: []string{"Verify=AbC123"}},
+			want: "Verify=AbC123",
+		},
+		{
+			name: "CAA value keeps its case",
+			rr:   &dns.CAA{Flag: 0, Tag: "issue", Value: "LetsEncrypt.org"},
+			want: `0 issue "LetsEncrypt.org"`,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			assert.Equal(t, tt.want, resolver.ExtractRecordValue(tt.rr))
+		})
+	}
+}

internal/resolver/livedns_harness_test.go

@@ -0,0 +1,192 @@
+package resolver_test
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+
+	"sneak.berlin/go/dnswatcher/internal/resolver"
+)
+
+// Tests for the live-DNS harness in livedns_test.go itself. These
+// exercise pure logic; they perform no DNS resolution of any kind, so
+// they neither mock DNS nor depend on it.
+
+// Names for the synthetic status maps below. Nothing is ever queried
+// at them: they are map keys handed to the package's pure counting
+// helpers, not a stand-in for a nameserver.
+const (
+	nsExample1 = "ns1.example."
+	nsExample2 = "ns2.example."
+	nsExample3 = "ns3.example."
+	nsExample4 = "ns4.example."
+)
+
+func TestLiveQuorumIsStrictMajority(t *testing.T) {
+	t.Parallel()
+
+	cases := map[int]int{
+		0:  1,
+		1:  1,
+		2:  2,
+		3:  2,
+		4:  3,
+		5:  3,
+		13: 7,
+	}
+
+	for total, want := range cases {
+		assert.Equal(
+			t, want, liveQuorum(total),
+			"liveQuorum(%d)", total,
+		)
+	}
+}
+
+func TestStatusCountingIgnoresSilentNameservers(t *testing.T) {
+	t.Parallel()
+
+	results := map[string]*resolver.NameserverResponse{
+		nsExample1: {
+			Nameserver: nsExample1,
+			Status:     resolver.StatusOK,
+		},
+		nsExample2: {
+			Nameserver: nsExample2,
+			Status:     resolver.StatusOK,
+		},
+		nsExample3: {
+			Nameserver: nsExample3,
+			Status:     resolver.StatusTimeout,
+		},
+		nsExample4: {
+			Nameserver: nsExample4,
+			Status:     resolver.StatusError,
+		},
+	}
+
+	assert.Equal(
+		t, 2, countStatus(results, resolver.StatusOK),
+	)
+	assert.Equal(
+		t, 0, countStatus(results, resolver.StatusNXDomain),
+	)
+
+	// Two of four answered, which is short of the quorum of
+	// three: this is the state that triggers a retry rather
+	// than an assertion failure.
+	assert.Equal(t, 2, answeredCount(results))
+	assert.Less(t, answeredCount(results), liveQuorum(len(results)))
+
+	assert.Equal(
+		t,
+		"ns1.example.=ok ns2.example.=ok "+
+			"ns3.example.=timeout ns4.example.=error",
+		describeStatuses(results),
+	)
+}
+
+// TestUnsanctionedStatusesRejectsWrongAnswers is the regression test
+// for the defect this allowlist exists to prevent: a minority of
+// nameservers answering WRONGLY while quorum keeps the suite green.
+// nodata is the case that motivated it — it is a wrong answer, not
+// silence, and it was previously banned by neither test.
+func TestUnsanctionedStatusesRejectsWrongAnswers(t *testing.T) {
+	t.Parallel()
+
+	// Four nameservers, three OK and one answering nodata: a
+	// quorum of three is satisfied and no NXDOMAIN is present, so
+	// the old blocklist assertions both passed on this input.
+	results := map[string]*resolver.NameserverResponse{
+		nsExample1: {
+			Nameserver: nsExample1,
+			Status:     resolver.StatusOK,
+		},
+		nsExample2: {
+			Nameserver: nsExample2,
+			Status:     resolver.StatusOK,
+		},
+		nsExample3: {
+			Nameserver: nsExample3,
+			Status:     resolver.StatusOK,
+		},
+		nsExample4: {
+			Nameserver: nsExample4,
+			Status:     resolver.StatusNoData,
+		},
+	}
+
+	assert.GreaterOrEqual(
+		t,
+		countStatus(results, resolver.StatusOK),
+		liveQuorum(len(results)),
+	)
+	assert.Zero(t, countStatus(results, resolver.StatusNXDomain))
+
+	// nodata is an ANSWER, so it never triggers a retry: nothing
+	// but the allowlist stands between it and a false green.
+	assert.Equal(t, len(results), answeredCount(results))
+
+	assert.Equal(
+		t,
+		[]string{nsExample4 + "=nodata"},
+		unsanctionedStatuses(
+			results,
+			resolver.StatusOK,
+			resolver.StatusTimeout,
+			resolver.StatusError,
+		),
+		"nodata must be reported as an unsanctioned status",
+	)
+}
+
+func TestUnsanctionedStatusesToleratesSilenceOnly(t *testing.T) {
+	t.Parallel()
+
+	results := map[string]*resolver.NameserverResponse{
+		nsExample1: {
+			Nameserver: nsExample1,
+			Status:     resolver.StatusNXDomain,
+		},
+		nsExample2: {
+			Nameserver: nsExample2,
+			Status:     resolver.StatusTimeout,
+		},
+		nsExample3: {
+			Nameserver: nsExample3,
+			Status:     resolver.StatusError,
+		},
+	}
+
+	allowed := []string{
+		resolver.StatusNXDomain,
+		resolver.StatusTimeout,
+		resolver.StatusError,
+	}
+
+	assert.Empty(
+		t,
+		unsanctionedStatuses(results, allowed...),
+		"timeout and error are non-answers and are tolerated",
+	)
+
+	// The same silent nameservers do not count towards a quorum.
+	assert.Equal(t, 1, answeredCount(results))
+
+	// An unknown status is treated as silence by answeredCount —
+	// so it retries and fails loudly — and is unsanctioned by the
+	// allowlist rather than quietly permitted.
+	const laterStatus = "some-status-added-later"
+
+	results[nsExample4] = &resolver.NameserverResponse{
+		Nameserver: nsExample4,
+		Status:     laterStatus,
+	}
+
+	assert.Equal(t, 1, answeredCount(results))
+	assert.Equal(
+		t,
+		[]string{nsExample4 + "=" + laterStatus},
+		unsanctionedStatuses(results, allowed...),
+	)
+}

internal/resolver/livedns_test.go

@@ -0,0 +1,385 @@
+package resolver_test
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"slices"
+	"sort"
+	"strings"
+	"testing"
+
+	"sneak.berlin/go/dnswatcher/internal/livednstest"
+	"sneak.berlin/go/dnswatcher/internal/resolver"
+)
+
+// ----------------------------------------------------------------
+// Live DNS test support
+// ----------------------------------------------------------------
+//
+// Tests that look something up in DNS query live DNS servers, never a
+// stand-in; logic that works on record data may be tested on that
+// data with no lookup (see TESTING.md). Each live operation below goes
+// through livednstest.Retry, which bounds how many resolutions are in
+// flight at once and retries an operation that got no answer (see
+// package livednstest).
+//
+// Where an assertion spans several independent nameservers, a quorum
+// is enough: a strict majority answering as expected. A server that
+// fails to answer is tolerated, while a server that answers *wrongly*
+// still fails the test.
+//
+// That tolerance is expressed as an ALLOWLIST of sanctioned statuses,
+// never as a blocklist of known-bad ones. A blocklist bans the one
+// wrong answer its author thought of and silently admits every other
+// status, including any added to the resolver later; an allowlist
+// fails on anything nobody explicitly sanctioned. Silence (timeout,
+// error) is the only thing quorum exists to tolerate. A *wrong
+// answer* — nxdomain for a name that exists, ok for one that does
+// not, nodata for either — is never tolerated at any count.
+
+// minNameservers is the smallest nameserver count a well-run zone is
+// expected to publish.
+const minNameservers = 2
+
+// errLiveNoQuorum reports that too few of a domain's nameservers
+// answered for a quorum assertion to be made.
+var errLiveNoQuorum = errors.New("no nameserver quorum")
+
+// liveQuorum is how many of total nameservers must agree for a
+// multi-nameserver assertion to hold: a strict majority.
+func liveQuorum(total int) int {
+	if total < 1 {
+		return 1
+	}
+
+	return total/2 + 1
+}
+
+// countStatus counts the responses carrying the given status.
+func countStatus(
+	results map[string]*resolver.NameserverResponse,
+	status string,
+) int {
+	n := 0
+
+	for _, resp := range results {
+		if resp.Status == status {
+			n++
+		}
+	}
+
+	return n
+}
+
+// liveAnswerStatuses is the closed set of statuses that count as a
+// nameserver having ANSWERED at all, whether or not the test agrees
+// with the answer. It is deliberately an allowlist: a status added
+// to the resolver later is treated as silence, so it can only ever
+// cause a retry and then a loud failure, never a quiet pass.
+func liveAnswerStatuses() []string {
+	return []string{
+		resolver.StatusOK,
+		resolver.StatusNXDomain,
+		resolver.StatusNoData,
+	}
+}
+
+// answeredCount counts the nameservers that produced an answer of
+// any kind, as opposed to failing or timing out.
+func answeredCount(
+	results map[string]*resolver.NameserverResponse,
+) int {
+	answers := liveAnswerStatuses()
+
+	n := 0
+
+	for _, resp := range results {
+		if slices.Contains(answers, resp.Status) {
+			n++
+		}
+	}
+
+	return n
+}
+
+// unsanctionedStatuses returns "nameserver=status" for every result
+// whose status the caller did not explicitly sanction, sorted for a
+// stable failure message. Callers pass the full closed set they will
+// accept — the expected answer plus whichever non-answers (timeout,
+// error) quorum is allowed to tolerate — so that any status outside
+// it fails the test by name.
+func unsanctionedStatuses(
+	results map[string]*resolver.NameserverResponse,
+	allowed ...string,
+) []string {
+	offenders := make([]string, 0, len(results))
+
+	for ns, resp := range results {
+		if slices.Contains(allowed, resp.Status) {
+			continue
+		}
+
+		offenders = append(
+			offenders, fmt.Sprintf("%s=%s", ns, resp.Status),
+		)
+	}
+
+	sort.Strings(offenders)
+
+	return offenders
+}
+
+// describeStatuses renders per-nameserver statuses for use in
+// assertion failure messages.
+func describeStatuses(
+	results map[string]*resolver.NameserverResponse,
+) string {
+	parts := make([]string, 0, len(results))
+	for ns, resp := range results {
+		parts = append(
+			parts, fmt.Sprintf("%s=%s", ns, resp.Status),
+		)
+	}
+
+	sort.Strings(parts)
+
+	return strings.Join(parts, " ")
+}
+
+// ----------------------------------------------------------------
+// Live operation wrappers
+// ----------------------------------------------------------------
+
+// liveFindAuthoritative resolves a domain's authoritative
+// nameservers, retrying until the delegation chain can be walked.
+func liveFindAuthoritative(
+	t *testing.T,
+	r *resolver.Resolver,
+	domain string,
+) []string {
+	t.Helper()
+
+	var out []string
+
+	livednstest.Retry(
+		t,
+		"FindAuthoritativeNameservers("+domain+")",
+		func(ctx context.Context) error {
+			ns, err := r.FindAuthoritativeNameservers(ctx, domain)
+			if err != nil {
+				return err
+			}
+
+			if len(ns) == 0 {
+				return fmt.Errorf(
+					"%w: %s has no nameservers",
+					livednstest.ErrNoAnswer, domain,
+				)
+			}
+
+			out = ns
+
+			return nil
+		},
+	)
+
+	return out
+}
+
+// liveLookupNS is liveFindAuthoritative through the LookupNS entry
+// point, so that both entry points stay independently exercised.
+func liveLookupNS(
+	t *testing.T,
+	r *resolver.Resolver,
+	domain string,
+) []string {
+	t.Helper()
+
+	var out []string
+
+	livednstest.Retry(
+		t,
+		"LookupNS("+domain+")",
+		func(ctx context.Context) error {
+			ns, err := r.LookupNS(ctx, domain)
+			if err != nil {
+				return err
+			}
+
+			if len(ns) == 0 {
+				return fmt.Errorf(
+					"%w: %s has no nameservers",
+					livednstest.ErrNoAnswer, domain,
+				)
+			}
+
+			out = ns
+
+			return nil
+		},
+	)
+
+	return out
+}
+
+// liveQueryNameserver queries one nameserver, retrying while that
+// nameserver fails to answer. NXDOMAIN and NODATA are answers and
+// are returned to the caller to assert on.
+func liveQueryNameserver(
+	t *testing.T,
+	r *resolver.Resolver,
+	nameserver string,
+	hostname string,
+) *resolver.NameserverResponse {
+	t.Helper()
+
+	what := fmt.Sprintf(
+		"QueryNameserver(%s, %s)", nameserver, hostname,
+	)
+
+	var out *resolver.NameserverResponse
+
+	livednstest.Retry(
+		t,
+		what,
+		func(ctx context.Context) error {
+			resp, err := r.QueryNameserver(
+				ctx, nameserver, hostname,
+			)
+			if err != nil {
+				return err
+			}
+
+			if resp.Status == resolver.StatusTimeout ||
+				resp.Status == resolver.StatusError {
+				return fmt.Errorf(
+					"%w: %s returned %s: %s",
+					livednstest.ErrNoAnswer, nameserver,
+					resp.Status, resp.Error,
+				)
+			}
+
+			out = resp
+
+			return nil
+		},
+	)
+
+	return out
+}
+
+// liveQueryAllNameservers queries every authoritative nameserver
+// for a hostname, retrying until a quorum of them has answered.
+// Individual nameservers that stay silent are left in the result
+// for the caller to account for.
+func liveQueryAllNameservers(
+	t *testing.T,
+	r *resolver.Resolver,
+	hostname string,
+) map[string]*resolver.NameserverResponse {
+	t.Helper()
+
+	var out map[string]*resolver.NameserverResponse
+
+	livednstest.Retry(
+		t,
+		"QueryAllNameservers("+hostname+")",
+		func(ctx context.Context) error {
+			results, err := r.QueryAllNameservers(ctx, hostname)
+			if err != nil {
+				return err
+			}
+
+			if len(results) == 0 {
+				return fmt.Errorf(
+					"%w: no nameservers queried for %s",
+					livednstest.ErrNoAnswer, hostname,
+				)
+			}
+
+			answered := answeredCount(results)
+			if answered < liveQuorum(len(results)) {
+				return fmt.Errorf(
+					"%w: %d of %d answered: %s",
+					errLiveNoQuorum, answered,
+					len(results), describeStatuses(results),
+				)
+			}
+
+			out = results
+
+			return nil
+		},
+	)
+
+	return out
+}
+
+// liveResolveIPs resolves a hostname that is expected to have
+// addresses, retrying until at least one is returned.
+func liveResolveIPs(
+	t *testing.T,
+	r *resolver.Resolver,
+	hostname string,
+) []string {
+	t.Helper()
+
+	var out []string
+
+	livednstest.Retry(
+		t,
+		"ResolveIPAddresses("+hostname+")",
+		func(ctx context.Context) error {
+			ips, err := r.ResolveIPAddresses(ctx, hostname)
+			if err != nil {
+				return err
+			}
+
+			if len(ips) == 0 {
+				return fmt.Errorf(
+					"%w: no addresses for %s",
+					livednstest.ErrNoAnswer, hostname,
+				)
+			}
+
+			out = ips
+
+			return nil
+		},
+	)
+
+	return out
+}
+
+// liveResolveIPsAllowingEmpty resolves a hostname that may legitimately
+// have no addresses, so the empty result is returned rather than
+// retried. Used for names that must not exist; the corresponding
+// QueryAllNameservers test is what proves the nameservers actively
+// said NXDOMAIN rather than merely staying silent.
+func liveResolveIPsAllowingEmpty(
+	t *testing.T,
+	r *resolver.Resolver,
+	hostname string,
+) []string {
+	t.Helper()
+
+	var out []string
+
+	livednstest.Retry(
+		t,
+		"ResolveIPAddresses("+hostname+")",
+		func(ctx context.Context) error {
+			ips, err := r.ResolveIPAddresses(ctx, hostname)
+			if err != nil {
+				return err
+			}
+
+			out = ips
+
+			return nil
+		},
+	)
+
+	return out
+}

