Report file writes run one at a time (closes #150)

Two writes that failed at the same time put their reports back in the
wrong order when the older one failed first. A mutex is now held
across each write, from taking the waiting reports out of the buffer
to putting them back if the write fails, so the next write takes them
in order. The write Append starts at the 10 MiB threshold waits for
one under way, and Append starts no other while it waits. Files
written since the start are now deleted to make room in the order they
were written. A test that needed two writes at once is removed; new
tests cover the one waiting write and the order files are deleted in.

Model: opus-5-5

Dockerfile

@@ -10,11 +10,22 @@
 # phase covers the frontend as well, through a copy from a node stage.
 # Inside them each tool is invoked directly, never through make or
 # script/, whose lint and test are themselves docker builds.
+#
+# In every stage ARG CACHEBUST comes right after the steps that install
+# system packages and download Go's modules, or right after FROM in a
+# stage with neither, and every step that lints, tests or compiles
+# comes after it, as does yarn install in the node stages. script/
+# passes a new value on every build, which reruns every RUN step after
+# it. Docker keeps the layers above it until the base image, their own
+# lines or go.mod and go.sum change. The steps that compile mount Go's
+# build cache, and golangci-lint run also mounts golangci-lint's; both
+# survive from build to build.
 
 # Frontend lint stage: eslint with the rules in eslint.config.js. The
 # lint phase below runs it.
 # node:22-alpine as of 2026-02-22
 FROM node@sha256:e4bf2a82ad0a4037d28035ae71529873c069b13eb0455466ae0bc13363826e34 AS frontend-lint
+ARG CACHEBUST
 WORKDIR /app
 COPY package.json yarn.lock ./
 RUN yarn install --frozen-lockfile
@@ -29,8 +40,11 @@ FROM golangci/golangci-lint@sha256:ad862ba6b3798cbe0fd9fd7408d498fd74fbd2623a924
 WORKDIR /src
 COPY go.mod go.sum ./
 RUN go mod download
+ARG CACHEBUST
 COPY . .
-RUN golangci-lint run --config .golangci.yml ./...
+RUN --mount=type=cache,target=/root/.cache/go-build \
+    --mount=type=cache,target=/root/.cache/golangci-lint \
+    golangci-lint run --config .golangci.yml ./...
 # Nothing is wanted from frontend-lint; the copy is what makes this
 # phase run it.
 COPY --from=frontend-lint /app/yarn.lock /dev/null
@@ -43,6 +57,7 @@ COPY --from=frontend-lint /app/yarn.lock /dev/null
 # node would take a reporter option for one more file.
 # node:22-alpine as of 2026-02-22
 FROM node@sha256:e4bf2a82ad0a4037d28035ae71529873c069b13eb0455466ae0bc13363826e34 AS frontend-test
+ARG CACHEBUST
 WORKDIR /app
 COPY package.json yarn.lock ./
 RUN yarn install --frozen-lockfile
@@ -58,19 +73,19 @@ RUN NODE_OPTIONS=--test-reporter=dot timeout 90 yarn --silent run test || \
 # ships and the alpine one does not. -timeout 90s is a backstop above
 # the 60-second cap on the suite. The rerun with -v only shows details:
 # the step fails however it ends, because the first run already failed.
-# Go's module and build caches are in memory (tmpfs) for the go test
+# -count=1 because the mounted cache also holds test results, which go
-# step alone, so the image make test tags does not carry them and the
+# test would otherwise replay.
-# build spends no time writing them into it. They start empty on every
-# build, so no stored test result can stand in for a run.
 # golang:1.25.7-trixie, 2026-10-07
 FROM golang@sha256:2b174ffcf56c7ad0c47d30d2630693265639ddf2a5141149c2da34db921791b4 AS test
 WORKDIR /src
+COPY go.mod go.sum ./
+RUN go mod download
+ARG CACHEBUST
 COPY . .
-RUN --mount=type=tmpfs,target=/go/pkg/mod \
+RUN --mount=type=cache,target=/root/.cache/go-build \
-    --mount=type=tmpfs,target=/root/.cache/go-build \
+    go test -count=1 -timeout 90s -race -cover ./... || \
-    go test -timeout 90s -race -cover ./... || \
     { echo "--- Rerunning with -v for details ---"; \
-      go test -timeout 90s -race -v ./...; exit 1; }
+      go test -count=1 -timeout 90s -race -v ./...; exit 1; }
 # Nothing is wanted from the frontend-test stage; the copy is what makes
 # this phase run its tests.
 COPY --from=frontend-test /app/yarn.lock /dev/null
@@ -83,23 +98,28 @@ COPY --from=frontend-test /app/yarn.lock /dev/null
 FROM frontend-test AS frontend-build
 COPY --from=lint /src/go.sum /dev/null
 COPY --from=test /src/go.sum /dev/null
-# vite.config.js reads the commit for the page's footer with git.
+# vite.config.js reads the commit for the page's footer with git. git is
+# installed again on every image build, although it comes above
+# ARG CACHEBUST, because this stage starts from frontend-test, whose
+# steps run again on every build.
 RUN apk add --no-cache git
+ARG CACHEBUST
 RUN yarn build
 
 # Backend build stage. Nothing is wanted from the two phases; the copies
-# are what make BuildKit build them first, so this stage cannot run
+# are what make BuildKit build them first, so nothing after them can
-# unless lint and test passed.
+# run unless lint and test passed.
 # golang:1.25-alpine (2026-02-27)
 FROM golang@sha256:f6751d823c26342f9506c03797d2527668d095b0a15f1862cddb4d927a7a4ced AS builder
-COPY --from=lint /src/go.sum /dev/null
-COPY --from=test /src/go.sum /dev/null
 RUN apk add --no-cache git
-# A tar-stream context keeps the sender's file owners, which git refuses.
-RUN git config --system --add safe.directory /src
 WORKDIR /src
 COPY go.mod go.sum ./
 RUN go mod download
+ARG CACHEBUST
+COPY --from=lint /src/go.sum /dev/null
+COPY --from=test /src/go.sum /dev/null
+# A tar-stream context keeps the sender's file owners, which git refuses.
+RUN git config --system --add safe.directory /src
 COPY . .
 
 # The same go build as in script/build, which builds the binary on the
@@ -112,7 +132,8 @@ COPY . .
 # still empty, dev or unknown fails the build: git is missing or could
 # not read the checkout.
 ARG VERSION
-RUN version="${VERSION:-$(git describe --tags --always)}"; \
+RUN --mount=type=cache,target=/root/.cache/go-build \
+    version="${VERSION:-$(git describe --tags --always)}"; \
     if [ -e .git ]; then \
         case "$version" in ""|dev|unknown) \
             echo "version is '$version' although .git is present" >&2; \
@@ -127,6 +148,7 @@ RUN version="${VERSION:-$(git describe --tags --always)}"; \
 # and the stages it copies from, the two phases included.
 # nginx:stable-alpine as of 2026-02-22
 FROM nginx@sha256:15e96e59aa3b0aada3a121296e3bce117721f42d88f5f64217ef4b18f458c6ab
+ARG CACHEBUST
 
 # netwatch-server runs as this user, which owns the report directory.
 # nginx keeps the image's own arrangement: its main process runs as

README.md

@@ -58,11 +58,11 @@ either one is broken. We provide:
   defaults (see [Settings](#settings))
 - `script/clean` — remove what `script/build` writes
 - `script/projectname` — print the project name (used for the Docker image tag)
-- `script/test` — build the `test` phase of `Dockerfile` without the cache: the
+- `script/test` — build the `test` phase of `Dockerfile`, writing no image: the
   frontend's unit tests in its `frontend-test` stage, and the backend's Go tests
   with the race detector and coverage. The frontend's production build is left
   to the image build, in the `frontend-build` stage
-- `script/lint` — build the `lint` phase of `Dockerfile` without the cache:
+- `script/lint` — build the `lint` phase of `Dockerfile`, writing no image:
   eslint in its `frontend-lint` stage, and golangci-lint over the Go packages
 - `script/fmt` — format all files (writes): prettier over the JavaScript, CSS,
   HTML and Markdown, then gofmt over `cmd/` and `internal/`. It runs on the
@@ -86,15 +86,29 @@ either one is broken. We provide:
 - `script/frontend-fmt-check` — check prettier formatting (read-only)
 - `script/frontend-viewport-test` — responsive-layout verification of the built
   frontend in a containerised headless Chrome (see
-  [test/viewport/README.md](test/viewport/README.md)). Not part of
+  [test/viewport/README.md](test/viewport/README.md)). Before the browser
+  starts, it checks that the page, and a path that falls back to it, are served
+  with `Cache-Control: no-cache` and `Content-Security-Policy`. Not part of
   `script/check`: it takes minutes.
-- `script/docker` — build the image from `Dockerfile` without the build cache,
+- `script/docker` — build the image from `Dockerfile`, tagged `netwatch` via
-  tagged `netwatch` via `script/projectname`
+  `script/projectname`
 - `script/cibuild` — CI entrypoint: runs `script/bootstrap` and `script/check`,
-  then builds the image as `script/docker` does, without the build cache
+  then builds the image as `script/docker` does
 - `script/precommit` — run by the git pre-commit hook; runs `script/check`
 - `script/install-precommit` — install the git pre-commit hook
 
+`script/test`, `script/lint`, `script/docker` and `script/cibuild` each pass a
+new `CACHEBUST` value, so every step in `Dockerfile` that lints, tests or
+compiles runs again even when nothing has changed. Docker keeps the system
+packages and the Go modules from earlier builds until the base image, the step
+that installs them, or `go.mod` and `go.sum` change. The `frontend-build` stage
+is the exception: it installs git again on every image build, because it starts
+from the `frontend-test` stage, whose steps run again on every build. Go's build
+cache and golangci-lint's are kept as well, and each reuses a result only when
+everything it was made from is unchanged. Both `go test` runs carry `-count=1`,
+so no test result is reused. The yarn packages are installed again on every
+build.
+
 ## Responsive layout
 
 The narrow-viewport layout lives in the `max-width: 768px` media block in
@@ -153,8 +167,26 @@ report a delta, so only new samples are sent; the mark advances only on a
 delivered report, and while paused nothing is sent. Changing the interval clears
 each host's history on the page; the samples in it not yet reported still go in
 the next report. Delivery failure is quiet — one debug-log line per outage,
-retried at the next interval, never blocking probing. The report-building step
+never blocking probing. The report-building step is a pure function of host
-is a pure function of host state.
+state.
+
+A sample in the history that no report has delivered is sent only as long as the
+history holds it, so after a refused or failed report the next report carries
+the samples the history still has. Samples an interval change cleared from the
+history are kept apart, the last 100 per host, until they are delivered. The
+history keeps each host's last 100 rounds: 100 seconds at the 1-second interval,
+200 seconds at 2 seconds, 5 minutes at the default of 3 seconds and 100 minutes
+at 60 seconds. At 1 second, one refused or failed report loses the oldest 20
+seconds of each host's samples, as the next delivered report comes 2 minutes
+after the last. At 2 seconds and longer one loses nothing; samples are lost from
+the third refused report in a row at 2 seconds, from the fifth at 3 seconds, and
+later still at the longer intervals.
+
+No report is sent while the page is paused, but each paused round still adds an
+entry to the history, so each report due during a pause counts as a refused one.
+A pause longer than the history spans loses the samples in the history recorded
+between the last delivered report and the pause: up to a minute's worth for each
+host if the report before the pause was delivered.
 
 ### Backend
 
@@ -199,7 +231,7 @@ image, `make run` builds it and runs it on port 8080.
 | `PORT`                 | `8080`               | HTTP listen port                                                                                         |
 | `DATA_DIR`             | `./data/reports`     | Directory for compressed reports                                                                         |
 | `DATA_DIR_MAX_BYTES`   | `1073741824` (1 GiB) | Most bytes of report files kept in `DATA_DIR`, oldest deleted first; see [Report limits](#report-limits) |
-| `DEBUG`                | `false`              | Enable debug logging                                                                                     |
+| `DEBUG`                | empty                | Debug logging; any non-empty value, `false` and `0` included, turns it on                                |
 | `TRUSTED_PROXIES`      | RFC 1918 ranges      | Comma-separated CIDRs whose `X-Forwarded-For` / `X-Real-IP` headers are trusted for client IP resolution |
 | `REPORTS_PER_MINUTE`   | `60`                 | Reports each client address may send a minute; see [Report limits](#report-limits)                       |
 | `CORS_ALLOWED_ORIGINS` | empty                | Comma-separated origins whose pages may call the API; see [CORS](#cors)                                  |
@@ -212,13 +244,20 @@ replaces the default; set and empty, the server trusts no proxy. Loopback is not
 in the default, so a reverse proxy on the same host has to be named, as in
 `TRUSTED_PROXIES=127.0.0.1/32`. A request whose direct peer is outside this set
 has its forwarded headers ignored, and the direct peer is logged and
-rate-limited instead. The container image does not use this default; see
+rate-limited instead. From a direct peer inside the set, the client address is
-[Deployment](#deployment).
+the right-most `X-Forwarded-For` entry that is not in the set: each proxy
+appends the address it took the request from, so the entries further left were
+written by the client and are not used. An entry written as an address with a
+port, such as `198.51.100.7:5000` or `[2001:db8::1]:443`, is read as that
+address. If the search from the right finds only addresses in the set, or meets
+an entry that is not an address at all, such as `unknown`, the direct peer is
+used. `X-Real-IP` is read only when `X-Forwarded-For` is missing or empty. The
+container image does not use this default; see [Deployment](#deployment).
 
-A variable set to a value the server cannot use, such as `PORT=abc`,
+A variable set to a value the server cannot use, such as `PORT=abc` or a
-`DEBUG=maybe` or a `BIND_ADDRESS` that is not an IP address, stops it from
+`BIND_ADDRESS` that is not an IP address, stops it from starting, with an error
-starting, with an error naming the variable. An empty variable counts as unset,
+naming the variable. An empty variable counts as unset, except
-except `TRUSTED_PROXIES`.
+`TRUSTED_PROXIES`. No value of `DEBUG` stops the start.
 