internal/resolver/resolver.go

@@ -67,17 +67,4 @@ func NewFromLogger(log *slog.Logger) *Resolver {
 	}
 }
 
-// NewFromLoggerWithClient creates a Resolver with a custom DNS
-// client, useful for testing with mock DNS responses.
-func NewFromLoggerWithClient(
-	log *slog.Logger,
-	client DNSClient,
-) *Resolver {
-	return &Resolver{
-		log:    log,
-		client: client,
-		tcp:    client,
-	}
-}
-
 // Method implementations are in iterative.go.

internal/resolver/resolver_test.go

@@ -10,7 +10,6 @@ import (
 	"testing"
 	"time"
 
-	"github.com/miekg/dns"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 
@@ -32,32 +31,17 @@ func newTestResolver(t *testing.T) *resolver.Resolver {
 	return resolver.NewFromLogger(log)
 }
 
-func testContext(t *testing.T) context.Context {
-	t.Helper()
-
-	ctx, cancel := context.WithTimeout(
-		context.Background(), 60*time.Second,
-	)
-	t.Cleanup(cancel)
-
-	return ctx
-}
-
+// findOneNSForDomain picks one authoritative nameserver to aim a
+// test at. Quorum handling lives in livedns_test.go, and the live-DNS
+// retry and concurrency limit in package livednstest.
 func findOneNSForDomain(
 	t *testing.T,
 	r *resolver.Resolver,
-	ctx context.Context, //nolint:revive // test helper
 	domain string,
 ) string {
 	t.Helper()
 
-	nameservers, err := r.FindAuthoritativeNameservers(
-		ctx, domain,
-	)
-	require.NoError(t, err)
-	require.NotEmpty(t, nameservers)
-
-	return nameservers[0]
+	return liveFindAuthoritative(t, r, domain)[0]
 }
 
 // ----------------------------------------------------------------
@@ -70,13 +54,7 @@ func TestFindAuthoritativeNameservers_ValidDomain(
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-
-	nameservers, err := r.FindAuthoritativeNameservers(
-		ctx, "google.com",
-	)
-	require.NoError(t, err)
-	require.NotEmpty(t, nameservers)
+	nameservers := liveFindAuthoritative(t, r, "google.com")
 
 	hasGoogleNS := false
 
@@ -99,13 +77,9 @@ func TestFindAuthoritativeNameservers_Subdomain(
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
+	nameservers := liveFindAuthoritative(t, r, "www.google.com")
 
-	nameservers, err := r.FindAuthoritativeNameservers(
-		ctx, "www.google.com",
-	)
-	require.NoError(t, err)
-	require.NotEmpty(t, nameservers)
+	assert.NotEmpty(t, nameservers)
 }
 
 func TestFindAuthoritativeNameservers_ReturnsSorted(
@@ -114,12 +88,7 @@ func TestFindAuthoritativeNameservers_ReturnsSorted(
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-
-	nameservers, err := r.FindAuthoritativeNameservers(
-		ctx, "google.com",
-	)
-	require.NoError(t, err)
+	nameservers := liveFindAuthoritative(t, r, "google.com")
 
 	assert.True(
 		t,
@@ -134,17 +103,8 @@ func TestFindAuthoritativeNameservers_Deterministic(
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-
-	first, err := r.FindAuthoritativeNameservers(
-		ctx, "google.com",
-	)
-	require.NoError(t, err)
-
-	second, err := r.FindAuthoritativeNameservers(
-		ctx, "google.com",
-	)
-	require.NoError(t, err)
+	first := liveFindAuthoritative(t, r, "google.com")
+	second := liveFindAuthoritative(t, r, "google.com")
 
 	assert.Equal(t, first, second)
 }
@@ -155,17 +115,8 @@ func TestFindAuthoritativeNameservers_TrailingDot(
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-
-	ns1, err := r.FindAuthoritativeNameservers(
-		ctx, "google.com",
-	)
-	require.NoError(t, err)
-
-	ns2, err := r.FindAuthoritativeNameservers(
-		ctx, "google.com.",
-	)
-	require.NoError(t, err)
+	ns1 := liveFindAuthoritative(t, r, "google.com")
+	ns2 := liveFindAuthoritative(t, r, "google.com.")
 
 	assert.Equal(t, ns1, ns2)
 }
@@ -176,13 +127,7 @@ func TestFindAuthoritativeNameservers_CloudflareDomain(
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-
-	nameservers, err := r.FindAuthoritativeNameservers(
-		ctx, "cloudflare.com",
-	)
-	require.NoError(t, err)
-	require.NotEmpty(t, nameservers)
+	nameservers := liveFindAuthoritative(t, r, "cloudflare.com")
 
 	for _, ns := range nameservers {
 		assert.True(t, strings.HasSuffix(ns, "."),
@@ -199,13 +144,9 @@ func TestQueryNameserver_BasicA(t *testing.T) {
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-	ns := findOneNSForDomain(t, r, ctx, "google.com")
+	ns := findOneNSForDomain(t, r, "google.com")
+	resp := liveQueryNameserver(t, r, ns, "www.google.com")
 
-	resp, err := r.QueryNameserver(
-		ctx, ns, "www.google.com",
-	)
-	require.NoError(t, err)
 	require.NotNil(t, resp)
 
 	assert.Equal(t, resolver.StatusOK, resp.Status)
@@ -222,13 +163,8 @@ func TestQueryNameserver_AAAA(t *testing.T) {
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-	ns := findOneNSForDomain(t, r, ctx, "cloudflare.com")
-
-	resp, err := r.QueryNameserver(
-		ctx, ns, "cloudflare.com",
-	)
-	require.NoError(t, err)
+	ns := findOneNSForDomain(t, r, "cloudflare.com")
+	resp := liveQueryNameserver(t, r, ns, "cloudflare.com")
 
 	aaaaRecords := resp.Records["AAAA"]
 	require.NotEmpty(t, aaaaRecords,
@@ -247,13 +183,8 @@ func TestQueryNameserver_MX(t *testing.T) {
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-	ns := findOneNSForDomain(t, r, ctx, "google.com")
-
-	resp, err := r.QueryNameserver(
-		ctx, ns, "google.com",
-	)
-	require.NoError(t, err)
+	ns := findOneNSForDomain(t, r, "google.com")
+	resp := liveQueryNameserver(t, r, ns, "google.com")
 
 	mxRecords := resp.Records["MX"]
 	require.NotEmpty(t, mxRecords,
@@ -265,13 +196,8 @@ func TestQueryNameserver_TXT(t *testing.T) {
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-	ns := findOneNSForDomain(t, r, ctx, "google.com")
-
-	resp, err := r.QueryNameserver(
-		ctx, ns, "google.com",
-	)
-	require.NoError(t, err)
+	ns := findOneNSForDomain(t, r, "google.com")
+	resp := liveQueryNameserver(t, r, ns, "google.com")
 
 	txtRecords := resp.Records["TXT"]
 	require.NotEmpty(t, txtRecords,
@@ -297,14 +223,10 @@ func TestQueryNameserver_NXDomain(t *testing.T) {
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-	ns := findOneNSForDomain(t, r, ctx, "google.com")
-
-	resp, err := r.QueryNameserver(
-		ctx, ns,
-		"this-surely-does-not-exist-xyz.google.com",
+	ns := findOneNSForDomain(t, r, "google.com")
+	resp := liveQueryNameserver(
+		t, r, ns, "this-surely-does-not-exist-xyz.google.com",
 	)
-	require.NoError(t, err)
 
 	assert.Equal(t, resolver.StatusNXDomain, resp.Status)
 }
@@ -313,13 +235,8 @@ func TestQueryNameserver_RecordsSorted(t *testing.T) {
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-	ns := findOneNSForDomain(t, r, ctx, "google.com")
-
-	resp, err := r.QueryNameserver(
-		ctx, ns, "google.com",
-	)
-	require.NoError(t, err)
+	ns := findOneNSForDomain(t, r, "google.com")
+	resp := liveQueryNameserver(t, r, ns, "google.com")
 
 	for recordType, values := range resp.Records {
 		assert.True(
@@ -336,13 +253,8 @@ func TestQueryNameserver_ResponseIncludesNameserver(
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-	ns := findOneNSForDomain(t, r, ctx, "cloudflare.com")
-
-	resp, err := r.QueryNameserver(
-		ctx, ns, "cloudflare.com",
-	)
-	require.NoError(t, err)
+	ns := findOneNSForDomain(t, r, "cloudflare.com")
+	resp := liveQueryNameserver(t, r, ns, "cloudflare.com")
 
 	assert.Equal(t, ns, resp.Nameserver)
 }
@@ -353,14 +265,10 @@ func TestQueryNameserver_EmptyRecordsOnNXDomain(
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-	ns := findOneNSForDomain(t, r, ctx, "google.com")
-
-	resp, err := r.QueryNameserver(
-		ctx, ns,
-		"this-surely-does-not-exist-xyz.google.com",
+	ns := findOneNSForDomain(t, r, "google.com")
+	resp := liveQueryNameserver(
+		t, r, ns, "this-surely-does-not-exist-xyz.google.com",
 	)
-	require.NoError(t, err)
 
 	totalRecords := 0
 	for _, values := range resp.Records {
@@ -374,18 +282,9 @@ func TestQueryNameserver_TrailingDotHandling(t *testing.T) {
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-	ns := findOneNSForDomain(t, r, ctx, "google.com")
-
-	resp1, err := r.QueryNameserver(
-		ctx, ns, "google.com",
-	)
-	require.NoError(t, err)
-
-	resp2, err := r.QueryNameserver(
-		ctx, ns, "google.com.",
-	)
-	require.NoError(t, err)
+	ns := findOneNSForDomain(t, r, "google.com")
+	resp1 := liveQueryNameserver(t, r, ns, "google.com")
+	resp2 := liveQueryNameserver(t, r, ns, "google.com.")
 