 #### Report storage
 
@@ -232,18 +271,22 @@ toward `DATA_DIR_MAX_BYTES` from the next start, only if removing it fails too.
 Files are created with `O_EXCL` to prevent overwrites.
 
 The reports of a failed write stay in memory, and the next flush, within a
-minute, tries again to write them. From the failed write until a write succeeds,
+minute, tries again to write them. The server writes one file at a time, so a
-`POST /api/v1/reports` refuses every report with 500 and keeps nothing of it, so
+file holds its reports in the order they arrived, those of a failed write
-the page keeps its samples and sends them again at its next report. Reports
+included. From the failed write until a write succeeds, `POST /api/v1/reports`
-already answered 200 are lost only if the server stops before a write succeeds;
+refuses every report with 500 and keeps nothing of it. The page sends the
-the final flush at stop tries them too, and if it fails, the process exits
+refused samples again in its next report, as long as it still holds them: at the
-non-zero.
+1-second interval one refused report already loses some of those in the history
+(see [Reporting](#reporting)). Reports already answered 200 are lost only if the
+server stops before a write succeeds; the final flush at stop tries them too,
+and if it fails, the process exits non-zero.
 
 #### Report limits
 
 `POST /api/v1/reports` takes reports from anyone who can reach it, without
 credentials, so it is bounded instead. Each refusal below answers with the same
-`{"status":"error"}` body as any other error.
+`{"status":"error"}` body as any other error, except nginx's refusal of a body
+over the limit in the image.
 
 - **JSON only.** A report must be sent with `Content-Type: application/json`,
   parameters such as `charset` allowed, as the page sends it; any other is
@@ -251,6 +294,16 @@ credentials, so it is bounded instead. Each refusal below answers with the same
   send a `text/plain` POST, or a form, without a CORS preflight; refusing them
   keeps any site from making its visitors' browsers store reports (see
   [CORS](#cors)).
+- **Body limit.** A report body may be at most 1 MiB, the backend's limit for
+  any request. Run on its own, the backend answers 413 before the route to a
+  request that declares a longer body. Without a declared length, the route
+  answers 413 only when the report it reads runs past 1 MiB; nothing after the
+  end of the report is read, so a longer body whose report ends sooner is
+  stored, and one that stops being valid JSON sooner is answered 400. In the
+  image, nginx refuses a body over 1 MiB first, with or without a declared
+  length, as its `client_max_body_size`, which `configs/nginx.conf` does not
+  set, is 1 MiB by default. nginx answers 413 with its own HTML error page, and
+  the backend never sees the request.
 - **Rate limit.** Each client address, resolved through `TRUSTED_PROXIES`, may
   send `REPORTS_PER_MINUTE` reports a minute; past that it gets 429 with
   `Retry-After: 60`. The minute slides: reports from the minute before still
@@ -274,6 +327,12 @@ credentials, so it is bounded instead. Each refusal below answers with the same
   how much of the newest reports is kept: the default of 1 GiB is small enough
   for any host; set it to the space you can give `DATA_DIR`.
 
+At the defaults, one address may send 60 reports of up to 1 MiB a minute, and a
+report made up to compress badly takes about as much room in the report files as
+its request body, so in about 17 minutes one address can fill the 1 GiB cap with
+its own reports, every report stored before them deleted to make room. Lowering
+`REPORTS_PER_MINUTE` or raising `DATA_DIR_MAX_BYTES` makes that take longer.
+
 #### CORS
 
 The page calls the API from the origin it is served from, so by default the
@@ -295,12 +354,16 @@ the metrics record its duration and response size, labelled with its path,
 method and status; they also count those requests in progress, and include Go's
 runtime and process metrics. No other request is recorded: not those to
 `/metrics` itself, and not those answered before they reach either route, such
-as a CORS preflight, or a request refused with 404 for a path no route has, 405
+as a CORS preflight, or a request refused with 404 for a path no route has or
-for a method its route does not take, or 413 for declaring a body length over
+405 for a method its route does not take. Run on its own, the backend also
-the 1 MiB limit. A report whose body goes over the limit without declaring its
+answers 413 before the route to a request declaring a body length over the 1 MiB
-length reaches the route, is answered 413 there, and is recorded with that
+limit, which is not recorded. A body over the limit sent without a declared
-status. Clients can make up any number of paths and methods, and each would add
+length reaches the route and is recorded with the status the route answers,
-labels to the metrics for as long as the server runs.
+which is 413 only when the report it reads runs past the limit: nothing after
+the end of the report is read. In the image neither reaches the backend: nginx
+answers both with 413 itself (see [Report limits](#report-limits)), and nothing
+of them is recorded. Clients can make up any number of paths and methods, and
+each would add labels to the metrics for as long as the server runs.
 
 `/metrics` is rate limited, so that its password cannot be guessed quickly: each
 client address, resolved through `TRUSTED_PROXIES`, may make 60 requests to it a
@@ -428,7 +491,10 @@ address nginx passes on and no other. The image:
   each entry with `netwatch-server check-cidr`, which parses it as the backend
   parses its own `TRUSTED_PROXIES`
 - Sends access logs to stdout
-- Caches static assets with immutable headers
+- Sends the page with `Cache-Control: no-cache`, so a browser that keeps a copy
+  checks with the server that it is current before using it again; only the
+  files under `/assets/`, whose names change whenever their contents do, are
+  cached as immutable
 - Sends the security headers `REPO_POLICIES.md` requires on every response, as
   `configs/security-headers.conf` sets them, in place of the backend's own
 - Stores reports in `DATA_DIR`, `/data/reports` by default, on the `/data`
@@ -456,7 +522,8 @@ What the [upaas](https://git.eeqj.de/sneak/upaas) app for netwatch needs:
       report files may take; the oldest are deleted to stay under it
     - `CORS_ALLOWED_ORIGINS`, default empty: other origins whose pages may call
       the API
-    - `DEBUG`, default `false`: debug logging
+    - `DEBUG`, default empty: debug logging; any non-empty value, `false` and
+      `0` included, turns it on
     - `DATA_DIR`, default `/data/reports`: the directory the reports are kept
       in: `/data` or a path below it, with no `.` or `..` part and no extra `/`.
       The container also stops if the path goes through a symbolic link that

REPO_POLICIES.md

@@ -1,6 +1,6 @@
 ---
 title: Repository Policies
-last_modified: 2026-10-04
+last_modified: 2026-10-18
 ---
 
 This document covers repository structure, tooling, and workflow standards. Code
@@ -54,34 +54,39 @@ style conventions are in separate documents:
   `cibuild`. `script/bootstrap` installs all dependencies idempotently and
   assumes nothing is present: base tools come from nix, apt, brew, or apk
   (detected in that order; apt runs noninteractive). For node it uses the
-  installed node if present; otherwise it installs a PINNED node version via
+  installed node only when its major version is the pinned one; otherwise (node
-  nvm, first installing nvm itself if missing — from a hash-verified GitHub
+  missing, or another major, such as the node 24 the CI runner image ships) it
-  release archive (never `curl | sh`), with bash installed as an explicit
+  installs the PINNED node version via nvm and installs yarn and the packages
-  prerequisite since nvm requires bash. yarn is then pinned via
+  under it. nvm itself is installed first if missing, from a hash-verified
-  `corepack prepare yarn@<version> --activate`. Never install "latest" or "lts";
+  GitHub release archive (never `curl | sh`), with bash installed as an explicit
-  always exact versions. `script/cibuild` runs the CI build: it changes to the
+  prerequisite since nvm requires bash. Only the major version is compared
-  repo root, runs `script/bootstrap`, runs `script/check`, and builds the image
+  because the `Dockerfile` stages start from a node image whose exact version is
-  with the version; the Gitea workflow calls it. **`script/cibuild` runs
+  not the pin, and nvm cannot install a prebuilt node on alpine. yarn is pinned
+  via `corepack prepare yarn@<version> --activate`. Never install "latest" or
+  "lts"; always exact versions. `script/cibuild` runs the CI build: it changes
+  to the repo root, runs `script/bootstrap`, runs `script/check`, and builds the
+  image with the version; the Gitea workflow calls it. **`script/cibuild` runs
   `script/bootstrap` first**, because the workflow checks out the repo and runs
   nothing else, while `script/fmt-check` runs the formatter on the host: on a
   pristine checkout with nothing installed the run dies there, after the
-  containerised gates have passed. **The bootstrap alone is not enough**:
+  containerised gates have passed. **The bootstrap alone is not enough**: when
-  `script/bootstrap` installs node and yarn under nvm and leaves neither on the
+  `script/bootstrap` installs node and yarn under nvm it leaves neither on the
-  `PATH` of the shell that called it, so a bare `yarn` still exits 127. The host
+  `PATH` of the shell that called it, so a bare `yarn` either exits 127 or runs
-  entrypoints that need yarn — `script/fmt` and `script/fmt-check` — therefore
+  under another node. The host entrypoints that need yarn — `script/fmt` and
-  source nvm for the pinned node version before invoking it, exactly as
+  `script/fmt-check` — therefore use the same test as `script/bootstrap`: when
-  `script/bootstrap`'s own install step does. A runner carrying nothing but
+  the node on `PATH` has the pinned major version they run the `yarn` on `PATH`,
-  docker and git then gets through `script/check`. Four further scripts are our
+  and otherwise they load nvm through `$HOME/.nvm/nvm.sh` and run yarn under the
-  own extensions to the standard: `script/check` runs `script/test`,
+  pinned node. A runner carrying nothing but docker and git then gets through
-  `script/lint` and `script/fmt-check`; `script/precommit` is what the git
+  `script/check`. Four further scripts are our own extensions to the standard:
-  pre-commit hook runs, and it calls `script/check`; `script/install-precommit`
+  `script/check` runs `script/test`, `script/lint` and `script/fmt-check`;
-  installs the git pre-commit hook (the `make hooks` target shims to it); and
+  `script/precommit` is what the git pre-commit hook runs, and it calls
-  `script/projectname` (literally that filename) simply outputs the project's
+  `script/check`; `script/install-precommit` installs the git pre-commit hook
-  name. Scripts that need the name call `script/projectname` — e.g.
+  (the `make hooks` target shims to it); and `script/projectname` (literally
-  `script/docker` assembles its image tag from it — so those scripts stay
+  that filename) simply outputs the project's name. Scripts that need the name
-  byte-identical across all repos. Repo-type-specific pre-commit extras (e.g.
+  call `script/projectname` — e.g. `script/docker` assembles its image tag from
-  `go mod tidy` verification in Go repos) belong in `script/precommit`, not in
+  it — so those scripts stay byte-identical across all repos. Repo-type-specific
-  the hook itself. Model scripts are at
+  pre-commit extras (e.g. `go mod tidy` verification in Go repos) belong in
+  `script/precommit`, not in the hook itself. Model scripts are at
   `https://git.eeqj.de/sneak/prompts/raw/branch/main/script/<name>`. The README
   must document the provided scripts in an **Entrypoints** section (see the
   README requirements below).
@@ -118,8 +123,10 @@ style conventions are in separate documents:
   and nothing else:
 
     ```sh
-    docker build --no-cache --target lint -t "$(script/projectname)-lint" .
+    docker build --build-arg CACHEBUST="$(date +%s)-$$" \
-    docker build --no-cache --target test -t "$(script/projectname)-test" .
+        --target lint --output type=cacheonly .
+    docker build --build-arg CACHEBUST="$(date +%s)-$$" \
+        --target test --output type=cacheonly .
     ```
 
     **A stage that is not the last one in the file is built only when the final
@@ -129,10 +136,15 @@ style conventions are in separate documents:
     plain `docker build .` builds the last stage alone and exits 0 having linted
     and tested nothing.
 
-    **Every `docker build` in `script/` is tagged**, here and in
+    **The gate builds write no image.** With `--output type=cacheonly` the phase
-    `script/cibuild` and `script/docker`. An untagged build leaves a dangling
+    runs and a failing step fails the build, but the result is not exported.
-    image behind on every invocation, on every developer host and every CI
+    Nothing uses those images, and writing one out is slow: a Go test phase's
-    runner; a tagged one replaces the previous image.
+    image holds the toolchain. A build given neither `--output` nor `-t` writes
+    an untagged image and leaves it dangling, on every developer host and every
+    CI runner. `script/cibuild` and `script/docker` build the image that ships
+    and tag it, so each build replaces the previous image; each assigns the tag
+    on its own line before the build, so `set -e` stops it where
+    `script/projectname` fails.
 
     Inside a phase the tool is invoked directly — `golangci-lint`, `go test`,
     `eslint`, `prettier` — never through `make lint` or `script/test`, which are
@@ -147,22 +159,42 @@ style conventions are in separate documents:
     one's findings, and a host-global lock in `$TMPDIR` makes concurrent runs
     exit non-zero with `parallel golangci-lint is running` — a status a caller
     cannot tell from real findings. Both have produced wrong verdicts in this
-    org, in both directions. A container has its own cache, its own `TMPDIR` and
+    org, in both directions. A container has its own `TMPDIR`, so the lock is
-    a digest-pinned binary, so neither is reachable.
+    out of reach. The lint phase does keep golangci-lint's cache between check
+    builds (below), but it lints every checkout at `/src`, and a finding is
+    taken from that cache only for the same files, settings and golangci-lint
+    version.
 