 	assert.Equal(t, resp1.Status, resp2.Status)
 }
@@ -398,15 +297,9 @@ func TestQueryAllNameservers_ReturnsAllNS(t *testing.T) {
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
+	results := liveQueryAllNameservers(t, r, "google.com")
 
-	results, err := r.QueryAllNameservers(
-		ctx, "google.com",
-	)
-	require.NoError(t, err)
-	require.NotEmpty(t, results)
-
-	assert.GreaterOrEqual(t, len(results), 2)
+	assert.GreaterOrEqual(t, len(results), minNameservers)
 
 	for ns, resp := range results {
 		assert.Equal(t, ns, resp.Nameserver)
@@ -417,20 +310,37 @@ func TestQueryAllNameservers_AllReturnOK(t *testing.T) {
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
+	results := liveQueryAllNameservers(t, r, "google.com")
 
-	results, err := r.QueryAllNameservers(
-		ctx, "google.com",
+	// A quorum, not unanimity: one authoritative server being
+	// slow or rate-limiting us is a property of the live
+	// internet, not a resolver defect.
+	assert.GreaterOrEqual(
+		t,
+		countStatus(results, resolver.StatusOK),
+		liveQuorum(len(results)),
+		"a quorum of nameservers should answer OK: %s",
+		describeStatuses(results),
 	)
-	require.NoError(t, err)
 
-	for ns, resp := range results {
-		assert.Equal(
-			t, resolver.StatusOK, resp.Status,
-			"NS %s should return OK", ns,
+	// Quorum tolerates SILENCE only. Every individual result must
+	// be either the expected answer or a non-answer: ok, timeout
+	// or error, and nothing else. Stated as a closed allowlist so
+	// that a wrong answer no one thought to ban — nxdomain and
+	// nodata today, any status added later — fails here rather
+	// than sliding through under the quorum.
+	assert.Empty(
+		t,
+		unsanctionedStatuses(
+			results,
+			resolver.StatusOK,
+			resolver.StatusTimeout,
+			resolver.StatusError,
+		),
+		"every nameserver must answer OK or not answer at all: %s",
+		describeStatuses(results),
 	)
 }
-}
 
 func TestQueryAllNameservers_NXDomainFromAllNS(
 	t *testing.T,
@@ -438,20 +348,34 @@ func TestQueryAllNameservers_NXDomainFromAllNS(
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-
-	results, err := r.QueryAllNameservers(
-		ctx,
-		"this-surely-does-not-exist-xyz.google.com",
+	results := liveQueryAllNameservers(
+		t, r, "this-surely-does-not-exist-xyz.google.com",
 	)
-	require.NoError(t, err)
 
-	for ns, resp := range results {
-		assert.Equal(
-			t, resolver.StatusNXDomain, resp.Status,
-			"NS %s should return nxdomain", ns,
+	assert.GreaterOrEqual(
+		t,
+		countStatus(results, resolver.StatusNXDomain),
+		liveQuorum(len(results)),
+		"a quorum of nameservers should report NXDOMAIN: %s",
+		describeStatuses(results),
+	)
+
+	// Silence is tolerated; any actual answer other than NXDOMAIN
+	// is not. Closed allowlist for the same reason as above: a
+	// server answering `ok` or `nodata` for a name that must not
+	// exist is a wrong answer, not a slow one.
+	assert.Empty(
+		t,
+		unsanctionedStatuses(
+			results,
+			resolver.StatusNXDomain,
+			resolver.StatusTimeout,
+			resolver.StatusError,
+		),
+		"every nameserver must report NXDOMAIN or not answer "+
+			"at all: %s",
+		describeStatuses(results),
 	)
-	}
 }
 
 // ----------------------------------------------------------------
@@ -462,11 +386,7 @@ func TestLookupNS_ValidDomain(t *testing.T) {
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-
-	nameservers, err := r.LookupNS(ctx, "google.com")
-	require.NoError(t, err)
-	require.NotEmpty(t, nameservers)
+	nameservers := liveLookupNS(t, r, "google.com")
 
 	for _, ns := range nameservers {
 		assert.True(t, strings.HasSuffix(ns, "."),
@@ -479,10 +399,7 @@ func TestLookupNS_Sorted(t *testing.T) {
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-
-	nameservers, err := r.LookupNS(ctx, "google.com")
-	require.NoError(t, err)
+	nameservers := liveLookupNS(t, r, "google.com")
 
 	assert.True(t, sort.StringsAreSorted(nameservers))
 }
@@ -491,15 +408,8 @@ func TestLookupNS_MatchesFindAuthoritative(t *testing.T) {
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-
-	fromLookup, err := r.LookupNS(ctx, "google.com")
-	require.NoError(t, err)
-
-	fromFind, err := r.FindAuthoritativeNameservers(
-		ctx, "google.com",
-	)
-	require.NoError(t, err)
+	fromLookup := liveLookupNS(t, r, "google.com")
+	fromFind := liveFindAuthoritative(t, r, "google.com")
 
 	assert.Equal(t, fromFind, fromLookup)
 }
@@ -512,11 +422,7 @@ func TestResolveIPAddresses_ReturnsIPs(t *testing.T) {
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-
-	ips, err := r.ResolveIPAddresses(ctx, "google.com")
-	require.NoError(t, err)
-	require.NotEmpty(t, ips)
+	ips := liveResolveIPs(t, r, "google.com")
 
 	for _, ip := range ips {
 		parsed := net.ParseIP(ip)
@@ -530,10 +436,7 @@ func TestResolveIPAddresses_Deduplicated(t *testing.T) {
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-
-	ips, err := r.ResolveIPAddresses(ctx, "google.com")
-	require.NoError(t, err)
+	ips := liveResolveIPs(t, r, "google.com")
 
 	seen := make(map[string]bool)
 
@@ -547,10 +450,7 @@ func TestResolveIPAddresses_Sorted(t *testing.T) {
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-
-	ips, err := r.ResolveIPAddresses(ctx, "google.com")
-	require.NoError(t, err)
+	ips := liveResolveIPs(t, r, "google.com")
 
 	assert.True(t, sort.StringsAreSorted(ips))
 }
@@ -561,13 +461,10 @@ func TestResolveIPAddresses_NXDomainReturnsEmpty(
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
-
-	ips, err := r.ResolveIPAddresses(
-		ctx,
-		"this-surely-does-not-exist-xyz.google.com",
+	ips := liveResolveIPsAllowingEmpty(
+		t, r, "this-surely-does-not-exist-xyz.google.com",
 	)
-	require.NoError(t, err)
+
 	assert.Empty(t, ips)
 }
 
@@ -575,11 +472,9 @@ func TestResolveIPAddresses_CloudflareDomain(t *testing.T) {
 	t.Parallel()
 
 	r := newTestResolver(t)
-	ctx := testContext(t)
+	ips := liveResolveIPs(t, r, "cloudflare.com")
 
-	ips, err := r.ResolveIPAddresses(ctx, "cloudflare.com")
-	require.NoError(t, err)
-	require.NotEmpty(t, ips)
+	assert.NotEmpty(t, ips)
 }
 
 // ----------------------------------------------------------------
@@ -630,21 +525,18 @@ func TestQueryAllNameservers_ContextCanceled(t *testing.T) {
 func TestQueryNameserverIP_Timeout(t *testing.T) {
 	t.Parallel()
 
-	log := slog.New(slog.NewTextHandler(
-		os.Stderr,
-		&slog.HandlerOptions{Level: slog.LevelDebug},
-	))
-
-	r := resolver.NewFromLoggerWithClient(
-		log, &timeoutClient{},
-	)
+	r := newTestResolver(t)
 
+	// Nothing answers at 192.0.2.1, a documentation address. The
+	// resolver tries each query twice, and the first try gives up
+	// after two seconds. A deadline that ends during the first try
+	// makes the status vary from run to run between nodata and
+	// timeout, so the deadline must outlast the first try.
 	ctx, cancel := context.WithTimeout(
-		context.Background(), 10*time.Second,
+		context.Background(), 3*time.Second,
 	)
 	t.Cleanup(cancel)
 
-	// Query any IP — the client always returns a timeout error.
 	resp, err := r.QueryNameserverIP(
 		ctx, "unreachable.test.", "192.0.2.1",
 		"example.com",
@@ -655,27 +547,6 @@ func TestQueryNameserverIP_Timeout(t *testing.T) {
 	assert.NotEmpty(t, resp.Error)
 }
 
-// timeoutClient simulates DNS timeout errors for testing.
-type timeoutClient struct{}
-
-func (c *timeoutClient) ExchangeContext(
-	_ context.Context,
-	_ *dns.Msg,
-	_ string,
-) (*dns.Msg, time.Duration, error) {
-	return nil, 0, &net.OpError{
-		Op:  "read",
-		Net: "udp",
-		Err: &timeoutError{},
-	}
-}
-
-type timeoutError struct{}
-
-func (e *timeoutError) Error() string   { return "i/o timeout" }
-func (e *timeoutError) Timeout() bool   { return true }
-func (e *timeoutError) Temporary() bool { return true }
-
 func TestResolveIPAddresses_ContextCanceled(t *testing.T) {
 	t.Parallel()
 

internal/server/export_test.go

@@ -0,0 +1,24 @@
+package server
+
+import (
+	"net/http"
+	"time"
+)
+
+// RequestTimeout exports the handler execution budget applied by
+// chimw.Timeout in SetupRoutes, so tests can assert the relationship
+// between it and the server's WriteTimeout.
+const RequestTimeout time.Duration = requestTimeout
+
+// SetListenPort overrides the port Run binds. A test uses it to hand
+// Run an unbindable port so ListenAndServe fails immediately and Run
+// returns after storing its http.Server.
+func SetListenPort(s *Server, port int) {
+	s.port = port
+}
+
+// HTTPServerOf returns the http.Server that Run built and stored, so a
+// test can inspect the timeouts the running server actually carries.
+func HTTPServerOf(s *Server) *http.Server {
+	return s.httpServer
+}

internal/server/routes.go

@@ -21,6 +21,7 @@ func (s *Server) SetupRoutes() {
 	// Global middleware
 	s.router.Use(chimw.Recoverer)
 	s.router.Use(chimw.RequestID)
+	s.router.Use(s.mw.SecurityHeaders())
 	s.router.Use(s.mw.Logging())
 	s.router.Use(s.mw.CORS())
 	s.router.Use(chimw.Timeout(requestTimeout))

internal/server/server.go

@@ -33,8 +33,52 @@ type Params struct {
 // shutdownTimeout is how long to wait for graceful shutdown.
 const shutdownTimeout = 30 * time.Second
 
-// readHeaderTimeout is the max duration for reading request headers.
-const readHeaderTimeout = 10 * time.Second
+// Socket-level timeouts for the HTTP server.
+//
+// These bound time spent on the connection itself and are a distinct
+// control from the per-request handler budget enforced by
+// chimw.Timeout(requestTimeout) in routes.go: that one cancels the
+// request context after requestTimeout but never touches the socket,
+// so without the values below a peer can hold a connection open
+// forever (slowloris, unreaped keep-alives).
+//
+// The one hard constraint between the two controls is
+// writeTimeout > requestTimeout. net/http arms the write deadline
+// once the request headers have been read, so on a plaintext
+// connection it covers handler execution AND the response flush. If
+// writeTimeout were <= requestTimeout the server would sever the
+// connection before a handler that legitimately consumed its full
+// budget could emit anything, making the 60s budget unreachable in
+// practice. The margin between them is the response-flush allowance.
+//
+// The only clients of this service are browsers loading the dashboard
+// and a Prometheus scraper; the values are sized for those.
+const (
+	// readHeaderTimeout is the max duration for reading request
+	// headers.
+	readHeaderTimeout = 10 * time.Second
+
+	// readTimeout bounds reading the entire request, headers plus
+	// body. Every route here is a GET with no body, so this only
+	// ever needs to cover headers; the extra 5s over
+	// readHeaderTimeout is slack, not a real allowance, and keeps a
+	// body dribbled one byte at a time from holding the read side
+	// open indefinitely.
+	readTimeout = 15 * time.Second
+
+	// writeTimeout must exceed the requestTimeout handler budget
+	// (60s) per the note above. The 15s difference is the allowance
+	// for flushing a completed response to a slow client.
+	writeTimeout = 75 * time.Second
+
+	// idleTimeout reaps keep-alive connections between requests. It
+	// is deliberately longer than the common Prometheus scrape
+	// intervals (15s/30s/60s) so the scraper reuses its connection
+	// rather than reconnecting every cycle, while a browser tab
+	// left open on the dashboard stops occupying a connection
+	// within two minutes of going quiet.
+	idleTimeout = 120 * time.Second
+)
 
 // Server is the HTTP server.
 type Server struct {
@@ -76,16 +120,29 @@ func New(
 	return srv, nil
 }
 
+// newHTTPServer builds the listening http.Server with every
+// socket-level timeout set. All four are set deliberately: a zero
+// value in net/http means "no limit", not "some default".
+func newHTTPServer(
+	listenAddr string,
+	handler http.Handler,
+) *http.Server {
+	return &http.Server{
+		Addr:              listenAddr,
+		Handler:           handler,
+		ReadTimeout:       readTimeout,
+		ReadHeaderTimeout: readHeaderTimeout,
+		WriteTimeout:      writeTimeout,
+		IdleTimeout:       idleTimeout,
+	}
+}
+
 // Run starts the HTTP server.
 func (s *Server) Run() {
 	s.SetupRoutes()
 
 	listenAddr := fmt.Sprintf(":%d", s.port)
-	s.httpServer = &http.Server{
-		Addr:              listenAddr,
-		Handler:           s,
-		ReadHeaderTimeout: readHeaderTimeout,
-	}
+	s.httpServer = newHTTPServer(listenAddr, s)
 
 	s.log.Info("http server starting", "addr", listenAddr)
 

internal/server/server_test.go

@@ -0,0 +1,130 @@
+package server_test
+
+import (
+	"testing"
+
+	"github.com/spf13/viper"
+	"go.uber.org/fx"
+
+	"sneak.berlin/go/dnswatcher/internal/config"
+	"sneak.berlin/go/dnswatcher/internal/globals"
+	"sneak.berlin/go/dnswatcher/internal/handlers"
+	"sneak.berlin/go/dnswatcher/internal/healthcheck"
+	"sneak.berlin/go/dnswatcher/internal/logger"
+	"sneak.berlin/go/dnswatcher/internal/middleware"
+	"sneak.berlin/go/dnswatcher/internal/notify"
+	"sneak.berlin/go/dnswatcher/internal/server"
+	"sneak.berlin/go/dnswatcher/internal/state"
+)
+
+// buildServer wires a *server.Server exactly as cmd/dnswatcher does,
+// minus the watcher/resolver subtree that would touch live DNS. fx
+// builds the object graph but the lifecycle is never started, so no
+// OnStart hook runs and nothing listens or resolves. The caller must
+// first configure viper (config.New reads it), which is also why the
+// caller cannot run in parallel.
+func buildServer(t *testing.T) *server.Server {
+	t.Helper()
+
+	var srv *server.Server
+
+	app := fx.New(
+		fx.NopLogger,
+		fx.Provide(
+			globals.New,
+			logger.New,
+			config.New,
+			state.New,
+			healthcheck.New,
+			notify.New,
+			middleware.New,
+			handlers.New,
+			server.New,
+		),
+		fx.Populate(&srv),
+	)
+
+	err := app.Err()
+	if err != nil {
+		t.Fatalf("building server graph: %v", err)
+	}
+
+	return srv
+}
+
+// TestRunWiresSocketTimeouts pins that the http.Server the running
+// server actually serves — the one Run builds and hands to
+// ListenAndServe — carries every socket-level timeout, plus the two
+// relationships the values must satisfy.
+//
+// Run is driven to completion with an unbindable port: it builds and
+// stores s.httpServer, then ListenAndServe fails at once and Run
+// returns without ever listening. The assertions run in the same
+// goroutine after Run returns, so reading s.httpServer is free of any
+// data race. Nothing here measures elapsed time.
+//
+// ReadTimeout must be at least ReadHeaderTimeout. net/http reads the
+// headers under ReadHeaderTimeout, then sets the read deadline for the
+// rest of the request to ReadTimeout, counted from when it started
+// reading the request. If ReadTimeout were smaller, a request whose
+// headers arrived after ReadTimeout but within ReadHeaderTimeout would
+// get a read deadline that had already passed, so reading its body
+// would fail at once.
+func TestRunWiresSocketTimeouts(t *testing.T) {
+	// Sets an env var and touches viper global state, so like the
+	// config tests it cannot use t.Parallel.
+	viper.Reset()
+	t.Setenv("DNSWATCHER_TARGETS", "example.com")
+
+	srv := buildServer(t)
+	server.SetListenPort(srv, -1)
+
+	srv.Run()
+
+	hs := server.HTTPServerOf(srv)
+	if hs == nil {
+		t.Fatal("Run did not build an http.Server")
+	}
+
+	if hs.ReadTimeout <= 0 {
+		t.Errorf("ReadTimeout must be non-zero, got %v", hs.ReadTimeout)
+	}
+
+	if hs.ReadHeaderTimeout <= 0 {
+		t.Errorf(
+			"ReadHeaderTimeout must be non-zero, got %v",
+			hs.ReadHeaderTimeout,
+		)
+	}
+
+	if hs.WriteTimeout <= 0 {
+		t.Errorf("WriteTimeout must be non-zero, got %v", hs.WriteTimeout)
+	}
+
+	if hs.IdleTimeout <= 0 {
+		t.Errorf("IdleTimeout must be non-zero, got %v", hs.IdleTimeout)
+	}
+
+	if hs.WriteTimeout <= server.RequestTimeout {
+		t.Errorf(
+			"WriteTimeout (%v) must exceed handler budget (%v)",
+			hs.WriteTimeout,
+			server.RequestTimeout,
+		)
+	}
+
+	if hs.ReadTimeout < hs.ReadHeaderTimeout {
+		t.Errorf(
+			"ReadTimeout (%v) must be >= ReadHeaderTimeout (%v)",
+			hs.ReadTimeout,
+			hs.ReadHeaderTimeout,
+		)
+	}
+
+	if hs.Handler != srv {
+		t.Errorf(
+			"Run wired handler %T, want the *server.Server",
+			hs.Handler,
+		)
+	}
+}

internal/state/state.go

@@ -148,6 +148,11 @@ func New(
 
 	lifecycle.Append(fx.Hook{
 		OnStart: func(_ context.Context) error {
+			err := state.checkDataDirWritable()
+			if err != nil {
+				return err
+			}
+
 			return state.Load()
 		},
 		OnStop: func(_ context.Context) error {
@@ -345,3 +350,27 @@ func (s *State) GetCertificateState(
 
 	return cs, ok
 }
+
+// checkDataDirWritable creates the data directory if needed, then writes
+// and removes the temp file that Save uses. It runs at startup so that an
+// unwritable directory stops the process, instead of the process running
+// with every save failing and only logged.
+func (s *State) checkDataDirWritable() error {
+	dir := s.config.DataDir
+	tmpPath := s.config.StatePath() + ".tmp"
+
+	err := os.MkdirAll(dir, dirPermissions)
+	if err == nil {
+		err = os.WriteFile(tmpPath, nil, filePermissions)
+	}
+
+	if err == nil {
+		err = os.Remove(tmpPath)
+	}
+
+	if err != nil {
+		return fmt.Errorf("data directory %s is not writable: %w", dir, err)
+	}
+
+	return nil
+}

internal/state/state_test.go

@@ -4,10 +4,16 @@ import (
 	"encoding/json"
 	"os"
 	"path/filepath"
+	"strings"
 	"sync"
 	"testing"
 	"time"
 
+	"go.uber.org/fx/fxtest"
+
+	"sneak.berlin/go/dnswatcher/internal/config"
+	"sneak.berlin/go/dnswatcher/internal/globals"
+	"sneak.berlin/go/dnswatcher/internal/logger"
 	"sneak.berlin/go/dnswatcher/internal/state"
 )
 
@@ -493,6 +499,107 @@ func TestSaveWritePermissionError(t *testing.T) {
 	}
 }
 
+// startState builds a State through the real constructor and runs its
+// startup hook against dataDir, returning the startup error.
+func startState(t *testing.T, dataDir string) error {
+	t.Helper()
+
+	g, err := globals.New(nil)
+	if err != nil {
+		t.Fatalf("globals.New: %v", err)
+	}
+
+	log, err := logger.New(nil, logger.Params{Globals: g})
+	if err != nil {
+		t.Fatalf("logger.New: %v", err)
+	}
+
+	lifecycle := fxtest.NewLifecycle(t)
+
+	_, err = state.New(lifecycle, state.Params{
+		Logger: log,
+		Config: &config.Config{DataDir: dataDir},
+	})
+	if err != nil {
+		t.Fatalf("state.New: %v", err)
+	}
+
+	return lifecycle.Start(t.Context())
+}
+
+// TestStartupFailsWhenDataDirNotWritable verifies that startup stops
+// with an error naming the data directory when it cannot be written.
+// The directory's parent is a regular file, which also fails as root.
+func TestStartupFailsWhenDataDirNotWritable(t *testing.T) {
+	t.Parallel()
+
+	parent := filepath.Join(t.TempDir(), "file")
+
+	err := os.WriteFile(parent, nil, 0o600)
+	if err != nil {
+		t.Fatalf("writing file: %v", err)
+	}
+
+	dataDir := filepath.Join(parent, "data")
+
+	err = startState(t, dataDir)
+	if err == nil {
+		t.Fatal("startup should fail when the data directory is not writable")
+	}
+
+	want := "data directory " + dataDir + " is not writable"
+	if !strings.Contains(err.Error(), want) {
+		t.Errorf("startup error %q does not contain %q", err, want)
+	}
+}
+
+// TestStartupFailsWhenExistingDataDirNotWritable verifies that startup
+// stops when the data directory exists but the temp file that saving uses
+// cannot be written in it. A directory sitting at the temp file's path
+// makes that write fail, which also holds as root.
+func TestStartupFailsWhenExistingDataDirNotWritable(t *testing.T) {
+	t.Parallel()
+
+	dataDir := t.TempDir()
+
+	err := os.Mkdir(filepath.Join(dataDir, "state.json.tmp"), 0o700)
+	if err != nil {
+		t.Fatalf("creating directory: %v", err)
+	}
+
+	err = startState(t, dataDir)
+	if err == nil {
+		t.Fatal("startup should fail when the data directory is not writable")
+	}
+
+	want := "data directory " + dataDir + " is not writable"
+	if !strings.Contains(err.Error(), want) {
+		t.Errorf("startup error %q does not contain %q", err, want)
+	}
+}
+
+// TestStartupCreatesDataDir verifies that startup creates a missing
+// data directory and leaves nothing behind in it.
+func TestStartupCreatesDataDir(t *testing.T) {
+	t.Parallel()
+
+	dataDir := filepath.Join(t.TempDir(), "data")
+
+	err := startState(t, dataDir)
+	if err != nil {
+		t.Fatalf("startup error: %v", err)
+	}
+
+	entries, err := os.ReadDir(dataDir)
+	if err != nil {
+		t.Fatalf("reading data directory: %v", err)
+	}
+
+	if len(entries) != 0 {
+		t.Errorf("startup left %d entries in the data directory", len(entries))
+	}
+}
+
 // TestPortStateUnmarshalJSON_NewFormat verifies deserialization of the
 // current multi-hostname format.
 func TestPortStateUnmarshalJSON_NewFormat(t *testing.T) {