-- **Any build that runs checks is built with `--no-cache`.** Docker invalidates
+- **Any build that runs checks is given a new `CACHEBUST` value.** Docker
-  a `COPY` layer only when the copied content changes, so on an unchanged tree
+  invalidates a `COPY` layer only when the copied content changes, so on an
-  the check `RUN` is served from cache, nothing executes, and the build still
+  unchanged tree the check `RUN` is served from cache, nothing executes, and the
-  exits 0. Every `docker build` in `script/` therefore passes `--no-cache`:
+  build still exits 0. Every `docker build` in `script/` therefore passes
-  `script/lint`, `script/test`, `script/cibuild` and `script/docker` are the
+  `--build-arg CACHEBUST="$(date +%s)-$$"`, the time and the script's process
-  four, and there is no fifth — `script/check` runs the two gate phases and
+  id, so no two builds share a value, not even the three in one
-  `script/fmt-check`, and builds no image of its own. A bare `docker build .` is
+  `script/cibuild`. `script/lint`, `script/test`, `script/cibuild` and
-  not evidence that anything ran: a sub-second build reporting success is a
+  `script/docker` are the four, and there is no fifth — `script/check` runs the
-  cache hit, not a result. Never invalidate by pruning — `docker builder prune`
+  two gate phases and `script/fmt-check`, and builds no image of its own. In
-  and friends destroy a build cache shared with every other build on the host.
+  every stage of the `Dockerfile`, `ARG CACHEBUST` comes right after the steps
-  When a check is added or changed, prove it works by planting a defect it must
+  that install system packages and download Go's modules, or right after `FROM`
-  catch and watching the run fail on it, then revert the defect. A green run
+  in a stage with neither, and every step that lints, tests or compiles comes
-  alone shows neither that the check ran nor that it covers what it should.
+  after it. A step that installs anything else as well, such as a
+  `script/bootstrap` that also installs node packages, comes after it too. A new
+  value reruns every `RUN` step after that line; `COPY` steps are compared by
+  content and need nothing. Docker keeps the layers above the line until the
+  base image, their own lines or the files they copy change. A stage without the
+  `ARG` line is served from cache on an unchanged tree, so a repository
+  re-vendors the scripts and adds the `ARG` lines in one commit. The scripts do
+  not pass `--no-cache`: it reruns every step too, but it also builds the layers
+  above the line again and empties every cache mount, so Go's build cache and
+  golangci-lint's, which the canonical Go `Dockerfile` below mounts so that they
+  survive between check builds, would start empty every time. A bare
+  `docker build .` is not evidence that anything ran: a sub-second build
+  reporting success is a cache hit, not a result. Never invalidate by pruning —
+  `docker builder prune` and friends destroy a build cache shared with every
+  other build on the host. When a check is added or changed, prove it works by
+  planting a defect it must catch and watching the run fail on it, then revert
+  the defect. A green run alone shows neither that the check ran nor that it
+  covers what it should.
 
 - **The gate phases are separate stages, and the build stage depends on both.**
   The lint phase is based on the `golangci/golangci-lint` image (pinned by
@@ -171,40 +203,56 @@ style conventions are in separate documents:
   `Dockerfile`:
 
     ```dockerfile
+    # In every stage ARG CACHEBUST comes right after the steps that install
+    # system packages and download Go's modules, or right after FROM in a
+    # stage with neither, and every step that lints, tests or compiles
+    # comes after it. script/ passes a new value on every build, which
+    # reruns every RUN step after it. Docker keeps the layers above it
+    # until the base image, their own lines or go.mod and go.sum change.
+    # The steps that compile mount Go's build cache, and golangci-lint run
+    # also mounts golangci-lint's; both survive from build to build.
+
     # Lint phase
     # golangci/golangci-lint:v2.x.x, YYYY-MM-DD
     FROM golangci/golangci-lint@sha256:... AS lint
     WORKDIR /src
     COPY go.mod go.sum ./
     RUN go mod download
+    ARG CACHEBUST
     COPY . .
-    RUN golangci-lint run --config .golangci.yml ./...
+    RUN --mount=type=cache,target=/root/.cache/go-build \
+        --mount=type=cache,target=/root/.cache/golangci-lint \
+        golangci-lint run --config .golangci.yml ./...
 
     # Test phase. -race needs cgo and so a C compiler, which the Debian Go
-    # image ships and the alpine one does not.
+    # image ships and the alpine one does not. -count=1 because the mounted
+    # cache also holds test results, which go test would otherwise replay.
     # golang:1.x, YYYY-MM-DD
     FROM golang@sha256:... AS test
     WORKDIR /src
     COPY go.mod go.sum ./
     RUN go mod download
+    ARG CACHEBUST
     COPY . .
-    RUN go test -timeout 90s -race -cover ./... || \
+    RUN --mount=type=cache,target=/root/.cache/go-build \
+        go test -count=1 -timeout 90s -race -cover ./... || \
         { echo "--- Rerunning with -v for details ---"; \
-          go test -timeout 90s -race -v ./...; exit 1; }
+          go test -count=1 -timeout 90s -race -v ./...; exit 1; }
 
     # Build stage. Nothing is wanted from either phase above; the copies
-    # are what make BuildKit build them first, so this stage cannot run
+    # are what make BuildKit build them first, so nothing after them can
-    # unless lint and test passed.
+    # run unless lint and test passed.
     # golang:1.x-alpine, YYYY-MM-DD
     FROM golang@sha256:... AS builder
-    COPY --from=lint /src/go.sum /dev/null
-    COPY --from=test /src/go.sum /dev/null
     RUN apk add --no-cache git
-    # A tar-stream context keeps the sender's file owners, which git refuses.
-    RUN git config --system --add safe.directory /src
     WORKDIR /src
     COPY go.mod go.sum ./
     RUN go mod download
+    ARG CACHEBUST
+    COPY --from=lint /src/go.sum /dev/null
+    COPY --from=test /src/go.sum /dev/null
+    # A tar-stream context keeps the sender's file owners, which git refuses.
+    RUN git config --system --add safe.directory /src
     COPY . .
 
     # The VERSION build arg when one is given, otherwise
@@ -212,7 +260,8 @@ style conventions are in separate documents:
     # .git present, a version that is still empty, dev or unknown fails the
     # build: git is missing or could not read the checkout.
     ARG VERSION
-    RUN VERSION="${VERSION:-$(git describe --tags --always)}"; \
+    RUN --mount=type=cache,target=/root/.cache/go-build \
+        VERSION="${VERSION:-$(git describe --tags --always)}"; \
         if [ -e .git ]; then \
             case "$VERSION" in ""|dev|unknown) \
                 echo "version is '$VERSION' although .git is present" >&2; \
@@ -225,6 +274,7 @@ style conventions are in separate documents:
 
     # Runtime stage, and the last one
     FROM alpine@sha256:...
+    ARG CACHEBUST
     COPY --from=builder /app /usr/local/bin/app
     ENTRYPOINT ["app"]
     ```
@@ -239,6 +289,26 @@ style conventions are in separate documents:
     - Keep the runtime stage last, and if you add a stage after it, give it the
       same two copies. A plain `docker build .` builds the last stage's chain
       and nothing else.
+    - Check builds keep these, and none of them can hand back a stale result:
+        - The layers above `ARG CACHEBUST`: the system packages and Go's
+          modules. Docker installs the packages again when the base image or the
+          install line changes, and downloads the modules again when `go.mod` or
+          `go.sum` changes. Go checks every module against `go.sum` as it
+          downloads it, and no later build can change a kept layer. The modules
+          are kept in a layer rather than a cache mount because a cache mount is
+          shared by every build on the host, and Go builds from whatever files
+          it holds without checking them again (only `go mod verify` does), so a
+          file changed there would go unnoticed.
+        - Go's build cache (`/root/.cache/go-build`, where Go keeps it for
+          root), mounted on the steps that compile: `golangci-lint run`, both
+          `go test` runs and `go build`. Go reuses a compiled package only when
+          all its inputs are unchanged. The cache also holds test results, which
+          is why both `go test` runs carry `-count=1`.
+        - golangci-lint's cache (`/root/.cache/golangci-lint`, where the pinned
+          image keeps it), mounted on `golangci-lint run`. golangci-lint reuses
+          a package's findings only when the package's files and those of every
+          package it imports, `go.mod`, the linters and their settings in
+          `.golangci.yml`, and the golangci-lint version are all unchanged.
     - If the project uses `//go:embed` directives that reference build artifacts
       (e.g. a web frontend compiled in a separate stage), the lint phase must
       create placeholder files so the embed directives resolve. Example:
@@ -247,12 +317,20 @@ style conventions are in separate documents:
       in the lint phase. The `golangci/golangci-lint` image is Debian-based and
       has no `apk`, so install with `apt-get` under the Debian package name
       (`libvips-dev`, where alpine says `vips-dev`), and delete the package
-      lists in the same `RUN`, so the layer does not keep them:
+      lists in the same `RUN`, so the layer does not keep them. The test phase,
+      on the Debian Go image, installs the libraries its tests need the same
+      way. Either install goes right after `FROM`, above `ARG CACHEBUST`, so
+      Docker keeps it until the base image or the install line changes:
 
         ```dockerfile
+        FROM golangci/golangci-lint@sha256:... AS lint
         RUN apt-get update \
             && apt-get install -y --no-install-recommends libvips-dev \
             && rm -rf /var/lib/apt/lists/*
+        WORKDIR /src
+        COPY go.mod go.sum ./
+        RUN go mod download
+        ARG CACHEBUST
         ```
 
     - `.dockerignore` lets `.git` into the build context. It keeps out every git
@@ -278,20 +356,39 @@ style conventions are in separate documents:
       `.git` and the version still comes out empty, `dev` or `unknown`. A plain
       `docker build .` with no build arguments must succeed; a Dockerfile that
       refuses an empty build argument drops that refusal and keeps the argument.
+      A checkout whose `.git` is a file (a linked worktree, or a repository
+      checked out as a submodule) is the exception: that file points to a git
+      directory outside the build context, so the build cannot read the version
+      and a plain `docker build .` fails; pass the version with
+      `--build-arg VERSION=...`, as `script/docker` and `script/cibuild` already
+      do.
 
 - Every repo should have a Gitea Actions workflow (`.gitea/workflows/`) that
-  runs `script/cibuild` on push, and checks out the repo as its only other step.
+  runs `script/cibuild` on push, and checks out the repo as its only other step,
-  That script bootstraps, runs the gate phases, and then builds the image, so a
+  with `persist-credentials: false`: `script/cibuild` needs no token, and
-  successful run means every check passed; a bare `docker build .` does not
+  without it the checkout leaves the job's token in `.git/config` for every
+  later step. The checkout step also sets `fetch-depth: 0`, which fetches the
+  tags `git describe` needs: by default it clones shallow with no tags, and a
+  tagged repository's CI build would stamp a bare short commit id. The
+  workflow's `concurrency` block groups runs by workflow and branch
+  (`${{ github.workflow }}-${{ github.ref }}`) with `cancel-in-progress: true`,
+  so a new push cancels the older run on the same branch, queued or running, and
+  no other: runs for replaced commits do not hold up the shared runner.
+  `script/cibuild` bootstraps, runs the gate phases, and then builds the image,
+  so a successful run means every check passed; a bare `docker build .` does not
   carry the same guarantee, because its gate phases may come from the cache. The
-  image build is uncached and so runs the gate phases a second time. That is the
+  image build gets a `CACHEBUST` value of its own and so runs the gate phases a
-  price of the rule above, and it is worth paying: the image that ships is built
+  second time. That is the price of the rule above, and it is worth paying: the
-  from a run of its own gates rather than from a cache entry. A separate
+  image that ships is built from a run of its own gates rather than from a cache
-  workflow limited to `main` by a `branches` list under `on: push` cannot be
+  entry. The `check` job sets `timeout-minutes: 20`, so a hung build frees the
-  checked by review: to try a change to it, add the feature branch to that list
+  shared runner after 20 minutes. That allows for the three Docker builds
-  and push, then remove the branch from the list again before merging. Keep any
+  described above (the test phase, the lint phase, then the image), each held to
-  job in it that publishes behind `if: github.ref_name == 'main'`, so the run
+  the 5-minute Docker build limit below, plus the bootstrap. A separate workflow
-  from the feature branch publishes nothing.
+  limited to `main` by a `branches` list under `on: push` cannot be checked by
+  review: to try a change to it, add the feature branch to that list and push,
+  then remove the branch from the list again before merging. Keep any job in it
+  that publishes behind `if: github.ref_name == 'main'`, so the run from the
+  feature branch publishes nothing.
 
 - Use platform-standard formatters: `black` for Python, `prettier` for
   JS/CSS/Markdown/HTML, `go fmt` for Go. Always use default configuration with
@@ -351,12 +448,11 @@ style conventions are in separate documents:
     That cache is Go's own, separate from Docker's layer cache. Go stores a
     passing result in its cache directory (`GOCACHE`), and when the same tests
     run again on unchanged code it prints that result, marked `(cached)`,
-    without running them. That matters on a developer's machine, where this
+    without running them. The directory lasts from one run to the next on a
-    target runs and the directory lasts from one run to the next. The `test`
+    developer's machine, where this target runs, and in the `test` phase of the
-    phase of the `Dockerfile` needs no `-count=1`: its base image holds no
+    `Dockerfile`, which mounts it to keep compiled packages between check
-    result for this repo's tests and nothing before its `go test` step runs a
+    builds, so the phase's `go test` runs carry `-count=1` too. A new
-    test, so there is nothing to replay. `--no-cache` (above) is what makes that
+    `CACHEBUST` value (above) is what makes that step run on an unchanged tree.
-    step run on an unchanged tree.
 
     Python example:
 
@@ -384,9 +480,12 @@ style conventions are in separate documents:
 
 - `.gitignore` should be comprehensive from the start: OS files (`.DS_Store`),
   editor files (`.swp`, `*~`), in-repo agent scratch directories (`.claude/`),
-  language build artifacts, and `node_modules/`. Fetch the standard `.gitignore`
+  `node_modules/`, and the repo's own build outputs. Fetch the standard
-  from `https://git.eeqj.de/sneak/prompts/raw/branch/main/.gitignore` when
+  `.gitignore` from
-  setting up a new repo. These patterns are written to `.gitignore`'s own
+  `https://git.eeqj.de/sneak/prompts/raw/branch/main/.gitignore` when setting up
+  a new repo. A repo's `.gitignore` is the standard file followed by the repo's
+  own entries, such as its binaries; a re-vendor replaces the standard part and
+  keeps those entries. These patterns are written to `.gitignore`'s own
   semantics, in which an unanchored pattern already matches at every depth; they
   are not a `.dockerignore` and must not be transplanted into one unmodified.
 