internal/watcher/export_test.go

@@ -0,0 +1,25 @@
+package watcher
+
+import (
+	"context"
+
+	"sneak.berlin/go/dnswatcher/internal/state"
+)
+
+// NewlyDisagreeingPairs exports newlyDisagreeingPairs for testing.
+func NewlyDisagreeingPairs(
+	prev *state.HostnameState,
+	current map[string]map[string][]string,
+) [][2]string {
+	return newlyDisagreeingPairs(prev, current)
+}
+
+// DetectHostnameChanges exports detectHostnameChanges for testing.
+func (w *Watcher) DetectHostnameChanges(
+	ctx context.Context,
+	hostname string,
+	prev *state.HostnameState,
+	current map[string]map[string][]string,
+) {
+	w.detectHostnameChanges(ctx, hostname, prev, current)
+}

internal/watcher/inconsistency_test.go

@@ -0,0 +1,182 @@
+package watcher_test
+
+import (
+	"slices"
+	"testing"
+
+	"sneak.berlin/go/dnswatcher/internal/state"
+	"sneak.berlin/go/dnswatcher/internal/watcher"
+)
+
+const (
+	host = "www.example.net"
+	nsA  = "a.ns.example.net."
+	nsB  = "b.ns.example.net."
+	nsC  = "c.ns.example.net."
+	ip1  = "192.0.2.1"
+	ip2  = "192.0.2.2"
+	ip3  = "192.0.2.3"
+)
+
+// hostnameState builds the state a check with these records leaves behind.
+func hostnameState(
+	records map[string]map[string][]string,
+) *state.HostnameState {
+	hs := &state.HostnameState{
+		RecordsByNameserver: make(map[string]*state.NameserverRecordState),
+	}
+
+	for ns, recs := range records {
+		hs.RecordsByNameserver[ns] = &state.NameserverRecordState{
+			Records: recs,
+			Status:  "ok",
+		}
+	}
+
+	return hs
+}
+
+func TestNewlyDisagreeingPairs(t *testing.T) {
+	t.Parallel()
+
+	onlyA := map[string]map[string][]string{nsA: {"A": {ip1}}}
+	agree := map[string]map[string][]string{nsA: {"A": {ip1}}, nsB: {"A": {ip1}}}
+	disagree := map[string]map[string][]string{nsA: {"A": {ip1}}, nsB: {"A": {ip2}}}
+	alert := [][2]string{{nsA, nsB}}
+
+	// b already disagrees with a and c; then c changes, so a and c,
+	// which agreed, now differ.
+	bDiffers := map[string]map[string][]string{
+		nsA: {"A": {ip1}}, nsB: {"A": {ip2}}, nsC: {"A": {ip1}},
+	}
+	cChanges := map[string]map[string][]string{
+		nsA: {"A": {ip1}}, nsB: {"A": {ip2}}, nsC: {"A": {ip3}},
+	}
+
+	// Each case starts from the state loaded at startup and runs the
+	// checks in order; want[i] is what check i alerts for.
+	tests := []struct {
+		name   string
+		loaded map[string]map[string][]string
+		checks []map[string]map[string][]string
+		want   [][][2]string
+	}{
+		{
+			name:   "disagreement persisting across checks alerts once",
+			loaded: agree,
+			checks: []map[string]map[string][]string{disagree, disagree, disagree},
+			want:   [][][2]string{alert, nil, nil},
+		},
+		{
+			name:   "disagreement starting on a later check alerts on it",
+			loaded: agree,
+			checks: []map[string]map[string][]string{agree, agree, disagree},
+			want:   [][][2]string{nil, nil, alert},
+		},
+		{
+			name:   "disagreement in the loaded state does not alert",
+			loaded: disagree,
+			checks: []map[string]map[string][]string{disagree, disagree},
+			want:   [][][2]string{nil, nil},
+		},
+		{
+			name:   "nameserver new on the first check and disagreeing alerts once",
+			loaded: onlyA,
+			checks: []map[string]map[string][]string{disagree, disagree},
+			want:   [][][2]string{alert, nil},
+		},
+		{
+			name:   "disagreement after agreeing again alerts again",
+			loaded: agree,
+			checks: []map[string]map[string][]string{disagree, agree, disagree},
+			want:   [][][2]string{alert, nil, alert},
+		},
+		{
+			name:   "new disagreement while another nameserver differs alerts",
+			loaded: bDiffers,
+			checks: []map[string]map[string][]string{cChanges, cChanges},
+			want:   [][][2]string{{{nsA, nsC}}, nil},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			prev := hostnameState(tt.loaded)
+
+			for i, current := range tt.checks {
+				got := watcher.NewlyDisagreeingPairs(prev, current)
+				if !slices.Equal(got, tt.want[i]) {
+					t.Errorf(
+						"check %d: alerted for %v, want %v",
+						i, got, tt.want[i],
+					)
+				}
+
+				prev = hostnameState(current)
+			}
+		})
+	}
+}
+
+func TestInconsistencyAlert(t *testing.T) {
+	t.Parallel()
+
+	onlyA := map[string]map[string][]string{nsA: {"A": {ip1}}}
+	agree := map[string]map[string][]string{nsA: {"A": {ip1}}, nsB: {"A": {ip1}}}
+	disagree := map[string]map[string][]string{nsA: {"A": {ip1}}, nsB: {"A": {ip2}}}
+
+	// Each case starts from the state loaded at startup and then sees
+	// the nameservers disagree on three checks in a row.
+	tests := []struct {
+		name   string
+		loaded map[string]map[string][]string
+		want   int
+	}{
+		{
+			name:   "disagreement lasting several checks alerts once",
+			loaded: agree,
+			want:   1,
+		},
+		{
+			name:   "disagreement in the loaded state does not alert",
+			loaded: disagree,
+			want:   0,
+		},
+		{
+			name:   "nameserver new on the first check and disagreeing alerts once",
+			loaded: onlyA,
+			want:   1,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			// The hostname change detection uses only the notifier.
+			notifier := &mockNotifier{}
+			w := watcher.NewForTest(nil, nil, nil, nil, nil, notifier)
+
+			prev := hostnameState(tt.loaded)
+
+			for range 3 {
+				w.DetectHostnameChanges(t.Context(), host, prev, disagree)
+				prev = hostnameState(disagree)
+			}
+
+			got := 0
+
+			for _, n := range notifier.getNotifications() {
+				if n.Title == "Inconsistency: "+host {
+					got++
+				}
+			}
+
+			if got != tt.want {
+				t.Errorf("sent %d inconsistency alerts, want %d", got, tt.want)
+			}
+		})
+	}
+}

internal/watcher/watcher.go

@@ -366,7 +366,7 @@ func (w *Watcher) detectHostnameChanges(
 ) {
 	w.detectRecordChanges(ctx, hostname, prev, current)
 	w.detectNSDisappearances(ctx, hostname, prev, current)
-	w.detectInconsistencies(ctx, hostname, current)
+	w.detectInconsistencies(ctx, hostname, prev, current)
 }
 
 func (w *Watcher) detectRecordChanges(
@@ -448,22 +448,11 @@ func (w *Watcher) detectNSDisappearances(
 func (w *Watcher) detectInconsistencies(
 	ctx context.Context,
 	hostname string,
+	prev *state.HostnameState,
 	current map[string]map[string][]string,
 ) {
-	nameservers := make([]string, 0, len(current))
-	for ns := range current {
-		nameservers = append(nameservers, ns)
-	}
-
-	sort.Strings(nameservers)
-
-	for i := range len(nameservers) - 1 {
-		ns1 := nameservers[i]
-		ns2 := nameservers[i+1]
-
-		if recordsEqual(current[ns1], current[ns2]) {
-			continue
-		}
+	for _, pair := range newlyDisagreeingPairs(prev, current) {
+		ns1, ns2 := pair[0], pair[1]
 
 		msg := fmt.Sprintf(
 			"Hostname: %s\n%s: %v\n%s: %v",
@@ -481,6 +470,43 @@ func (w *Watcher) detectInconsistencies(
 	}
 }
 
+// newlyDisagreeingPairs returns every pair of nameservers whose records
+// differ in current, in sorted order of name, except pairs where both
+// nameservers were in prev and already differed there. A nameserver
+// missing from prev is paired with every nameserver it differs from.
+func newlyDisagreeingPairs(
+	prev *state.HostnameState,
+	current map[string]map[string][]string,
+) [][2]string {
+	nameservers := make([]string, 0, len(current))
+	for ns := range current {
+		nameservers = append(nameservers, ns)
+	}
+
+	sort.Strings(nameservers)
+
+	var pairs [][2]string
+
+	for i, ns1 := range nameservers {
+		for _, ns2 := range nameservers[i+1:] {
+			if recordsEqual(current[ns1], current[ns2]) {
+				continue
+			}
+
+			prev1, ok1 := prev.RecordsByNameserver[ns1]
+			prev2, ok2 := prev.RecordsByNameserver[ns2]
+
+			if ok1 && ok2 && !recordsEqual(prev1.Records, prev2.Records) {
+				continue
+			}
+
+			pairs = append(pairs, [2]string{ns1, ns2})
+		}
+	}
+
+	return pairs
+}
+
 func (w *Watcher) checkAllPorts(ctx context.Context) {
 	// Phase 1: Build current IP:port → hostname associations
 	// from fresh DNS data.

internal/watcher/watcher_test.go

@@ -2,151 +2,77 @@ package watcher_test
 
 import (
 	"context"
-	"errors"
 	"fmt"
+	"log/slog"
+	"slices"
 	"sync"
 	"testing"
 	"time"
 
 	"sneak.berlin/go/dnswatcher/internal/config"
+	"sneak.berlin/go/dnswatcher/internal/livednstest"
 	"sneak.berlin/go/dnswatcher/internal/portcheck"
+	"sneak.berlin/go/dnswatcher/internal/resolver"
 	"sneak.berlin/go/dnswatcher/internal/state"
 	"sneak.berlin/go/dnswatcher/internal/tlscheck"
 	"sneak.berlin/go/dnswatcher/internal/watcher"
 )
 
-// errNotFound is returned when mock data is missing.
-var errNotFound = errors.New("not found")
-
-// Fixture values shared across tests.
+// The watcher looks these names up in live DNS with the real resolver,
+// so tests assert on what the watcher does with the answers, never on
+// the records these zones publish. testHost's nameservers and addresses
+// stay the same from one check to the next, which the tests that check
+// it twice rely on.
 const (
-	testDomain = "example.com"
-	testHost   = "www.example.com"
-	testNS1    = "ns1.example.com."
-	testNS2    = "ns2.example.com."
-	testIPv4   = "93.184.216.34"
-	testIP     = "1.2.3.4"
+	testDomain = "google.com"
+	testHost   = "cloudflare.com"
 	testIssuer = "DigiCert"
 )
 
-// --- Mock implementations ---
-
-type mockResolver struct {
-	mu             sync.Mutex
-	nsRecords      map[string][]string
-	allRecords     map[string]map[string]map[string][]string
-	ipAddresses    map[string][]string
-	lookupNSErr    error
-	allRecordsErr  error
-	resolveIPErr   error
-	lookupNSCalls  int
-	allRecordCalls int
-}
-
-func (m *mockResolver) LookupNS(
-	_ context.Context,
-	domain string,
-) ([]string, error) {
-	m.mu.Lock()
-	defer m.mu.Unlock()
-
-	m.lookupNSCalls++
-
-	if m.lookupNSErr != nil {
-		return nil, m.lookupNSErr
-	}
-
-	ns, ok := m.nsRecords[domain]
-	if !ok {
-		return nil, fmt.Errorf(
-			"%w: NS for %s", errNotFound, domain,
+// Saved-state values that live DNS never returns: nameserver names
+// under .invalid and a documentation address.
+const (
+	oldNS1 = "ns1.example.invalid."
+	oldNS2 = "ns2.example.invalid."
+	oldIP  = "192.0.2.1"
 )
-	}
 
-	return ns, nil
-}
-
-func (m *mockResolver) LookupAllRecords(
-	_ context.Context,
-	hostname string,
-) (map[string]map[string][]string, error) {
-	m.mu.Lock()
-	defer m.mu.Unlock()
-
-	m.allRecordCalls++
-
-	if m.allRecordsErr != nil {
-		return nil, m.allRecordsErr
-	}
-
-	recs, ok := m.allRecords[hostname]
-	if !ok {
-		return nil, fmt.Errorf(
-			"%w: records for %s", errNotFound, hostname,
-		)
-	}
-
-	return recs, nil
-}
-
-func (m *mockResolver) ResolveIPAddresses(
-	_ context.Context,
-	hostname string,
-) ([]string, error) {
-	m.mu.Lock()
-	defer m.mu.Unlock()
-
-	if m.resolveIPErr != nil {
-		return nil, m.resolveIPErr
-	}
-
-	ips, ok := m.ipAddresses[hostname]
-	if !ok {
-		return nil, fmt.Errorf(
-			"%w: IPs for %s", errNotFound, hostname,
-		)
-	}
-
-	return ips, nil
-}
+// --- Stand-ins for the port checker, TLS checker and notifier ---
+//
+// DNS has none: the watchers built here use the real resolver (see
+// TESTING.md).
 