@@ -434,13 +533,15 @@ style conventions are in separate documents:
   byte-identically across repos:
 
     ```sh
-    # Own line: a failing command substitution inside an argument does not
+    # The version and the tag each get their own line: a failing command
-    # trip `set -e`, so the inline form degrades to an empty constant.
+    # substitution inside an argument does not trip `set -e`, so the inline
+    # form degrades to an empty constant.
     version="$(git describe --tags --always --dirty 2>/dev/null || true)"
     [ -n "$version" ] || version="unknown"
-    docker build --no-cache \
+    tag="$(script/projectname)"
+    docker build --build-arg CACHEBUST="$(date +%s)-$$" \
         --build-arg VERSION="$version" \
-        -t "$(script/projectname)" .
+        -t "$tag" .
     ```
 
     `--always` makes an untagged repo yield an abbreviated commit hash rather
@@ -452,8 +553,8 @@ style conventions are in separate documents:
     there because `ARG` is stage-scoped; passing `VERSION` to a repo whose
     Dockerfile declares no such `ARG` is ignored and costs nothing, which is why
     the scripts stay byte-identical. One consequence for CI: the standard
-    checkout action clones shallow and fetches no tags, so a repo that embeds a
+    checkout action clones shallow and fetches no tags, so the canonical
-    tag-derived version must set `fetch-depth: 0` on its checkout step.
+    `.gitea/workflows/check.yml` sets `fetch-depth: 0` on its checkout step.
 
 - **Verify `.dockerignore` by enumerating the image, not by reading the
   patterns.** Plant files at the root _and_ at least two directories deep, build
@@ -520,6 +621,11 @@ style conventions are in separate documents:
       function defined and never invoked has the same exit status and the same
       empty output as one that worked.
 
+    Node is the exception to the whole-token comparison: the canonical
+    `script/bootstrap` compares only its major version, because the node 22
+    alpine image the `Dockerfile` stages start from is not the exact pin, and
+    nvm cannot install a prebuilt node on alpine.
+
     Keep it POSIX sh: no arrays, no `[[`, no `grep -P`.
 
     A Go tool a repo needs on the host is installed with `go install` pinned to
@@ -636,7 +742,11 @@ style conventions are in separate documents:
       Never edit existing migrations after release.
 
 - All repos should have an `.editorconfig` enforcing the project's indentation
-  settings.
+  settings: the standard file from
+  `https://git.eeqj.de/sneak/prompts/raw/branch/main/.editorconfig`, which sets
+  tabs for `Makefile` and Go files, followed by the repo's own sections, such as
+  one for another language it uses. A re-vendor replaces the standard part and
+  keeps those sections.
 
 - Avoid putting files in the repo root unless necessary. Root should contain
   only project-level config files (`README.md`, `AGENTS.md`, `Makefile`,

TODO.md

@@ -21,6 +21,97 @@ Run `make frontend-viewport-test` in CI as its own step; it is not part of
 
 # Completed Steps
 
+- 2026-10-19: report file writes no longer overlap
+  ([#150](https://git.eeqj.de/sneak/netwatch/issues/150)), so the reports of
+  failed writes are written in the order they arrived. Two writes that failed at
+  the same time put their reports back in the wrong order when the older failed
+  first. One write runs at a time, and each takes the waiting reports only when
+  it starts, after a failed write has put its own back. The write `Append`
+  starts when 10 MiB are waiting waits for any write under way, and `Append`
+  starts no other while it waits. Files written since the start are deleted to
+  make room in the order they were written, after those found at start. A unit
+  test holds a write open, starts a second at 10 MiB, fails both and checks the
+  order in the next file. Others check that `Append` starts no second write at
+  10 MiB while one waits, that it starts one again the next time 10 MiB are
+  waiting, and that the first file written since the start is the first deleted
+- 2026-10-19: `DEBUG` unset or empty leaves debug logging off, and any other
+  value, `false` and `0` included, turns it on, as the code style guide says
+  ([#159](https://git.eeqj.de/sneak/netwatch/issues/159)). No value of it stops
+  the start any more; before, only `1`, `t`, `T`, `TRUE`, `true`, `True`, `0`,
+  `f`, `F`, `FALSE`, `false` and `False` were accepted. `README.md` no longer
+  gives `false` as its default
+- 2026-10-19: check builds keep Go's modules and compiled packages between runs
+  ([#118](https://git.eeqj.de/sneak/netwatch/issues/118)), on sneak's rulings in
+  [prompts#113](https://git.eeqj.de/sneak/prompts/issues/113) and
+  [prompts#130](https://git.eeqj.de/sneak/prompts/issues/130).
+  `REPO_POLICIES.md`, `script/test`, `script/lint`, `script/docker` and
+  `script/cibuild` are the sneak/prompts copies at `0653579`: the scripts pass a
+  new `CACHEBUST` value where they passed `--no-cache`, and the lint and test
+  builds write no image. In each stage of `Dockerfile`, `ARG CACHEBUST` comes
+  after the system packages and the Go module download, and before every step
+  that lints, tests or compiles and the node stages' `yarn install`. Go's build
+  cache and golangci-lint's are cache mounts, and both `go test` runs carry
+  `-count=1`. The test phase's Go caches used to be in memory, empty on every
+  build
+- 2026-10-19: the three unit tests of a check whose result is discarded (its
+  round given up, the user paused, or the first round) also check that the
+  host's history is still empty, its status still `pending`, and nothing was
+  written to the console
+  ([#151](https://git.eeqj.de/sneak/netwatch/issues/151)). They checked only the
+  row, so a change that recorded the result and skipped only the row would have
+  kept them passing
+- 2026-10-18: when the backend runs on its own, a `DATA_DIR` that cannot be
+  created or read, or where no file can be created, stops the start with an
+  error naming `DATA_DIR` and its value, as the other settings' errors do
+  ([#148](https://git.eeqj.de/sneak/netwatch/issues/148)); the first two named
+  only the path, and the third was found only at the first write of a report
+  file. At start the server creates a file in `DATA_DIR` and removes it. Unit
+  tests start the buffer with `DATA_DIR` below a regular file and with
+  `DATA_DIR` set to `/proc`, and check the error. The report file number is an
+  `atomic.Int64`, as the Go style guide asks for int types
+  ([#161](https://git.eeqj.de/sneak/netwatch/issues/161)); file names are
+  unchanged
+- 2026-10-18: the image's nginx sends the page with `Cache-Control: no-cache`
+  ([#141](https://git.eeqj.de/sneak/netwatch/issues/141)), so a browser that
+  keeps a copy checks with the server that it is current before using it again.
+  The page went out with no `Cache-Control`, so a browser could keep it after a
+  deploy and ask for files under `/assets/` that the new build no longer has,
+  and show an empty page. A path that `location /` answers with the page through
+  its `try_files` fallback gets the same header. `location /` now includes the
+  security headers again, as `/assets/` does, since its own `add_header` drops
+  the server's. `/assets/` keeps its immutable caching.
+  `make frontend-viewport-test` fails unless `/` and a path that falls back to
+  the page carry `Cache-Control: no-cache` and `Content-Security-Policy`, which
+  comes from the same file as the other security headers
+- 2026-10-18: `README.md` says what the page loses while reports are refused or
+  the page is paused ([#145](https://git.eeqj.de/sneak/netwatch/issues/145)),
+  how fast one address can replace the stored reports
+  ([#146](https://git.eeqj.de/sneak/netwatch/issues/146)), and that in the image
+  nginx answers a body over 1 MiB itself
+  ([#147](https://git.eeqj.de/sneak/netwatch/issues/147)). A sample in the
+  history that no report has delivered is sent only as long as the history holds
+  it: at the 1-second interval one refused or failed report loses the oldest 20
+  seconds of each host's samples, and a pause longer than the history loses
+  every sample between the last delivered report and the pause. Samples an
+  interval change cleared are kept apart, the last 100 per host, until they are
+  delivered. At the defaults, one address sending 1 MiB reports made up to
+  compress badly can fill the 1 GiB cap in about 17 minutes. nginx's
+  `client_max_body_size` is also 1 MiB by default, so in the image it answers a
+  larger body with 413 and its own HTML page, and the backend never sees or
+  records it; the backend's own 413 sentences now say they apply to it run on
+  its own. The reporter, the body limit and nginx are unchanged
+- 2026-10-18: a browser that cannot format the header's clocks still probes
+  ([#149](https://git.eeqj.de/sneak/netwatch/issues/149)). The clocks name the
+  local time zone through `Intl.DateTimeFormat`'s `shortGeneric` option, which
+  Chrome before 95, Firefox before 91 and Safari before 15.4 reject with an
+  error; it came before the first round started, so such a browser never checked
+  a target. The clocks now start last, after the rounds, and an error formatting
+  them leaves both blank. The footer's color legend reads `>=500ms` for red,
+  where it read `>500ms`, so 500 ms, which is red, is in one of its colors
+  ([#152](https://git.eeqj.de/sneak/netwatch/issues/152)). Unit tests check the
+  clocks with a working `Intl.DateTimeFormat` and with one that rejects
+  `shortGeneric`, and draw the page and check the legend against every entry of
+  `CONFIG.latencyColors`
 - 2026-10-18: `netwatch-server prepare-data-dir` no longer changes the owner of
   a file outside `/data` through a hard link in the volume
   ([#143](https://git.eeqj.de/sneak/netwatch/issues/143)). A file with a second
@@ -29,12 +120,29 @@ Run `make frontend-viewport-test` in CI as its own step; it is not part of
   swapped in after the check is caught too; a symbolic link is opened without
   following it and given to the netwatch user itself, so the server can still
   follow one in a sticky directory that anyone can write to. Devices, named
-  pipes and sockets keep their owner. Opening a file without following it and
+  pipes and sockets keep their owner. Setting the owner on an open file that is
-  setting the owner on the open file needs Linux, the system of the image; on
+  not a directory is built only for Linux: `prepare-data-dir` runs only in the
-  any other system the package still builds, and `prepare-data-dir` fails at the
+  Linux image, and macOS, where the backend is also built for development, has
-  first file that is not a directory. The test that runs as root hard-links a
+  neither `O_PATH` nor `AT_EMPTY_PATH`. On any other system the package still
-  file, a symbolic link and a named pipe from outside the volume into it and
+  builds, and `prepare-data-dir` fails at the first file that is not a
-  finds their owners unchanged
+  directory. The test that runs as root hard-links a file, a symbolic link and a
+  named pipe from outside the volume into it and finds their owners unchanged
+- 2026-10-18: from a trusted proxy, the backend takes the client address from
+  the right-most `X-Forwarded-For` entry that is not in `TRUSTED_PROXIES`
+  ([#144](https://git.eeqj.de/sneak/netwatch/issues/144)). It took the left-most
+  entry, which the client writes itself behind a proxy that appends to the
+  header, so a client could choose the address the rate limit counts and the
+  request log records. An entry written as an address with a port is read as
+  that address. With only trusted addresses in the header, or an entry that is
+  not an address at all (such as `unknown`) met on the way, the direct peer is
+  used. `X-Real-IP` is read only without `X-Forwarded-For`, and every
+  `X-Forwarded-For` header line is read, not only the first. The image's nginx,
+  and the `TRUSTED_PROXIES=127.0.0.1/32` it starts the backend with, are
+  unchanged; the comment in `configs/nginx.conf` that said the backend takes the
+  first entry now says it takes the one address nginx passes on. Unit tests
+  resolve the address from chains ending in an untrusted address, in trusted
+  ones, of only trusted ones, with ports, with an entry that is not an address,
+  and split over two header lines
 - 2026-10-18: after an interval change, the summary's Reachable slot reads
   `--/--` in white, as at page load, until the second round since the change
   ends ([#140](https://git.eeqj.de/sneak/netwatch/issues/140)). It used to read
@@ -545,9 +653,3 @@ Run `make frontend-viewport-test` in CI as its own step; it is not part of
 - A configurable host list (an environment variable or a config file)
 - Export of the latency history (CSV or JSON)
 - A notification when the health status changes to DEGRADED
-- `make test` within the 20-second target
-  ([#118](https://git.eeqj.de/sneak/netwatch/issues/118)): every run compiles
-  the Go standard library and every dependency with the race detector from
-  nothing. Whether test builds may keep Go's compiled packages between runs
-  waits on sneak's ruling in
-  [prompts#113](https://git.eeqj.de/sneak/prompts/issues/113)

configs/nginx.conf

@@ -30,8 +30,13 @@ server {
     access_log /dev/stdout combined;
     error_log  /dev/stderr warn;
 
+    # With no-cache, a browser never uses a copy of the page that names
+    # assets a newer build no longer has without first checking with the
+    # server that it is current.
     location / {
         try_files $uri $uri/ /index.html;
+        add_header Cache-Control no-cache;
+        include /etc/nginx/security-headers.conf;
     }
 
     # Cache static assets aggressively
@@ -46,7 +51,9 @@ server {
     # 127.0.0.1:8081. These headers go with every request passed to it.
     # X-Forwarded-For carries only the client address, as resolved by
     # the real IP settings above, and not the chain the request came
-    # with: the backend takes the first entry, which a client can write.
+    # with. bin/entrypoint.sh starts the backend with
+    # TRUSTED_PROXIES=127.0.0.1/32, the address nginx connects from, so
+    # the backend takes that one entry as the client address.
     proxy_set_header Host $host;
     proxy_set_header X-Real-IP $remote_addr;
     proxy_set_header X-Forwarded-For $remote_addr;

internal/config/config.go

@@ -41,7 +41,6 @@ var (
 		"must be an origin, scheme://host with an optional port",
 	)
 	errNotPort = errors.New("must be a port number, 1 to 65535")
-	errNotBool = errors.New("must be true or false")
 	errNotIP   = errors.New("must be an IP address, or empty")
 
 	errMetricsCredentials = errors.New(
@@ -101,7 +100,7 @@ func New(
 	viper.SetDefault("CORS_ALLOWED_ORIGINS", "")
 	viper.SetDefault("DATA_DIR", "./data/reports")
 	viper.SetDefault("DATA_DIR_MAX_BYTES", defaultDataDirMaxBytes)
-	viper.SetDefault("DEBUG", "false")
+	viper.SetDefault("DEBUG", "")
 	// An empty BIND_ADDRESS listens on every interface.
 	viper.SetDefault("BIND_ADDRESS", "")
 	viper.SetDefault("PORT", "8080")
@@ -120,19 +119,17 @@ func New(
 		}
 	}
 
-	// Read with strconv: viper's GetInt and GetBool would read a value
+	// Read with strconv: viper's GetInt would read a value it cannot
-	// they cannot parse as 0 or false instead of failing.
+	// parse as 0 instead of failing.
 	port, err := strconv.Atoi(viper.GetString("PORT"))
 	if err != nil || port < 1 || port > math.MaxUint16 {
 		return nil, fmt.Errorf("PORT %q: %w",
 			viper.GetString("PORT"), errNotPort)
 	}
 
-	debug, err := strconv.ParseBool(viper.GetString("DEBUG"))
+	// Any non-empty value turns debug logging on, "false" and "0"
-	if err != nil {
+	// included, as the code style guide says.
-		return nil, fmt.Errorf("DEBUG %q: %w",
+	debug := viper.GetString("DEBUG") != ""
-			viper.GetString("DEBUG"), errNotBool)
-	}
 
 	// An empty TRUSTED_PROXIES trusts no proxy, but viper reads an
 	// empty environment variable as unset and would give the default,
@@ -172,8 +169,8 @@ func New(
 }
 
 // check fails with an error naming the first setting here whose value
-// the server cannot use. New checks PORT and DEBUG as it reads them,
+// the server cannot use. New checks PORT as it reads it, and the
-// and the middleware checks TRUSTED_PROXIES as it parses it.
+// middleware checks TRUSTED_PROXIES as it parses it.
 func (s *Config) check() error {
 	// viper reads a value that is not a number as 0, so this also
 	// catches a mistyped setting.

internal/config/config_test.go

@@ -127,12 +127,52 @@ func TestPortMustBeAPortNumber(t *testing.T) {
 	}
 }
 
-// TestDebugMustBeTrueOrFalse: viper reads any other value, such as
+// TestDebugIsOnForAnyNonEmptyValue: unset or empty, debug logging is
-// "yes", as false.
+// off; any other value turns it on, "false" and "0" included, and none
-func TestDebugMustBeTrueOrFalse(t *testing.T) {
+// stops the start.
-	t.Setenv("DEBUG", "yes")
+func TestDebugIsOnForAnyNonEmptyValue(t *testing.T) {
+	for _, tc := range []struct {
+		name  string
+		value string
+		unset bool
+		want  bool
+	}{
+		{name: "unset", unset: true, want: false},
+		{name: "empty", value: "", want: false},
+		{name: "false", value: "false", want: true},
+		{name: "0", value: "0", want: true},
+		{name: "yes", value: "yes", want: true},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			// t.Setenv first, so the variable is put back as it was
+			// after the test.
+			t.Setenv("DEBUG", tc.value)
 
-	requireConfigError(t, "DEBUG")
+			if tc.unset {
+				err := os.Unsetenv("DEBUG")
+				if err != nil {
+					t.Fatal(err)
+				}
+			}
+
+			var cfg *config.Config
+
+			app := fx.New(
+				fx.NopLogger,
+				fx.Provide(globals.New, logger.New, config.New),
+				fx.Populate(&cfg),
+			)
+
+			err := app.Err()
+			if err != nil {
+				t.Fatalf("config error = %v", err)
+			}
+
+			if cfg.Debug != tc.want {
+				t.Fatalf("Debug = %t, want %t", cfg.Debug, tc.want)
+			}
+		})
+	}
 }
 
 // TestBindAddressMustBeAnIPAddress: a host name would be looked up

internal/middleware/middleware.go

@@ -11,6 +11,7 @@ import (
 	"net/http"
 	"net/netip"
 	"runtime/debug"
+	"slices"
 	"strings"
 	"time"
 
@@ -133,6 +134,17 @@ func ipFromHostPort(hostPort string) string {
 // X-Real-IP are honoured only when the direct peer is a
 // trusted proxy; otherwise the direct peer is returned so a
 // spoofed header cannot forge the logged address.
+//
+// Each proxy appends to X-Forwarded-For the address it took the
+// request from, so the entries on the right are written by the
+// trusted proxies and those further left by the client. The
+// client's address is therefore the right-most entry that is not
+// a trusted proxy. An entry written as an address with a port is
+// read as that address, both for the trust check and as the
+// result. When every entry is a trusted proxy, or the walk from
+// the right meets an entry that is not an address at all (such as
+// "unknown"), the peer is used. X-Real-IP is read only when there
+// is no X-Forwarded-For entry at all.
 func clientIP(
 	remoteAddr string,
 	header http.Header,
@@ -144,28 +156,49 @@ func clientIP(
 		return peer
 	}
 
-	if xff := firstForwardedFor(header.Get("X-Forwarded-For")); xff != "" {
+	forwarded := forwardedFor(header)
-		return xff
+
+	if len(forwarded) == 0 {
+		if xr := strings.TrimSpace(header.Get("X-Real-IP")); validIP(xr) {
+			return xr
+		}
+
+		return peer
 	}
 
-	if xr := strings.TrimSpace(header.Get("X-Real-IP")); validIP(xr) {
+	for _, entry := range slices.Backward(forwarded) {
-		return xr
+		if !validIP(entry) {
+			return peer
+		}
+
+		if !addrInAny(entry, trusted) {
+			return entry
+		}
 	}
 
 	return peer
 }
 
-// firstForwardedFor returns the left-most valid address in an
+// forwardedFor returns the entries of X-Forwarded-For in order,
-// X-Forwarded-For list (the original client), or "" if none.
+// leaving out blank ones. A proxy may add its entry as a header
-func firstForwardedFor(value string) string {
+// line of its own rather than append it to the line the client
+// sent, so every line is read, not only the first. Some proxies
+// write an entry as an address with a port, such as
+// 198.51.100.7:5000 or [2001:db8::1]:443; the port is removed so
+// that the entry is read as its address.
+func forwardedFor(header http.Header) []string {
+	value := strings.Join(header.Values("X-Forwarded-For"), ",")
+
+	var entries []string
+
 	for part := range strings.SplitSeq(value, ",") {
-		candidate := strings.TrimSpace(part)
+		entry := strings.TrimSpace(part)
-		if validIP(candidate) {
+		if entry != "" {
-			return candidate
+			entries = append(entries, ipFromHostPort(entry))
 		}
 	}
 
-	return ""
+	return entries
 }
 
 func validIP(s string) bool {

internal/middleware/middleware_test.go

@@ -25,10 +25,16 @@ import (
 const (
 	// loopbackPeer is a remote address inside the trusted-proxy allowlist.
 	loopbackPeer = "127.0.0.1:5000"
+	// loopbackIP is the address of loopbackPeer, which ClientIP returns
+	// when it uses the peer.
+	loopbackIP = "127.0.0.1"
 	// forwardedIP is the client address presented via X-Forwarded-For.
 	forwardedIP = "203.0.113.7"
 	// realIP is the client address presented via X-Real-IP.
 	realIP = "203.0.113.9"
+	// appendedIP is the X-Forwarded-For entry a trusted proxy appended
+	// after the entries the client wrote.
+	appendedIP = "198.51.100.7"
 )
 
 func mustPrefixes(t *testing.T, cidrs ...string) []netip.Prefix {
@@ -85,10 +91,35 @@ func clientIPCases() []clientIPCase {
 			want:       forwardedIP,
 		},
 		{
-			name:       "trusted proxy uses left-most of chain",
+			name:       "trusted proxy uses right-most of chain",
-			remoteAddr: "10.1.2.3:5000",
+			remoteAddr: loopbackPeer,
-			xff:        forwardedIP + ", 10.1.2.3",
+			xff:        "203.0.113.9, " + appendedIP,
-			want:       forwardedIP,
+			want:       appendedIP,
+		},
+		{
+			name:       "trusted proxy entries are skipped from the right",
+			remoteAddr: loopbackPeer,
+			xff:        "203.0.113.9, 10.0.0.5",
+			want:       "203.0.113.9",
+		},
+		{
+			name:       "chain of only trusted proxies uses peer",
+			remoteAddr: loopbackPeer,
+			xff:        "10.0.0.5, 10.1.2.3",
+			want:       loopbackIP,
+		},
+		{
+			name:       "entry that is not an address uses peer",
+			remoteAddr: loopbackPeer,
+			xff:        "203.0.113.9, unknown",
+			want:       loopbackIP,
+		},
+		{
+			name:       "x-real-ip ignored when forwarded-for has entries",
+			remoteAddr: loopbackPeer,
+			xff:        "10.0.0.5",
+			xRealIP:    realIP,
+			want:       loopbackIP,
 		},
 		{
 			name:       "trusted proxy falls back to x-real-ip",
@@ -117,7 +148,7 @@ func clientIPCases() []clientIPCase {
 			name:       "trusted proxy with garbage header uses peer",
 			remoteAddr: loopbackPeer,
 			xff:        "not-an-ip",
-			want:       "127.0.0.1",
+			want:       loopbackIP,
 		},
 	}
 }
@@ -148,6 +179,47 @@ func TestClientIP(t *testing.T) {
 	}
 }
 
+// TestClientIPReadsEveryForwardedForLine checks that an entry a proxy
+// adds as an X-Forwarded-For line of its own is read, and the line the
+// client sent before it is not taken.
+func TestClientIPReadsEveryForwardedForLine(t *testing.T) {
+	t.Parallel()
+
+	header := http.Header{}
+	header.Add("X-Forwarded-For", "203.0.113.9")
+	header.Add("X-Forwarded-For", appendedIP)
+
+	got := middleware.ClientIP(loopbackPeer, header,
+		mustPrefixes(t, "127.0.0.1/32"))
+	if got != appendedIP {
+		t.Errorf("ClientIP() = %q, want %q", got, appendedIP)
+	}
+}
+
+// TestClientIPReadsAddressWithPort checks that an X-Forwarded-For entry
+// written as an address with a port is read as that address, both for
+// the trust check and as the result.
+func TestClientIPReadsAddressWithPort(t *testing.T) {
+	t.Parallel()
+
+	trusted := mustPrefixes(t, "127.0.0.1/32", "10.0.0.0/8")
+
+	for xff, want := range map[string]string{
+		"203.0.113.9, " + appendedIP + ":5000": appendedIP,
+		"203.0.113.9, [2001:db8::1]:443":       "2001:db8::1",
+		"203.0.113.9, 10.0.0.5:5000":           "203.0.113.9",
+	} {
+		header := http.Header{}
+		header.Set("X-Forwarded-For", xff)
+
+		got := middleware.ClientIP(loopbackPeer, header, trusted)
+		if got != want {
+			t.Errorf("X-Forwarded-For %q: ClientIP() = %q, want %q",
+				xff, got, want)
+		}
+	}
+}
+
 func TestSecurityHeaders(t *testing.T) {
 	t.Parallel()
 

internal/reportbuf/datadir_linux.go

@@ -24,9 +24,10 @@ import (
 // its number of names from the open file and gives that file to them,
 // so a hard link swapped in at name is caught too.
 //
-// This file is built only on Linux, because O_PATH and AT_EMPTY_PATH
+// This file is built only on Linux: prepare-data-dir runs only in the
-// exist only there; prepare-data-dir runs only in the Linux image. On
+// Linux image, and macOS, where the backend is also built for
-// any other system chownFile returns an error (see datadir_other.go).
+// development, has neither O_PATH nor AT_EMPTY_PATH. On any other
+// system chownFile returns an error (see datadir_other.go).
 func chownFile(root *os.Root, name string, uid, gid int) error {
 	file, err := root.OpenFile(name, unix.O_PATH, 0)
 	if err != nil {

internal/reportbuf/datadir_other.go

@@ -10,11 +10,12 @@ import (
 
 // chownFile returns an error on every system but Linux. Opening a file
 // that is not a directory without reading it or following it, and
-// giving it to a user on the open file, never by name, needs O_PATH
+// giving it to a user on the open file, never by name, is built only
-// and AT_EMPTY_PATH, which only Linux has (see datadir_linux.go), and
+// for Linux (see datadir_linux.go): prepare-data-dir runs only in the
-// prepare-data-dir runs only in the Linux image. This file lets the
+// Linux image, and macOS, where the backend is also built for
-// package build elsewhere, such as on macOS, where a developer runs
+// development, has neither O_PATH nor AT_EMPTY_PATH. This file lets
-// make build and make run.
+// the package build elsewhere, such as on macOS, where a developer
+// runs make build and make run.
 func chownFile(_ *os.Root, name string, _, _ int) error {
 	return &fs.PathError{Op: "chown", Path: name, Err: errors.ErrUnsupported}
 }

internal/reportbuf/datadir_test.go

@@ -277,6 +277,16 @@ func TestPrepareDataDirGivesVolumeToOwner(t *testing.T) {
 		t.Fatal(err)
 	}
 
+	err = os.Symlink("f", filepath.Join(outside, "link"))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	err = syscall.Mkfifo(filepath.Join(outside, "fifo"), 0o600)
+	if err != nil {
+		t.Fatal(err)
+	}
+
 	err = os.Mkdir(old, 0o700)
 	if err != nil {
 		t.Fatal(err)
@@ -292,38 +302,19 @@ func TestPrepareDataDirGivesVolumeToOwner(t *testing.T) {
 		t.Fatal(err)
 	}
 