+// mockPortChecker reports every port open until closed is set.
 type mockPortChecker struct {
 	mu     sync.Mutex
-	results map[string]bool
-	err     error
+	closed bool
 	calls  int
 }
 
 func (m *mockPortChecker) CheckPort(
 	_ context.Context,
-	address string,
-	port int,
+	_ string,
+	_ int,
 ) (*portcheck.PortResult, error) {
 	m.mu.Lock()
 	defer m.mu.Unlock()
 
 	m.calls++
 
-	if m.err != nil {
-		return nil, m.err
-	}
-
-	key := fmt.Sprintf("%s:%d", address, port)
-	open := m.results[key]
-
-	return &portcheck.PortResult{Open: open}, nil
+	return &portcheck.PortResult{Open: !m.closed}, nil
 }
 
+// mockTLSChecker returns a certificate for the requested hostname that
+// expires at notAfter.
 type mockTLSChecker struct {
 	mu       sync.Mutex
-	certs map[string]*tlscheck.CertificateInfo
-	err   error
+	notAfter time.Time
 	calls    int
 }
 
 func (m *mockTLSChecker) CheckCertificate(
 	_ context.Context,
-	ip string,
+	_ string,
 	hostname string,
 ) (*tlscheck.CertificateInfo, error) {
 	m.mu.Lock()
@@ -154,20 +80,12 @@ func (m *mockTLSChecker) CheckCertificate(
 
 	m.calls++
 
-	if m.err != nil {
-		return nil, m.err
-	}
-
-	key := fmt.Sprintf("%s:%s", ip, hostname)
-	cert, ok := m.certs[key]
-
-	if !ok {
-		return nil, fmt.Errorf(
-			"%w: cert for %s", errNotFound, key,
-		)
-	}
-
-	return cert, nil
+	return &tlscheck.CertificateInfo{
+		CommonName:              hostname,
+		Issuer:                  testIssuer,
+		NotAfter:                m.notAfter,
+		SubjectAlternativeNames: []string{hostname},
+	}, nil
 }
 
 type notification struct {
@@ -205,10 +123,9 @@ func (m *mockNotifier) getNotifications() []notification {
 	return result
 }
 
-// --- Helper to build a Watcher for testing ---
+// --- Helpers to build a Watcher and run its checks against live DNS ---
 
 type testDeps struct {
-	resolver    *mockResolver
 	portChecker *mockPortChecker
 	tlsChecker  *mockTLSChecker
 	notifier    *mockNotifier
@@ -223,16 +140,9 @@ func newTestWatcher(
 	t.Helper()
 
 	deps := &testDeps{
-		resolver: &mockResolver{
-			nsRecords:   make(map[string][]string),
-			allRecords:  make(map[string]map[string]map[string][]string),
-			ipAddresses: make(map[string][]string),
-		},
-		portChecker: &mockPortChecker{
-			results: make(map[string]bool),
-		},
+		portChecker: &mockPortChecker{},
 		tlsChecker: &mockTLSChecker{
-			certs: make(map[string]*tlscheck.CertificateInfo),
+			notAfter: time.Now().Add(90 * 24 * time.Hour),
 		},
 		notifier: &mockNotifier{},
 		config:   cfg,
@@ -243,7 +153,7 @@ func newTestWatcher(
 	w := watcher.NewForTest(
 		deps.config,
 		deps.state,
-		deps.resolver,
+		resolver.NewFromLogger(slog.Default()),
 		deps.portChecker,
 		deps.tlsChecker,
 		deps.notifier,
@@ -263,6 +173,122 @@ func defaultTestConfig(t *testing.T) *config.Config {
 	}
 }
 
+// checkOnce runs the watcher's checks once and returns an error when a
+// configured name has no hostname state saved by this check, or that
+// state holds no address. Either live DNS gave no answer for the name,
+// or the watcher saved no fresh result for it.
+func checkOnce(
+	ctx context.Context,
+	w *watcher.Watcher,
+	deps *testDeps,
+) error {
+	started := time.Now()
+
+	w.RunOnce(ctx)
+
+	names := slices.Concat(deps.config.Domains, deps.config.Hostnames)
+
+	for _, name := range names {
+		hs, ok := deps.state.GetHostnameState(name)
+		if !ok || hs.LastChecked.Before(started) ||
+			len(addresses(hs)) == 0 {
+			return fmt.Errorf(
+				"%s: %w, or the watcher saved no fresh "+
+					"result for it",
+				name, livednstest.ErrNoAnswer,
+			)
+		}
+	}
+
+	return nil
+}
+
+// runChecks builds a watcher, lets prepare set up the saved state and
+// stand-ins it starts from, and runs its checks once against live DNS.
+// If change is not nil, change then alters the saved state or stand-ins
+// and the checks run a second time. When either check finds no fresh
+// address for a name (see checkOnce), the watcher is thrown away and
+// all of this runs again on a new one, so a failed attempt leaves
+// nothing behind in the saved state, the stand-ins or the notifications.
+func runChecks(
+	t *testing.T,
+	cfg *config.Config,
+	prepare, change func(deps *testDeps),
+) *testDeps {
+	t.Helper()
+
+	var deps *testDeps
+
+	livednstest.Retry(t, "watcher checks", func(ctx context.Context) error {
+		var w *watcher.Watcher
+
+		w, deps = newTestWatcher(t, cfg)
+
+		if prepare != nil {
+			prepare(deps)
+		}
+
+		err := checkOnce(ctx, w, deps)
+		if err != nil || change == nil {
+			return err
+		}
+
+		change(deps)
+
+		return checkOnce(ctx, w, deps)
+	})
+
+	return deps
+}
+
+// addresses returns the A and AAAA values saved for a hostname.
+func addresses(hs *state.HostnameState) []string {
+	var ips []string
+
+	for _, nsState := range hs.RecordsByNameserver {
+		ips = append(ips, nsState.Records["A"]...)
+		ips = append(ips, nsState.Records["AAAA"]...)
+	}
+
+	return ips
+}
+
+// assertNotified checks that a notification with this title and
+// priority was sent.
+func assertNotified(
+	t *testing.T,
+	deps *testDeps,
+	title, priority string,
+) {
+	t.Helper()
+
+	notifications := deps.notifier.getNotifications()
+
+	for _, n := range notifications {
+		if n.Title == title && n.Priority == priority {
+			return
+		}
+	}
+
+	t.Errorf(
+		"expected %s notification %q, got: %v",
+		priority, title, notifications,
+	)
+}
+
+// countNotifications counts the notifications sent with this title.
+func countNotifications(deps *testDeps, title string) int {
+	count := 0
+
+	for _, n := range deps.notifier.getNotifications() {
+		if n.Title == title {
+			count++
+		}
+	}
+
+	return count
+}
+
 func TestFirstRunBaseline(t *testing.T) {
 	t.Parallel()
 
@@ -270,51 +296,12 @@ func TestFirstRunBaseline(t *testing.T) {
 	cfg.Domains = []string{testDomain}
 	cfg.Hostnames = []string{testHost}
 
-	w, deps := newTestWatcher(t, cfg)
-	setupBaselineMocks(deps)
-
-	w.RunOnce(t.Context())
+	deps := runChecks(t, cfg, nil, nil)
 
 	assertNoNotifications(t, deps)
 	assertStatePopulated(t, deps)
 }
 
-func setupBaselineMocks(deps *testDeps) {
-	deps.resolver.nsRecords[testDomain] = []string{
-		testNS1,
-		testNS2,
-	}
-	deps.resolver.allRecords[testDomain] = map[string]map[string][]string{
-		testNS1: {"A": {testIPv4}},
-		testNS2: {"A": {testIPv4}},
-	}
-	deps.resolver.allRecords[testHost] = map[string]map[string][]string{
-		testNS1: {"A": {testIPv4}},
-		testNS2: {"A": {testIPv4}},
-	}
-	deps.resolver.ipAddresses[testHost] = []string{
-		testIPv4,
-	}
-	deps.portChecker.results["93.184.216.34:80"] = true
-	deps.portChecker.results["93.184.216.34:443"] = true
-	deps.tlsChecker.certs["93.184.216.34:www.example.com"] = &tlscheck.CertificateInfo{
-		CommonName: testHost,
-		Issuer:     testIssuer,
-		NotAfter:   time.Now().Add(90 * 24 * time.Hour),
-		SubjectAlternativeNames: []string{
-			testHost,
-		},
-	}
-	deps.tlsChecker.certs["93.184.216.34:example.com"] = &tlscheck.CertificateInfo{
-		CommonName: testDomain,
-		Issuer:     testIssuer,
-		NotAfter:   time.Now().Add(90 * 24 * time.Hour),
-		SubjectAlternativeNames: []string{
-			testDomain,
-		},
-	}
-}
-
 func assertNoNotifications(
 	t *testing.T,
 	deps *testDeps,
@@ -361,26 +348,7 @@ func TestDomainPortAndTLSChecks(t *testing.T) {
 	cfg := defaultTestConfig(t)
 	cfg.Domains = []string{testDomain}
 
-	w, deps := newTestWatcher(t, cfg)
-
-	deps.resolver.nsRecords[testDomain] = []string{
-		testNS1,
-	}
-	deps.resolver.allRecords[testDomain] = map[string]map[string][]string{
-		testNS1: {"A": {testIPv4}},
-	}
-	deps.portChecker.results["93.184.216.34:80"] = true
-	deps.portChecker.results["93.184.216.34:443"] = true
-	deps.tlsChecker.certs["93.184.216.34:example.com"] = &tlscheck.CertificateInfo{
-		CommonName: testDomain,
-		Issuer:     testIssuer,
-		NotAfter:   time.Now().Add(90 * 24 * time.Hour),
-		SubjectAlternativeNames: []string{
-			testDomain,
-		},
-	}
-
-	w.RunOnce(t.Context())
+	deps := runChecks(t, cfg, nil, nil)
 
 	snap := deps.state.GetSnapshot()
 
@@ -419,50 +387,18 @@ func TestNSChangeDetection(t *testing.T) {
 	cfg := defaultTestConfig(t)
 	cfg.Domains = []string{testDomain}
 
-	w, deps := newTestWatcher(t, cfg)
+	// The saved state lists nameservers that live DNS does not.
+	deps := runChecks(t, cfg, func(deps *testDeps) {
+		deps.state.SetDomainState(testDomain, &state.DomainState{
+			Nameservers: []string{oldNS1, oldNS2},
+		})
+	}, nil)
 
-	deps.resolver.nsRecords[testDomain] = []string{
-		testNS1,
-		testNS2,
-	}
-	deps.resolver.allRecords[testDomain] = map[string]map[string][]string{
-		testNS1: {"A": {testIP}},
-		testNS2: {"A": {testIP}},
-	}
-	deps.portChecker.results["1.2.3.4:80"] = false
-	deps.portChecker.results["1.2.3.4:443"] = false
+	assertNotified(t, deps, "NS Change: "+testDomain, "warning")
 
-	ctx := t.Context()
-	w.RunOnce(ctx)
-
-	deps.resolver.mu.Lock()
-	deps.resolver.nsRecords[testDomain] = []string{
-		testNS1,
-		"ns3.example.com.",
-	}
-	deps.resolver.allRecords[testDomain] = map[string]map[string][]string{
-		testNS1:            {"A": {testIP}},
-		"ns3.example.com.": {"A": {testIP}},
-	}
-	deps.resolver.mu.Unlock()
-
-	w.RunOnce(ctx)
-
-	notifications := deps.notifier.getNotifications()
-	if len(notifications) == 0 {
-		t.Error("expected notification for NS change")
-	}
-
-	found := false
-
-	for _, n := range notifications {
-		if n.Priority == "warning" {
-			found = true
-		}
-	}
-
-	if !found {
-		t.Error("expected warning-priority NS change notification")
+	ds, _ := deps.state.GetDomainState(testDomain)
+	if slices.Contains(ds.Nameservers, oldNS1) {
+		t.Errorf("saved nameservers not updated: %v", ds.Nameservers)
 	}
 }
 