-	err = os.Link(filepath.Join(outside, "f"), filepath.Join(old, "hardlink"))
+	err = syscall.Mkfifo(filepath.Join(old, "fifo"), 0o600)
 	if err != nil {
 		t.Fatal(err)
 	}
 
 	// os.Link does not follow a symbolic link: it gives the link itself
 	// a second name.
-	err = os.Symlink("f", filepath.Join(outside, "link"))
+	for _, name := range []string{"f", "link", "fifo"} {
-	if err != nil {
+		err = os.Link(filepath.Join(outside, name),
-		t.Fatal(err)
+			filepath.Join(old, "hardlink-to-"+name))
-	}
+		if err != nil {
-
+			t.Fatal(err)
-	err = os.Link(filepath.Join(outside, "link"),
+		}
-		filepath.Join(old, "hardlink-to-link"))
-	if err != nil {
-		t.Fatal(err)
-	}
-
-	err = syscall.Mkfifo(filepath.Join(outside, "fifo"), 0o600)
-	if err != nil {
-		t.Fatal(err)
-	}
-
-	err = os.Link(filepath.Join(outside, "fifo"),
-		filepath.Join(old, "hardlink-to-fifo"))
-	if err != nil {
-		t.Fatal(err)
-	}
-
-	err = syscall.Mkfifo(filepath.Join(old, "fifo"), 0o600)
-	if err != nil {
-		t.Fatal(err)
 	}
 
 	err = reportbuf.PrepareDataDir(volume, filepath.Join(volume, reports),

internal/reportbuf/reportbuf.go

@@ -16,7 +16,6 @@ import (
 	"log/slog"
 	"os"
 	"path/filepath"
-	"slices"
 	"strings"
 	"sync"
 	"sync/atomic"
@@ -73,11 +72,9 @@ type Buffer struct {
 	// except in tests that hold the write open or make it fail.
 	fileCreated func(f *os.File)
 	// files are the report files that may be deleted to make room,
-	// in name order, which is oldest first: those in dataDir at
+	// oldest first: those in dataDir at start, in name order, then
-	// start, and each one this buffer writes, put in at its place by
+	// each one this buffer writes, once it is complete. A file still
-	// name once it is complete, even when an older file's write
+	// being written is not among them. filesBytes is their total size.
-	// completes after a newer one's. A file still being written is
-	// not among them. filesBytes is their total size.
 	files      []reportFile
 	filesBytes int64
 	log        *slog.Logger
@@ -88,8 +85,12 @@ type Buffer struct {
 	now func() time.Time
 	// seq numbers the report files, so that two named in the same
 	// millisecond still get different names.
-	seq      atomic.Uint64
+	seq      atomic.Int64
 	stopOnce sync.Once
+	// thresholdWriteWaiting is set while a write Append started at
+	// flushSizeThreshold waits for writeMu; Append starts no other
+	// while it is set.
+	thresholdWriteWaiting atomic.Bool
 	// usedBytes is what Append checks against maxBytes: the size
 	// of the report files in dataDir, plus the reports waiting to
 	// be written to one at their uncompressed size.
@@ -99,6 +100,11 @@ type Buffer struct {
 	// refuses reports, and the reports of that write wait in buf for
 	// the next flush.
 	writeErr error
+	// writeMu is held across each write of a report file, so writes
+	// never overlap: a failed write puts its reports back in buf before
+	// the next write takes them, which keeps them in the order they
+	// were appended.
+	writeMu sync.Mutex
 	// writers are the goroutines that write report files: flushLoop,
 	// and each one Append starts at flushSizeThreshold. OnStop waits
 	// for them before the final flush.
@@ -127,16 +133,33 @@ func New(
 
 	lc.Append(fx.Hook{
 		OnStart: func(_ context.Context) error {
+			// The errors name DATA_DIR, as config.New's do for the
+			// settings it checks.
 			err := os.MkdirAll(b.dataDir, dirPerms)
 			if err != nil {
-				return fmt.Errorf("create data dir: %w", err)
+				return fmt.Errorf("DATA_DIR %q: %w", b.dataDir, err)
+			}
+
+			// Creating and removing a file stops the start when no
+			// file can be created in DATA_DIR, instead of the first
+			// write of a report file failing after reports were
+			// answered 200. A full disk is not caught: the empty
+			// file is still created.
+			f, err := os.CreateTemp(b.dataDir, "")
+			if err != nil {
+				return fmt.Errorf("DATA_DIR %q: %w", b.dataDir, err)
+			}
+
+			err = errors.Join(f.Close(), os.Remove(f.Name()))
+			if err != nil {
+				return fmt.Errorf("DATA_DIR %q: %w", b.dataDir, err)
 			}
 
 			// Report files left by earlier runs count too, and are
 			// the first to be deleted to make room.
 			files, err := reportFiles(b.dataDir)
 			if err != nil {
-				return err
+				return fmt.Errorf("DATA_DIR %q: %w", b.dataDir, err)
 			}
 
 			b.mu.Lock()
@@ -190,8 +213,9 @@ func New(
 // write of a report file failed, until a write succeeds again. If the
 // line would take usedBytes past maxBytes, the oldest report files are
 // deleted to make room; it stores nothing and returns ErrFull if that
-// cannot make room. If the buffer reaches the size threshold, it is
+// cannot make room. If the buffer reaches the size threshold, it starts
-// drained and written to disk asynchronously.
+// a write of the reports in the background (see flushAtThreshold),
+// unless one it started is still waiting.
 func (b *Buffer) Append(v any) error {
 	line, err := json.Marshal(v)
 	if err != nil {
@@ -221,18 +245,9 @@ func (b *Buffer) Append(v any) error {
 	b.buf.Write(line)
 	b.buf.WriteByte('\n')
 
-	if b.buf.Len() >= flushSizeThreshold {
+	if b.buf.Len() >= flushSizeThreshold &&
-		data := b.drainBuf()
+		b.thresholdWriteWaiting.CompareAndSwap(false, true) {
-		b.mu.Unlock()
+		b.writers.Go(b.flushAtThreshold)
-
-		b.writers.Go(func() {
-			writeErr := b.writeFile(data)
-			if writeErr != nil {
-				b.log.Error("flush reports failed", "error", writeErr)
-			}
-		})
-
-		return nil
 	}
 
 	b.mu.Unlock()
@@ -290,21 +305,28 @@ func (b *Buffer) flushLoop() {
 	}
 }
 
-// flushLocked acquires the lock, drains the buffer, and
+// flushLocked waits for any write of a report file under way, then
-// writes the data to a compressed file.
+// writes the reports waiting in the buffer to a new one.
 func (b *Buffer) flushLocked() error {
-	b.mu.Lock()
+	b.writeMu.Lock()
+	defer b.writeMu.Unlock()
 
-	if b.buf.Len() == 0 {
+	return b.writeFile()
-		b.mu.Unlock()
+}
 
-		return nil
+// flushAtThreshold is the write Append starts when the buffer reaches
+// flushSizeThreshold. Once any write under way is done, it lets Append
+// start another and writes the reports waiting in the buffer.
+func (b *Buffer) flushAtThreshold() {
+	b.writeMu.Lock()
+	defer b.writeMu.Unlock()
+
+	b.thresholdWriteWaiting.Store(false)
+
+	err := b.writeFile()
+	if err != nil {
+		b.log.Error("flush reports failed", "error", err)
 	}
-
-	data := b.drainBuf()
-	b.mu.Unlock()
-
-	return b.writeFile(data)
 }
 
 // drainBuf copies the buffer contents and resets it.
@@ -317,10 +339,22 @@ func (b *Buffer) drainBuf() []byte {
 	return data
 }
 
-// writeFile writes data, reports drained from the buffer, to a new
+// writeFile takes the reports waiting in the buffer, if any, and writes
-// timestamped zstd-compressed JSONL file in the data directory. If the
+// them to a new timestamped zstd-compressed JSONL file in the data
-// write fails, it puts the reports back in the buffer.
+// directory. If the write fails, it puts the reports back in the
-func (b *Buffer) writeFile(data []byte) error {
+// buffer. The caller must hold b.writeMu.
+func (b *Buffer) writeFile() error {
+	b.mu.Lock()
+
+	if b.buf.Len() == 0 {
+		b.mu.Unlock()
+
+		return nil
+	}
+
+	data := b.drainBuf()
+	b.mu.Unlock()
+
 	// The timestamp comes first, so the names sort by time; the number
 	// after it tells apart files named in the same millisecond.
 	ts := b.now().UTC().Format("2006-01-02T15-04-05.000Z")
@@ -351,14 +385,7 @@ func (b *Buffer) writeFile(data []byte) error {
 	b.usedBytes -= int64(len(data))
 	b.usedBytes += size
 	b.filesBytes += size
-
+	b.files = append(b.files, reportFile{name: name, size: size})
-	// At its place by name, not at the end: another write, of a newer
-	// file, may have completed while this one was being written.
-	i, _ := slices.BinarySearchFunc(b.files, name,
-		func(f reportFile, target string) int {
-			return strings.Compare(f.name, target)
-		})
-	b.files = slices.Insert(b.files, i, reportFile{name: name, size: size})
 
 	return nil
 }

internal/reportbuf/reportbuf_test.go

@@ -108,6 +108,68 @@ func TestFailedFinalFlushFailsStop(t *testing.T) {
 	}
 }
 
+// TestUncreatableDataDirStopsStart sets DATA_DIR below a regular file,
+// where no directory can be created, even by root, as the tests run in
+// the backend image. The start must fail with an error naming DATA_DIR
+// and its value, as the errors for the other settings do.
+func TestUncreatableDataDirStopsStart(t *testing.T) {
+	file := filepath.Join(t.TempDir(), "file")
+	writeBytes(t, file, 1)
+
+	dir := filepath.Join(file, "reports")
+	t.Setenv("DATA_DIR", dir)
+
+	app := fxtest.New(t,
+		fx.Provide(
+			globals.New,
+			logger.New,
+			config.New,
+			reportbuf.New,
+		),
+		fx.Invoke(func(*reportbuf.Buffer) {}),
+	)
+
+	err := app.Start(t.Context())
+	if err == nil {
+		t.Fatal("start succeeded with DATA_DIR below a regular file")
+	}
+
+	want := fmt.Sprintf("DATA_DIR %q: ", dir)
+	if !strings.Contains(err.Error(), want) {
+		t.Fatalf("start error = %q, want it to contain %q", err, want)
+	}
+}
+
+// TestUnwritableDataDirStopsStart sets DATA_DIR to /proc, which exists
+// and can be read, but where no file can be created, even by root, as
+// the tests run in the backend image: root ignores a directory the test
+// makes read-only. The start must fail with an error naming DATA_DIR and
+// its value, rather than the first write of a report file.
+func TestUnwritableDataDirStopsStart(t *testing.T) {
+	dir := "/proc"
+	t.Setenv("DATA_DIR", dir)
+
+	app := fxtest.New(t,
+		fx.Provide(
+			globals.New,
+			logger.New,
+			config.New,
+			reportbuf.New,
+		),
+		fx.Invoke(func(*reportbuf.Buffer) {}),
+	)
+
+	err := app.Start(t.Context())
+	if err == nil {
+		t.Fatal("start succeeded with DATA_DIR /proc")
+	}
+
+	want := fmt.Sprintf("DATA_DIR %q: ", dir)
+	if !strings.Contains(err.Error(), want) {
+		t.Fatalf("start error = %q, want it to contain %q", err, want)
+	}
+}
+
 // TestStopWaitsForPeriodicWrite holds open the write the periodic flush
 // starts, then stops the app. The stop must wait until the write is
 // released, and the report must then be in its file. The test runs in
@@ -519,6 +581,62 @@ func TestWrittenFilesDeletedToMakeRoom(t *testing.T) {
 	}
 }
 
+// TestWrittenFilesDeletedOldestFirst writes three report files, then
+// appends a report that fits only once one of them is deleted. The
+// first one written must be the one deleted.
+func TestWrittenFilesDeletedOldestFirst(t *testing.T) {
+	const maxBytes = 1000
+
+	report := map[string]string{"id": "written"}
+
+	t.Setenv("DATA_DIR", t.TempDir())
+	t.Setenv("DATA_DIR_MAX_BYTES", strconv.Itoa(maxBytes))
+
+	buf := startBuffer(t)
+
+	var written []string
+
+	buf.OnFileCreated(func(f *os.File) { written = append(written, f.Name()) })
+
+	for range 3 {
+		err := buf.Append(report)
+		if err != nil {
+			t.Fatalf("append report: %v", err)
+		}
+
+		err = buf.Flush()
+		if err != nil {
+			t.Fatalf("flush: %v", err)
+		}
+	}
+
+	// The three files hold the same report, so they are the same size.
+	info, err := os.Stat(written[0])
+	if err != nil {
+		t.Fatalf("stat %s: %v", written[0], err)
+	}
+
+	// A report that fits beside two of the files, so deleting one makes
+	// exactly the room it needs.
+	pad := maxBytes - 2*int(info.Size()) -
+		lineBytes(t, map[string]string{"id": ""})
+
+	err = buf.Append(map[string]string{"id": strings.Repeat("a", pad)})
+	if err != nil {
+		t.Fatalf("report that needs room: %v", err)
+	}
+
+	if exists(t, written[0]) {
+		t.Fatal("first report file written kept when room was needed")
+	}
+
+	for _, path := range written[1:] {
+		if !exists(t, path) {
+			t.Fatalf("%s deleted before the first report file written", path)
+		}
+	}
+}
+
 // TestFailedDeletionStillCounts makes deleting the oldest report file
 // fail. It is still there, so it still takes room, and the next oldest
 // is deleted in its place.
@@ -604,17 +722,16 @@ func TestFileDeletedByHandFreesRoom(t *testing.T) {
 	}
 }
 