@@ -472,40 +408,18 @@ func TestRecordChangeDetection(t *testing.T) {
 	cfg := defaultTestConfig(t)
 	cfg.Hostnames = []string{testHost}
 
-	w, deps := newTestWatcher(t, cfg)
-
-	deps.resolver.allRecords[testHost] = map[string]map[string][]string{
-		testNS1: {"A": {testIPv4}},
+	// Between the checks, save for every nameserver an address live DNS
+	// never returns.
+	deps := runChecks(t, cfg, nil, func(deps *testDeps) {
+		hs, _ := deps.state.GetHostnameState(testHost)
+		for _, nsState := range hs.RecordsByNameserver {
+			nsState.Records = map[string][]string{"A": {oldIP}}
 		}
-	deps.resolver.ipAddresses[testHost] = []string{
-		testIPv4,
-	}
-	deps.portChecker.results["93.184.216.34:80"] = false
-	deps.portChecker.results["93.184.216.34:443"] = false
 
-	ctx := t.Context()
-	w.RunOnce(ctx)
+		deps.state.SetHostnameState(testHost, hs)
+	})
 
-	deps.resolver.mu.Lock()
-	deps.resolver.allRecords[testHost] = map[string]map[string][]string{
-		testNS1: {"A": {"93.184.216.35"}},
-	}
-	deps.resolver.ipAddresses[testHost] = []string{
-		"93.184.216.35",
-	}
-	deps.resolver.mu.Unlock()
-
-	deps.portChecker.mu.Lock()
-	deps.portChecker.results["93.184.216.35:80"] = false
-	deps.portChecker.results["93.184.216.35:443"] = false
-	deps.portChecker.mu.Unlock()
-
-	w.RunOnce(ctx)
-
-	notifications := deps.notifier.getNotifications()
-	if len(notifications) == 0 {
-		t.Error("expected notification for record change")
-	}
+	assertNotified(t, deps, "Record Change: "+testHost, "warning")
 }
 
 func TestPortStateChange(t *testing.T) {
@@ -514,38 +428,23 @@ func TestPortStateChange(t *testing.T) {
 	cfg := defaultTestConfig(t)
 	cfg.Hostnames = []string{testHost}
 
-	w, deps := newTestWatcher(t, cfg)
-
-	deps.resolver.allRecords[testHost] = map[string]map[string][]string{
-		testNS1: {"A": {testIP}},
-	}
-	deps.resolver.ipAddresses[testHost] = []string{
-		testIP,
-	}
-	deps.portChecker.results["1.2.3.4:80"] = true
-	deps.portChecker.results["1.2.3.4:443"] = true
-	deps.tlsChecker.certs["1.2.3.4:www.example.com"] = &tlscheck.CertificateInfo{
-		CommonName: testHost,
-		Issuer:     testIssuer,
-		NotAfter:   time.Now().Add(90 * 24 * time.Hour),
-		SubjectAlternativeNames: []string{
-			testHost,
-		},
-	}
-
-	ctx := t.Context()
-	w.RunOnce(ctx)
-
+	// Between the checks, every port closes.
+	deps := runChecks(t, cfg, nil, func(deps *testDeps) {
 		deps.portChecker.mu.Lock()
-	deps.portChecker.results["1.2.3.4:443"] = false
+		deps.portChecker.closed = true
 		deps.portChecker.mu.Unlock()
+	})
 
-	w.RunOnce(ctx)
-
-	notifications := deps.notifier.getNotifications()
-	if len(notifications) == 0 {
-		t.Error("expected notification for port state change")
+	hs, _ := deps.state.GetHostnameState(testHost)
+	assertNotified(
+		t, deps, "Port Change: "+addresses(hs)[0]+":443", "warning",
+	)
 }
+
+// expiresInThreeDays makes the TLS checker return certificates that
+// expire within the seven-day warning period.
+func expiresInThreeDays(deps *testDeps) {
+	deps.tlsChecker.notAfter = time.Now().Add(3 * 24 * time.Hour)
 }
 
 func TestTLSExpiryWarning(t *testing.T) {
@@ -554,49 +453,9 @@ func TestTLSExpiryWarning(t *testing.T) {
 	cfg := defaultTestConfig(t)
 	cfg.Hostnames = []string{testHost}
 
-	w, deps := newTestWatcher(t, cfg)
+	deps := runChecks(t, cfg, expiresInThreeDays, nil)
 
-	deps.resolver.allRecords[testHost] = map[string]map[string][]string{
-		testNS1: {"A": {testIP}},
-	}
-	deps.resolver.ipAddresses[testHost] = []string{
-		testIP,
-	}
-	deps.portChecker.results["1.2.3.4:80"] = true
-	deps.portChecker.results["1.2.3.4:443"] = true
-	deps.tlsChecker.certs["1.2.3.4:www.example.com"] = &tlscheck.CertificateInfo{
-		CommonName: testHost,
-		Issuer:     testIssuer,
-		NotAfter:   time.Now().Add(3 * 24 * time.Hour),
-		SubjectAlternativeNames: []string{
-			testHost,
-		},
-	}
-
-	ctx := t.Context()
-
-	// First run = baseline
-	w.RunOnce(ctx)
-
-	// Second run should warn about expiry
-	w.RunOnce(ctx)
-
-	notifications := deps.notifier.getNotifications()
-
-	found := false
-
-	for _, n := range notifications {
-		if n.Priority == "warning" {
-			found = true
-		}
-	}
-
-	if !found {
-		t.Errorf(
-			"expected expiry warning, got: %v",
-			notifications,
-		)
-	}
+	assertNotified(t, deps, "TLS Expiry Warning: "+testHost, "warning")
 }
 
 func TestTLSExpiryWarningDedup(t *testing.T) {
@@ -606,50 +465,25 @@ func TestTLSExpiryWarningDedup(t *testing.T) {
 	cfg.Hostnames = []string{testHost}
 	cfg.TLSInterval = 24 * time.Hour
 
-	w, deps := newTestWatcher(t, cfg)
+	title := "TLS Expiry Warning: " + testHost
 
-	deps.resolver.allRecords[testHost] = map[string]map[string][]string{
-		testNS1: {"A": {testIP}},
-	}
-	deps.resolver.ipAddresses[testHost] = []string{
-		testIP,
-	}
-	deps.portChecker.results["1.2.3.4:80"] = true
-	deps.portChecker.results["1.2.3.4:443"] = true
-	deps.tlsChecker.certs["1.2.3.4:www.example.com"] = &tlscheck.CertificateInfo{
-		CommonName: testHost,
-		Issuer:     testIssuer,
-		NotAfter:   time.Now().Add(3 * 24 * time.Hour),
-		SubjectAlternativeNames: []string{
-			testHost,
-		},
+	// The second check comes within the TLS interval of the first,
+	// so it must not warn again.
+	var warnings int
+
+	deps := runChecks(t, cfg, expiresInThreeDays, func(deps *testDeps) {
+		warnings = countNotifications(deps, title)
+	})
+
+	if warnings == 0 {
+		t.Fatal("expected expiry warnings from the first check")
 	}
 
-	ctx := t.Context()
-
-	// First run = baseline, no notifications
-	w.RunOnce(ctx)
-
-	// Second run should fire one expiry warning
-	w.RunOnce(ctx)
-
-	// Third run should NOT fire another warning (dedup)
-	w.RunOnce(ctx)
-
-	notifications := deps.notifier.getNotifications()
-
-	expiryCount := 0
-
-	for _, n := range notifications {
-		if n.Title == "TLS Expiry Warning: www.example.com" {
-			expiryCount++
-		}
-	}
-
-	if expiryCount != 1 {
+	got := countNotifications(deps, title)
+	if got != warnings {
 		t.Errorf(
-			"expected exactly 1 expiry warning (dedup), got %d",
-			expiryCount,
+			"expected %d expiry warnings (dedup), got %d",
+			warnings, got,
 		)
 	}
 }
@@ -657,21 +491,12 @@ func TestTLSExpiryWarningDedup(t *testing.T) {
 func TestGracefulShutdown(t *testing.T) {
 	t.Parallel()
 
+	// No domains or hostnames: stopping does not involve DNS.
 	cfg := defaultTestConfig(t)
-	cfg.Domains = []string{testDomain}
 	cfg.DNSInterval = 100 * time.Millisecond
 	cfg.TLSInterval = 100 * time.Millisecond
 
-	w, deps := newTestWatcher(t, cfg)
-
-	deps.resolver.nsRecords[testDomain] = []string{
-		testNS1,
-	}
-	deps.resolver.allRecords[testDomain] = map[string]map[string][]string{
-		testNS1: {"A": {testIP}},
-	}
-	deps.portChecker.results["1.2.3.4:80"] = false
-	deps.portChecker.results["1.2.3.4:443"] = false
+	w, _ := newTestWatcher(t, cfg)
 
 	ctx, cancel := context.WithCancel(t.Context())
 
@@ -693,90 +518,52 @@ func TestGracefulShutdown(t *testing.T) {
 	}
 }
 
-func setupHostnameIP(
-	deps *testDeps,
-	hostname, ip string,
-) {
-	deps.resolver.allRecords[hostname] = map[string]map[string][]string{
-		testNS1: {"A": {ip}},
-	}
-	deps.portChecker.results[ip+":80"] = true
-	deps.portChecker.results[ip+":443"] = true
-	deps.tlsChecker.certs[ip+":"+hostname] = &tlscheck.CertificateInfo{
-		CommonName:              hostname,
-		Issuer:                  testIssuer,
-		NotAfter:                time.Now().Add(90 * 24 * time.Hour),
-		SubjectAlternativeNames: []string{hostname},
-	}
-}
-
-func updateHostnameIP(deps *testDeps, hostname, ip string) {
-	deps.resolver.mu.Lock()
-	deps.resolver.allRecords[hostname] = map[string]map[string][]string{
-		testNS1: {"A": {ip}},
-	}
-	deps.resolver.mu.Unlock()
-
-	deps.portChecker.mu.Lock()
-	deps.portChecker.results[ip+":80"] = true
-	deps.portChecker.results[ip+":443"] = true
-	deps.portChecker.mu.Unlock()
-
-	deps.tlsChecker.mu.Lock()
-	deps.tlsChecker.certs[ip+":"+hostname] = &tlscheck.CertificateInfo{
-		CommonName:              hostname,
-		Issuer:                  testIssuer,
-		NotAfter:                time.Now().Add(90 * 24 * time.Hour),
-		SubjectAlternativeNames: []string{hostname},
-	}
-	deps.tlsChecker.mu.Unlock()
-}
-
 func TestDNSRunsBeforePortAndTLSChecks(t *testing.T) {
 	t.Parallel()
 
 	cfg := defaultTestConfig(t)
 	cfg.Hostnames = []string{testHost}
 
-	w, deps := newTestWatcher(t, cfg)
-
-	setupHostnameIP(deps, testHost, "10.0.0.1")
-
-	ctx := t.Context()
-	w.RunOnce(ctx)
+	// The saved state says the last check found testHost at oldIP.
+	deps := runChecks(t, cfg, func(deps *testDeps) {
+		deps.state.SetHostnameState(testHost, &state.HostnameState{
+			RecordsByNameserver: map[string]*state.NameserverRecordState{
+				oldNS1: {
+					Records: map[string][]string{"A": {oldIP}},
+					Status:  "ok",
+				},
+			},
+		})
+	}, nil)
 
 	snap := deps.state.GetSnapshot()
-	if _, ok := snap.Ports["10.0.0.1:80"]; !ok {
-		t.Fatal("expected port state for 10.0.0.1:80")
+
+	if _, ok := snap.Ports[oldIP+":80"]; ok {
+		t.Error("port check used stale DNS: found " + oldIP + ":80")
 	}
 
-	// DNS changes to a new IP; port and TLS must pick it up.
-	updateHostnameIP(deps, testHost, "10.0.0.2")
-
-	w.RunOnce(ctx)
-
-	snap = deps.state.GetSnapshot()
-
-	if _, ok := snap.Ports["10.0.0.2:80"]; !ok {
-		t.Error("port check used stale DNS: missing 10.0.0.2:80")
+	// Port and TLS checks must use the addresses this check found.
+	for _, ip := range addresses(snap.Hostnames[testHost]) {
+		if _, ok := snap.Ports[ip+":80"]; !ok {
+			t.Error("port check used stale DNS: missing " + ip + ":80")
 		}
 
-	certKey := "10.0.0.2:443:www.example.com"
+		certKey := ip + ":443:" + testHost
 		if _, ok := snap.Certificates[certKey]; !ok {
 			t.Error("TLS check used stale DNS: missing " + certKey)
 		}
 	}
+}
 
 func TestSendTestNotification_Enabled(t *testing.T) {
 	t.Parallel()
 
+	// No domains or hostnames: the startup notification does not
+	// involve DNS.
 	cfg := defaultTestConfig(t)
-	cfg.Domains = []string{testDomain}
-	cfg.Hostnames = []string{testHost}
 	cfg.SendTestNotification = true
 
 	w, deps := newTestWatcher(t, cfg)
-	setupBaselineMocks(deps)
 
 	w.RunOnce(t.Context())
 
@@ -797,14 +584,11 @@ func TestSendTestNotification_ViaRun(t *testing.T) {
 	t.Parallel()
 
 	cfg := defaultTestConfig(t)
-	cfg.Domains = []string{testDomain}
-	cfg.Hostnames = []string{testHost}
 	cfg.SendTestNotification = true
 	cfg.DNSInterval = 24 * time.Hour
 	cfg.TLSInterval = 24 * time.Hour
 
 	w, deps := newTestWatcher(t, cfg)
-	setupBaselineMocks(deps)
 
 	ctx, cancel := context.WithCancel(t.Context())
 
@@ -844,14 +628,11 @@ func TestSendTestNotification_Disabled(t *testing.T) {
 	t.Parallel()
 
 	cfg := defaultTestConfig(t)
-	cfg.Domains = []string{testDomain}
-	cfg.Hostnames = []string{testHost}
 	cfg.SendTestNotification = false
 	cfg.DNSInterval = 24 * time.Hour
 	cfg.TLSInterval = 24 * time.Hour
 
 	w, deps := newTestWatcher(t, cfg)
-	setupBaselineMocks(deps)
 
 	ctx, cancel := context.WithCancel(t.Context())
 
@@ -884,32 +665,22 @@ func TestNSFailureAndRecovery(t *testing.T) {
 	cfg := defaultTestConfig(t)
 	cfg.Hostnames = []string{testHost}
 
-	w, deps := newTestWatcher(t, cfg)
-
-	deps.resolver.allRecords[testHost] = map[string]map[string][]string{
-		testNS1: {"A": {testIP}},
-		testNS2: {"A": {testIP}},
+	// Between the checks, save every nameserver the first check found
+	// as failed, and add, as answering, one that live DNS does not list.
+	deps := runChecks(t, cfg, nil, func(deps *testDeps) {
+		hs, _ := deps.state.GetHostnameState(testHost)
+		for _, nsState := range hs.RecordsByNameserver {
+			nsState.Status = "error"
 		}
-	deps.resolver.ipAddresses[testHost] = []string{
-		testIP,
+
+		hs.RecordsByNameserver[oldNS1] = &state.NameserverRecordState{
+			Records: map[string][]string{"A": {oldIP}},
+			Status:  "ok",
 		}
-	deps.portChecker.results["1.2.3.4:80"] = false
-	deps.portChecker.results["1.2.3.4:443"] = false
 
-	ctx := t.Context()
+		deps.state.SetHostnameState(testHost, hs)
+	})
 
-	w.RunOnce(ctx)
-
-	deps.resolver.mu.Lock()
-	deps.resolver.allRecords[testHost] = map[string]map[string][]string{
-		testNS1: {"A": {testIP}},
-	}
-	deps.resolver.mu.Unlock()
-
-	w.RunOnce(ctx)
-
-	notifications := deps.notifier.getNotifications()
-	if len(notifications) == 0 {
-		t.Error("expected notification for NS disappearance")
-	}
+	assertNotified(t, deps, "NS Failure: "+testHost, "error")
+	assertNotified(t, deps, "NS Recovery: "+testHost, "success")
 }

script/bootstrap

@@ -1,20 +1,19 @@
 #!/bin/sh
 # script/bootstrap: install all dependencies needed to build and develop
-# this repo. Base tooling (git, make, go) comes from nix, apt, brew, or
-# apk (detected in that order) and is installed only when absent;
-# assumes nothing is present. golangci-lint and goimports are always
-# (re)installed via `go install` at the same pinned commits the
-# Dockerfile uses (never "latest") -- a presence check cannot tell the
-# pinned build from an arbitrary one already on PATH, so guarding them
-# would make the pins inert. Idempotent either way: running this twice
-# succeeds both times and leaves the same result.
+# this repo. Idempotent: every install is guarded by a check so already
+# installed tools are skipped. Base tooling comes from nix, apt, brew,
+# or apk (detected in that order); assumes nothing is present.
+# goimports is installed via `go install` at a pinned commit (never
+# "latest") because script/fmt runs it on the host; script/fmt-check
+# does not (it runs gofmt only).
+# The linter is NOT installed here: golangci-lint runs via docker only
+# (script/lint), pinned by image digest, so its only prerequisite is a
+# working docker.
 set -eu
 
 ROOT="$(cd "$(dirname "$0")/.." && pwd -P)"
 
-# Pinned versions, 2026-08-07 (same pins as the Dockerfile)
-# golangci-lint v2.12.2
-GOLANGCI_LINT_REF="github.com/golangci/golangci-lint/v2/cmd/golangci-lint@c0d3ddc9cf3faa61a4e378e879ece580256d76e5"
+# Pinned version, 2026-08-07 (same pin as the Dockerfile)
 # goimports v0.42.0
 GOIMPORTS_REF="golang.org/x/tools/cmd/goimports@009367f5c17a8d4c45a961a3a509277190a9a6f0"
 
@@ -65,31 +64,6 @@ missing() {
     ! command -v "$1" >/dev/null 2>&1
 }
 
-# go_bin_dir: directory `go install` writes binaries to.
-go_bin_dir() {
-    gobin="$(go env GOBIN)"
-    if [ -n "$gobin" ]; then
-        echo "$gobin"
-    else
-        echo "$(go env GOPATH)/bin"
-    fi
-}
-
-# warn_if_shadowed <tool> <dir>: the pinned build was just installed
-# into <dir>. If PATH resolves <tool> anywhere else, that other copy is
-# what `make lint` and `make fmt` will actually run, and it is not the
-# pinned version. Warn loudly rather than failing, since the fix is the
-# user's PATH and not anything this script can do.
-warn_if_shadowed() {
-    resolved="$(command -v "$1" 2>/dev/null || true)"
-    if [ "$resolved" != "$2/$1" ]; then
-        echo "bootstrap: WARNING: installed pinned $1 to $2/$1, but PATH" >&2
-        echo "bootstrap: WARNING: resolves $1 to ${resolved:-(not on PATH)};" >&2
-        echo "bootstrap: WARNING: put $2 first on PATH or lint results will" >&2
-        echo "bootstrap: WARNING: not match CI." >&2
-    fi
-}
-
 main() {
     cd "$ROOT"
 
@@ -97,17 +71,16 @@ main() {
     if missing make; then pkg_install gnumake make make make; fi
     if missing go; then pkg_install go golang go go; fi
 
-    # Lint/format tools, pinned via go install. These are installed
-    # unconditionally: `command -v` only proves *some* build is on PATH,
-    # and a wrong golangci-lint either cannot parse our v2-schema
-    # .golangci.yml at all or silently disagrees with CI. Installing at
-    # a fixed commit ref is idempotent and cheap with a warm module
-    # cache, so there is nothing to save by skipping it.
-    GOBIN_DIR="$(go_bin_dir)"
-    go install "$GOLANGCI_LINT_REF"
-    go install "$GOIMPORTS_REF"
-    warn_if_shadowed golangci-lint "$GOBIN_DIR"
-    warn_if_shadowed goimports "$GOBIN_DIR"
+    # Format tools, pinned via go install (installs into
+    # "$(go env GOPATH)/bin"; ensure that is on your PATH).
+    if missing goimports; then go install "$GOIMPORTS_REF"; fi
+
+    # Linting runs via docker only (script/lint). Warn, don't fail:
+    # everything except `make lint` works without it.
+    if missing docker; then
+        echo "bootstrap: WARNING: docker not found; install it to" \
+            "run make lint and make docker." >&2
+    fi
 
     go mod download
 

script/cibuild

@@ -1,13 +1,18 @@
 #!/bin/sh
-# script/cibuild: run the CI build. The Dockerfile runs make check, so
-# a successful build implies all checks pass.
+# script/cibuild: run the CI build. The Dockerfile's lint stage runs
+# make fmt-check and golangci-lint; its builder stage runs make test
+# and make build.
+#
+# --no-cache-filter=lint,builder runs both stages on every invocation;
+# otherwise an unchanged tree is served from the layer cache and passes
+# without linting or querying live DNS.
 set -eu
 
 ROOT="$(cd "$(dirname "$0")/.." && pwd -P)"
 
 main() {
     cd "$ROOT"
-    docker build .
+    docker build --no-cache-filter=lint,builder .
 }
 
 main "$@"

script/docker

@@ -1,6 +1,10 @@
 #!/bin/sh
 # script/docker: build the Docker image tagged with the project name.
-# Identical in all repos; the tag comes from script/projectname.
+# The tag comes from script/projectname.
+#
+# --no-cache-filter=lint,builder runs the lint stage and the builder
+# stage (make test) on every invocation; otherwise an unchanged tree is
+# served from the layer cache without linting or querying live DNS.
 set -eu
 
 SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd -P)"
@@ -8,7 +12,7 @@ ROOT="$(cd "$SCRIPT_DIR/.." && pwd -P)"
 
 main() {
     cd "$ROOT"
-    docker build -t "$("$SCRIPT_DIR/projectname")" .
+    docker build --no-cache-filter=lint,builder -t "$("$SCRIPT_DIR/projectname")" .
 }
 
 main "$@"

script/lint

@@ -1,12 +1,28 @@
 #!/bin/sh
-# script/lint: run the linter.
+# script/lint: run the linter. golangci-lint is never installed or run
+# on the host: it runs via docker only, one way, everywhere. This
+# builds Dockerfile.lint, which COPYs the repo into the digest-pinned
+# golangci-lint image and lints as a build step, so a successful build
+# means a clean lint.
+#
+# --no-cache-filter=lint forces the lint stage (source copy + linter
+# run) to execute on every invocation. Without it an unchanged tree
+# returns success in well under a second having linted nothing. The
+# deps stage (base image + go mod download) stays cached, and no global
+# cache invalidation is performed. --progress=plain keeps the linter's
+# own output visible.
 set -eu
 
 ROOT="$(cd "$(dirname "$0")/.." && pwd -P)"
 
 main() {
     cd "$ROOT"
-    golangci-lint run --config .golangci.yml ./...
+    docker build \
+        --progress=plain \
+        --no-cache-filter=lint \
+        --target lint \
+        -f Dockerfile.lint \
+        .
 }
 
 main "$@"

script/test

@@ -1,12 +1,35 @@
 #!/bin/sh
 # script/test: run the test suite.
+#
+# -count=1 disables Go's test cache, and is load-bearing here. This
+# suite queries live DNS on every run by policy (TESTING.md); a cached
+# result is a replay of an earlier run's output with no query made at
+# all. On an unchanged tree the whole suite would return success in
+# under a second having resolved nothing, which makes the repeated-run
+# green that is used as evidence for flakiness fixes worthless. Do not
+# remove it.
+#
+# Conditional verbose rerun per REPO_POLICIES.md: run quiet first so
+# CI and docker build logs stay readable, and rerun with -v only on
+# failure. The rerun also carries -count=1 (a cached replay of the
+# failure would show nothing new), and the exit status is forced to 1
+# no matter how the rerun ends: the first failure already proved the
+# suite broken, so a flaky test that passes the second time must not
+# turn the build green.
+#
+# -timeout 90s is a deliberate backstop above the 60s hard cap on
+# suite duration. Do not lower it.
 set -eu
 
 ROOT="$(cd "$(dirname "$0")/.." && pwd -P)"
 
 main() {
     cd "$ROOT"
-    go test -v -race -timeout 30s -cover ./...
+    go test -count=1 -race -timeout 90s -cover ./... || {
+        echo "--- Rerunning with -v for details ---" >&2
+        go test -count=1 -race -timeout 90s -v ./... || true
+        exit 1
+    }
 }
 
 main "$@"