-// TestFileBeingWrittenIsNeverDeleted holds the write of one report file
+// TestFileBeingWrittenIsNeverDeleted holds the write of a report file
-// open while a second write completes, then sends a report that needs
+// open under a cap with room for two reports, then appends a second
-// room. Deleting either file would make it, and the one being written is
+// report and a third that needs room. The file being written is the
-// the older, but only the complete one may be deleted. Once the first
+// only report file, and it is not deleted to make room, so the third
-// write is complete, its file is deleted when room is needed.
+// report is refused. Once the write is complete, its file is deleted
+// when room is needed.
 func TestFileBeingWrittenIsNeverDeleted(t *testing.T) {
 	report := map[string]string{"id": "writing"}
 
 	t.Setenv("DATA_DIR", t.TempDir())
-	// Room for two reports waiting to be written, but not for two
-	// beside a report file.
 	t.Setenv("DATA_DIR_MAX_BYTES", strconv.Itoa(2*lineBytes(t, report)))
 
 	buf := startBuffer(t)
@@ -639,36 +756,25 @@ func TestFileBeingWrittenIsNeverDeleted(t *testing.T) {
 
 	writing := <-created
 
-	// Only the first write is held; the second goes through, and
+	// Errorf, not Fatalf, until the write is released, so that a
-	// so do the writes after it, the final one at stop included.
-	var complete string
-
-	buf.OnFileCreated(func(f *os.File) { complete = f.Name() })
-
-	// Errorf, not Fatalf, until the first write is released, so that a
 	// failure here does not leave it held.
 	err = buf.Append(report)
 	if err != nil {
 		t.Errorf("second report: %v", err)
 	}
 
-	err = buf.Flush()
-	if err != nil {
-		t.Errorf("flush of the second report: %v", err)
-	}
-
 	err = buf.Append(report)
-	if err != nil {
+	if !errors.Is(err, reportbuf.ErrFull) {
-		t.Errorf("report that needs room: %v", err)
+		t.Errorf("report that needs room: error = %v, want ErrFull", err)
 	}
 
 	if !exists(t, writing) {
 		t.Error("report file deleted while it was being written")
 	}
 
-	if exists(t, complete) {
+	// Only this write is held; the writes after it, the final one at
-		t.Error("complete report file kept, though room was needed")
+	// stop included, go through.
-	}
+	buf.OnFileCreated(func(*os.File) {})
 
 	close(release)
 
@@ -677,9 +783,11 @@ func TestFileBeingWrittenIsNeverDeleted(t *testing.T) {
 		t.Fatalf("flush of the first report: %v", err)
 	}
 
+	// The second report still waits to be written, so this one fits
+	// only once the file just written is deleted.
 	err = buf.Append(report)
 	if err != nil {
-		t.Fatalf("report after the first write was complete: %v", err)
+		t.Fatalf("report after the write was complete: %v", err)
 	}
 
 	if exists(t, writing) {
@@ -821,6 +929,82 @@ func TestFailedWriteKeepsReportsAppendedDuringIt(t *testing.T) {
 	}
 }
 
+// TestFailedWritesKeepReportsInOrder holds a write open, then appends a
+// report that takes the buffer to FlushSizeThreshold, so that Append
+// starts a second write. The held write fails first, then the second,
+// which waited for it and took both reports. The next flush writes them
+// in the order they were appended, the held write's first.
+func TestFailedWritesKeepReportsInOrder(t *testing.T) {
+	dir := t.TempDir()
+	t.Setenv("DATA_DIR", dir)
+
+	buf := startBuffer(t)
+
+	created := make(chan struct{})
+	release := make(chan struct{})
+
+	// Each write is held until release is closed, then fails, as its
+	// file is closed under it.
+	buf.OnFileCreated(func(f *os.File) {
+		created <- struct{}{}
+
+		<-release
+
+		_ = f.Close()
+	})
+
+	err := buf.Append(map[string]string{"id": "held"})
+	if err != nil {
+		t.Fatalf("report of the held write: %v", err)
+	}
+
+	flushed := make(chan error)
+
+	go func() { flushed <- buf.Flush() }()
+
+	<-created
+
+	big := strings.Repeat("a", reportbuf.FlushSizeThreshold)
+
+	// Errorf, not Fatalf, until the held write is released, so that a
+	// failure here does not leave it held.
+	err = buf.Append(map[string]string{"id": big})
+	if err != nil {
+		t.Errorf("report that starts the second write: %v", err)
+	}
+
+	close(release)
+
+	err = <-flushed
+	if err == nil {
+		t.Error("held write succeeded, though its file was closed under it")
+	}
+
+	// The second write has created its file, so it has taken the
+	// reports.
+	<-created
+
+	// Writes from here on, the final one at stop included, succeed.
+	buf.OnFileCreated(func(*os.File) {})
+
+	// Flush waits for the second write to fail before it writes.
+	err = buf.Flush()
+	if err != nil {
+		t.Fatalf("flush after the failed writes: %v", err)
+	}
+
+	files, err := readReportFiles(dir)
+	if err != nil {
+		t.Fatalf("read report files: %v", err)
+	}
+
+	want := `{"id":"held"}` + "\n" + `{"id":"` + big + `"}` + "\n"
+	if len(files) != 1 || files[0] != want {
+		t.Fatalf("%d report files, want one holding the held write's "+
+			"report, then the second one", len(files))
+	}
+}
+
 // TestReportsRefusedWhileWritesFail posts reports through the report
 // handler, as the page does. Once a write has failed, a report is
 // refused with 500 until a write succeeds again; the reports taken
@@ -1088,6 +1272,106 @@ func TestFlushAtSizeThreshold(t *testing.T) {
 	}
 }
 
+// TestEachSizeThresholdStartsAWrite reaches FlushSizeThreshold twice,
+// with no call to Flush and the periodic flush a minute away. Append
+// must start a write each time, so that the reports end up in two
+// report files. The test runs in a synctest bubble, so that
+// synctest.Wait returns once each write is done.
+func TestEachSizeThresholdStartsAWrite(t *testing.T) {
+	dir := t.TempDir()
+	t.Setenv("DATA_DIR", dir)
+
+	synctest.Test(t, func(t *testing.T) {
+		buf := startBuffer(t)
+
+		var appended string
+
+		for range 2 {
+			appended = appendToThreshold(t, buf)
+
+			synctest.Wait()
+		}
+
+		files, err := readReportFiles(dir)
+		if err != nil {
+			t.Fatalf("read report files: %v", err)
+		}
+
+		if !slices.Equal(files, []string{appended, appended}) {
+			t.Fatalf("%d report files, want two, each holding the %d "+
+				"bytes appended to reach the threshold",
+				len(files), len(appended))
+		}
+	})
+}
+
+// TestAppendStartsOneWaitingWrite holds a write open, then appends two
+// reports of FlushSizeThreshold bytes each. Append starts a write at
+// the first, which waits for the held one, and none at the second,
+// since that write is still waiting. Every write fails, so a write
+// started at the second report would find the reports back in the
+// buffer and try them again: once the held write is released, two
+// writes must have been started, the held one and one more.
+func TestAppendStartsOneWaitingWrite(t *testing.T) {
+	t.Setenv("DATA_DIR", t.TempDir())
+
+	synctest.Test(t, func(t *testing.T) {
+		buf := startBuffer(t)
+
+		var writes atomic.Int64
+
+		release := make(chan struct{})
+
+		// Each write is held until release is closed, then fails, as its
+		// file is closed under it.
+		buf.OnFileCreated(func(f *os.File) {
+			writes.Add(1)
+
+			<-release
+
+			_ = f.Close()
+		})
+
+		err := buf.Append(map[string]string{"id": "one-waiting"})
+		if err != nil {
+			t.Fatalf("report of the held write: %v", err)
+		}
+
+		go func() { _ = buf.Flush() }()
+
+		// Wait returns once the flush's write is held.
+		synctest.Wait()
+
+		big := map[string]string{
+			"id": strings.Repeat("a", reportbuf.FlushSizeThreshold),
+		}
+
+		// Errorf, not Fatalf, until the held write is released, so that a
+		// failure here does not leave it held.
+		for range 2 {
+			err = buf.Append(big)
+			if err != nil {
+				t.Errorf("report of FlushSizeThreshold bytes: %v", err)
+			}
+		}
+
+		close(release)
+
+		// Wait returns once every write started is done.
+		synctest.Wait()
+
+		started := writes.Load()
+
+		// Writes from here on, the final one at stop included, succeed.
+		buf.OnFileCreated(func(*os.File) {})
+
+		if started != 2 {
+			t.Fatalf("%d writes started, want 2: the held one and one "+
+				"at the threshold", started)
+		}
+	})
+}
+
 // appendToThreshold appends reports to buf until it holds
 // FlushSizeThreshold bytes, so that the last append starts writing them
 // to a report file, and returns the lines appended.
@@ -1233,85 +1517,3 @@ func hasReportFile(t *testing.T, dir string) bool {
 
 	return false
 }
-
-// TestFilesDeletedOldestFirstWhenWritesOverlap holds the write of an
-// older report file open until a newer one's write completes, then
-// releases it. When room is needed, the older file is deleted first,
-// though its write was the last to complete.
-func TestFilesDeletedOldestFirstWhenWritesOverlap(t *testing.T) {
-	const maxBytes = 1000
-
-	report := map[string]string{"id": "overlap"}
-
-	t.Setenv("DATA_DIR", t.TempDir())
-	t.Setenv("DATA_DIR_MAX_BYTES", strconv.Itoa(maxBytes))
-
-	buf := startBuffer(t)
-
-	created := make(chan string)
-	release := make(chan struct{})
-
-	buf.OnFileCreated(func(f *os.File) {
-		created <- f.Name()
-
-		<-release
-	})
-
-	err := buf.Append(report)
-	if err != nil {
-		t.Fatalf("older report: %v", err)
-	}
-
-	flushed := make(chan error)
-
-	go func() { flushed <- buf.Flush() }()
-
-	older := <-created
-
-	// Only the older write is held; the newer one goes through, and
-	// so do the writes after it, the final one at stop included.
-	var newer string
-
-	buf.OnFileCreated(func(f *os.File) { newer = f.Name() })
-
-	// Errorf, not Fatalf, until the older write is released, so that a
-	// failure here does not leave it held.
-	err = buf.Append(report)
-	if err != nil {
-		t.Errorf("newer report: %v", err)
-	}
-
-	err = buf.Flush()
-	if err != nil {
-		t.Errorf("flush of the newer report: %v", err)
-	}
-
-	close(release)
-
-	err = <-flushed
-	if err != nil {
-		t.Fatalf("flush of the older report: %v", err)
-	}
-
-	info, err := os.Stat(newer)
-	if err != nil {
-		t.Fatalf("stat %s: %v", newer, err)
-	}
-
-	// A report that fits beside the newer file alone, so deleting the
-	// older one makes exactly the room it needs.
-	pad := maxBytes - int(info.Size()) - lineBytes(t, map[string]string{"id": ""})
-
-	err = buf.Append(map[string]string{"id": strings.Repeat("a", pad)})
-	if err != nil {
-		t.Fatalf("report that needs room: %v", err)
-	}
-
-	if exists(t, older) {
-		t.Fatal("older report file kept when room was needed")
-	}
-
-	if !exists(t, newer) {
-		t.Fatal("newer report file deleted before the older one")
-	}
-}

script/cibuild

@@ -2,9 +2,12 @@
 # script/cibuild: run the CI build. It bootstraps first: a CI runner
 # checks out and runs this and nothing else, and script/fmt-check runs
 # the formatter on the host, which a pristine checkout cannot do.
-# --no-cache for the same reason as script/docker: the gate phases the
+# A new CACHEBUST value reruns every RUN step after the ARG CACHEBUST
-# final stage depends on are RUN steps, and a cached one is a check that
+# line, since a cached gate phase is a check that did not run; unlike
-# did not run.
+# --no-cache, it keeps the layers above that line and the cache mounts.
+# This relies on every stage of the Dockerfile declaring ARG CACHEBUST
+# before every step that lints, tests or compiles; a stage without that
+# line is served from cache on an unchanged tree, with no warning.
 set -eu
 
 SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd -P)"
@@ -14,15 +17,17 @@ main() {
     cd "$ROOT"
     "$SCRIPT_DIR/bootstrap"
     "$SCRIPT_DIR/check"
-    # Own line: a failing command substitution inside an argument does
+    # The version and the tag each get their own line: a failing
-    # not trip `set -e`, so the inline form degrades silently to an
+    # command substitution inside an argument does not trip `set -e`,
-    # empty constant. The VERSION build argument takes precedence over
+    # so the inline form degrades silently to an empty constant. The
-    # the version a build stage derives from the .git in the context.
+    # VERSION build argument takes precedence over the version a build
+    # stage derives from the .git in the context.
     version="$(git describe --tags --always --dirty 2>/dev/null || true)"
     [ -n "$version" ] || version="unknown"
-    docker build --no-cache \
+    tag="$("$SCRIPT_DIR/projectname")"
+    docker build --build-arg CACHEBUST="$(date +%s)-$$" \
         --build-arg VERSION="$version" \
-        -t "$("$SCRIPT_DIR/projectname")" .
+        -t "$tag" .
 }
 
 main "$@"

script/docker

@@ -1,8 +1,12 @@
 #!/bin/sh
 # script/docker: build the Docker image tagged with the project name.
 # Identical in all repos; the tag comes from script/projectname.
-# --no-cache because the gate phases the final stage depends on are RUN
+# A new CACHEBUST value reruns every RUN step after the ARG CACHEBUST
-# steps, and a cached one is a check that did not run.
+# line, since a cached gate phase is a check that did not run; unlike
+# --no-cache, it keeps the layers above that line and the cache mounts.
+# This relies on every stage of the Dockerfile declaring ARG CACHEBUST
+# before every step that lints, tests or compiles; a stage without that
+# line is served from cache on an unchanged tree, with no warning.
 set -eu
 
 SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd -P)"
@@ -10,15 +14,17 @@ ROOT="$(cd "$SCRIPT_DIR/.." && pwd -P)"
 
 main() {
     cd "$ROOT"
-    # Own line: a failing command substitution inside an argument does
+    # The version and the tag each get their own line: a failing
-    # not trip `set -e`, so the inline form degrades silently to an
+    # command substitution inside an argument does not trip `set -e`,
-    # empty constant. The VERSION build argument takes precedence over
+    # so the inline form degrades silently to an empty constant. The
-    # the version a build stage derives from the .git in the context.
+    # VERSION build argument takes precedence over the version a build
+    # stage derives from the .git in the context.
     version="$(git describe --tags --always --dirty 2>/dev/null || true)"
     [ -n "$version" ] || version="unknown"
-    docker build --no-cache \
+    tag="$("$SCRIPT_DIR/projectname")"
+    docker build --build-arg CACHEBUST="$(date +%s)-$$" \
         --build-arg VERSION="$version" \
-        -t "$("$SCRIPT_DIR/projectname")" .
+        -t "$tag" .
 }
 
 main "$@"

script/frontend-viewport-test

@@ -7,6 +7,8 @@
 # headless Chrome at it over CDP, and asserts on computed layout at every
 # viewport width derived from the app's own CSS. See
 # test/viewport/README.md for what this covers and what it cannot.
+# Before the browser starts, it checks that the page is served with
+# Cache-Control: no-cache and Content-Security-Policy.
 #
 # Deliberately not part of script/check: it needs Docker and takes far
 # longer than the 20s budget make test has to stay inside.
@@ -75,6 +77,32 @@ main() {
         -v "$ROOT/configs/security-headers.conf:/etc/nginx/security-headers.conf:ro" \
         "$SERVER_IMAGE" > /dev/null
 
+    # The page, and /no-such-page, which location / answers with the
+    # page through its try_files fallback, must go out with
+    # Cache-Control: no-cache. That add_header in its location drops the
+    # server's own unless the location includes the security headers
+    # again; only Content-Security-Policy is checked, as it comes from
+    # the same file as the rest. curl tries up to ten more times, a
+    # second apart, while nginx starts. It runs without -S, which would
+    # print an error for every failed attempt, even when a later one
+    # succeeds.
+    for path in / /no-such-page; do
+        if ! headers="$(docker exec "$SERVER" curl -s -o /dev/null -D - \
+            --retry 10 --retry-connrefused --retry-delay 1 \
+            "http://127.0.0.1:8080$path")"; then
+            echo "frontend-viewport-test: nginx did not answer $path" >&2
+            exit 1
+        fi
+        headers="$(printf '%s\n' "$headers" | tr -d '\r')"
+        if ! printf '%s\n' "$headers" | grep -qix 'Cache-Control: no-cache' ||
+            ! printf '%s\n' "$headers" | grep -qi '^Content-Security-Policy:'; then
+            echo "frontend-viewport-test: $path needs" \
+                "Cache-Control: no-cache and Content-Security-Policy; got:" >&2
+            printf '%s\n' "$headers" >&2
+            exit 1
+        fi
+    done
+
     # The image's own entrypoint already exposes CDP on 9222 and passes
     # --no-sandbox, so only extra flags belong here; re-specifying the
     # debugging port collides with it and leaves the endpoint bound to

script/lint

@@ -5,9 +5,14 @@
 # make its answer untrustworthy.
 #
 # The phase is not the last stage in the file, so it is built only when
-# --target names it. --no-cache because a cached lint layer is a lint
+# --target names it. A new CACHEBUST value reruns every RUN step after
-# that did not run. The tag makes each build replace the previous image
+# the ARG CACHEBUST line, since a cached lint layer is a lint that did
-# instead of leaving a dangling one behind.
+# not run; unlike --no-cache, it keeps the layers above that line and
+# the cache mounts. This relies on every stage of the Dockerfile
+# declaring ARG CACHEBUST before every step that lints, tests or
+# compiles; a stage without that line is served from cache on an
+# unchanged tree, with no warning.
+# --output type=cacheonly writes no image, since nothing uses one.
 set -eu
 
 SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd -P)"
@@ -15,9 +20,9 @@ ROOT="$(cd "$SCRIPT_DIR/.." && pwd -P)"
 
 main() {
     cd "$ROOT"
-    docker build --no-cache \
+    docker build --build-arg CACHEBUST="$(date +%s)-$$" \
         --target lint \
-        -t "$("$SCRIPT_DIR/projectname")-lint" .
+        --output type=cacheonly .
 }
 
 main "$@"

script/test

@@ -2,8 +2,13 @@
 # script/test: run the test suite. Testing is a phase of the Dockerfile
 # and this builds that phase alone, on the same terms as script/lint:
 # --target because a phase that is not the last stage is built only when
-# named, --no-cache because a cached test layer is a test that did not
+# named, and a new CACHEBUST value because a cached test layer is a test
-# run, and a tag so each build replaces the previous image.
+# that did not run; unlike --no-cache, it keeps the layers above the ARG
+# CACHEBUST line and the cache mounts. This relies on every stage of the
+# Dockerfile declaring ARG CACHEBUST before every step that lints, tests
+# or compiles; a stage without that line is served from cache on an
+# unchanged tree, with no warning. --output type=cacheonly writes no
+# image, since nothing uses one.
 set -eu
 
 SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd -P)"
@@ -11,9 +16,9 @@ ROOT="$(cd "$SCRIPT_DIR/.." && pwd -P)"
 
 main() {
     cd "$ROOT"
-    docker build --no-cache \
+    docker build --build-arg CACHEBUST="$(date +%s)-$$" \
         --target test \
-        -t "$("$SCRIPT_DIR/projectname")-test" .
+        --output type=cacheonly .
 }
 
 main "$@"

test/unit/main.test.js

@@ -6,6 +6,7 @@ import { beforeEach, test } from "node:test";
 import assert from "node:assert/strict";
 import {
     AppState,
+    buildUI,
     CONFIG,
     greyOutUI,
     hostRowHTML,
@@ -17,6 +18,7 @@ import {
     measureLatency,
     Reporter,
     tick,
+    updateClocks,
     updateSummary,
 } from "../../web/src/main.js";
 
@@ -28,8 +30,10 @@ import {
 // selector until the next test starts. As on a page, writing its text
 // replaces its markup; the status dot greyOutUI looks for in it is not
 // there. Drawing a sparkline does nothing; it looks for the pixel ratio on
-// window and finds none. In each test, console.error and console.info
+// window and finds none. The commit the page's footer names, which
-// print nothing and keep what they are given.
+// vite.config.js fills in when it builds the page, is blank. In each test,
+// console.error and console.info print nothing and keep what they are
+// given.
 const doNothing = () => {};
 let elements;
 beforeEach((t) => {
@@ -51,6 +55,8 @@ const canvasContext = {
 globalThis.window = {};
 globalThis.requestAnimationFrame = doNothing;
 globalThis.cancelAnimationFrame = doNothing;
+globalThis.__COMMIT_HASH__ = "";
+globalThis.__COMMIT_FULL__ = "";
 globalThis.document = {
     getElementById: () => null,
     createElement: () => ({}),
@@ -77,6 +83,11 @@ function statusText(state, host) {
     return elements[`.status-text[data-host="${index}"]`]?.innerHTML;
 }
 
+// What this test wrote to the browser console with console[method].
+function consoleLines(method) {
+    return console[method].mock.calls.map((call) => call.arguments[0]);
+}
+
 // Mocks the clock for test t, so that a check lasting seconds takes no real
 // time, and replaces fetch with targets that each answer after
 // answerAfter(url) milliseconds of that clock, or never when that is
@@ -196,9 +207,10 @@ test("at a 30000ms interval, a target answering after 1000ms shows in its row wh
 });
 
 // In the next three tests, the answering target's check is still waiting
-// when something happens after which its result must not show.
+// when something happens after which its result must not be recorded or
+// show in its row, and the check must write nothing to the console.
 
-test("at a 30000ms interval, a check still waiting when its round is given up does not show in its row", async (t) => {
+test("at a 30000ms interval, a check still waiting when its round is given up is not recorded and does not show in its row", async (t) => {
     CONFIG.updateInterval = 30000;
     const state = new AppState([
         { name: "Answering", url: "https://answering.test" },
@@ -214,9 +226,13 @@ test("at a 30000ms interval, a check still waiting when its round is given up do
     roundChecks.abort();
     assert.notEqual(await settled(round), "still waiting");
     assert.equal(latencyFigure(state, answering), undefined);
+    assert.deepEqual(answering.history, []);
+    assert.equal(answering.status, "pending");
+    assert.deepEqual(consoleLines("error"), []);
+    assert.deepEqual(consoleLines("info"), []);
 });
 
-test("at a 30000ms interval, a check still waiting when the user pauses does not show in its row", async (t) => {
+test("at a 30000ms interval, a check still waiting when the user pauses is not recorded and does not show in its row", async (t) => {
     CONFIG.updateInterval = 30000;
     const state = new AppState([
         { name: "Answering", url: "https://answering.test" },
@@ -231,9 +247,13 @@ test("at a 30000ms interval, a check still waiting when the user pauses does not
     t.mock.timers.tick(500);
     assert.equal(await settled(round), "still waiting");
     assert.equal(latencyFigure(state, answering), undefined);
+    assert.deepEqual(answering.history, []);
+    assert.equal(answering.status, "pending");
+    assert.deepEqual(consoleLines("error"), []);
+    assert.deepEqual(consoleLines("info"), []);
 });
 
-test("at a 30000ms interval, a check in the first round does not show in its row", async (t) => {
+test("at a 30000ms interval, a check in the first round is not recorded and does not show in its row", async (t) => {
     CONFIG.updateInterval = 30000;
     const state = new AppState([
         { name: "Answering", url: "https://answering.test" },
@@ -245,6 +265,10 @@ test("at a 30000ms interval, a check in the first round does not show in its row
     t.mock.timers.tick(1000);
     assert.equal(await settled(round), "still waiting");
     assert.equal(latencyFigure(state, answering), undefined);
+    assert.deepEqual(answering.history, []);
+    assert.equal(answering.status, "pending");
+    assert.deepEqual(consoleLines("error"), []);
+    assert.deepEqual(consoleLines("info"), []);
 });
 
 test("at a 30000ms interval, after the user pauses and resumes during a round, no row reads paused once its last check ends", async (t) => {
@@ -294,11 +318,6 @@ function stateWithOneTarget() {
     return { state, target: state.local[0] };
 }
 
-// What this test wrote to the browser console with console[method].
-function consoleLines(method) {
-    return console[method].mock.calls.map((call) => call.arguments[0]);
-}
-
 test("a check that fails with a network error writes one console line with the target, the time, the error and how long the request took", async (t) => {
     const { state, target } = stateWithOneTarget();
     mockTargets(t, () => 23, target.url);
@@ -579,6 +598,32 @@ test("a check that timed out or found its target unreachable has a grey figure a
     assert.equal(latencyHex(null), "#6b7280");
 });
 
+// buildUI draws the page into #app, here a plain object. Each entry in the
+// footer's color legend is a colored dot and the latencies it stands for,
+// with < and > written as &lt; and &gt;. A latency takes the color of the
+// first entry in CONFIG.latencyColors whose limit it is below, so the
+// legend has, for each of those entries in order, a dot of its color (a
+// bg- class where CONFIG.latencyColors has a text- one) and the latencies
+// below its limit, or, for the last, which has no limit, the latencies
+// from the limit before it up. A gray dot for offline comes after them.
+test("the footer's color legend has each entry in CONFIG.latencyColors, with its color and limit, then offline", (t) => {
+    const app = {};
+    t.mock.method(globalThis.document, "getElementById", () => app);
+    buildUI(new AppState([]));
+
+    const legend = [
+        ...app.innerHTML.matchAll(/(bg-\w+-500) [^"]*"><\/span>(\S+)/g),
+    ].map(([, dot, latencies]) => [dot, latencies]);
+    const colors = CONFIG.latencyColors;
+    const expected = colors.map((color, i) => [
+        color.className.replace("text-", "bg-"),
+        color.below === Infinity
+            ? `&gt;=${colors[i - 1].below}ms`
+            : `&lt;${color.below}ms`,
+    ]);
+    assert.deepEqual(legend, [...expected, ["bg-gray-500", "offline"]]);
+});
+
 // A target whose checks, in turn, answered after each of latencies ms, or,
 // for null, found it unreachable.
 function hostAfter(latencies) {
@@ -701,6 +746,56 @@ test("once every WAN target's check has failed, the summary's Reachable slot rea
     assert.equal(elements["#summary-reachable"].className, "text-red-400");
 });
 
+// In the next two tests each of the header's clocks is a plain object, made
+// the first time updateClocks looks it up and kept in elements under "#" and
+// its id.
+
+// The mocked clock reads 1970-01-01T00:00:00.000Z.
+test("the header's clocks show the local time with its time zone's name, and the UTC time", (t) => {
+    t.mock.method(
+        globalThis.document,
+        "getElementById",
+        (id) => (elements[`#${id}`] ??= {}),
+    );
+    t.mock.timers.enable({ apis: ["Date"] });
+
+    updateClocks();
+
+    assert.match(
+        elements["#clock-local"].textContent,
+        /^Local: \d{4}-\d\d-\d\d \d\d:\d\d:\d\d \S/,
+    );
+    assert.equal(
+        elements["#clock-utc"].textContent,
+        "UTC: 1970-01-01 00:00:00 UTC",
+    );
+});
+
+// Here Intl.DateTimeFormat is as in Chrome before 95, Firefox before 91 or
+// Safari before 15.4: it throws a RangeError when asked for the time zone's
+// "shortGeneric" name.
+test("in a browser that cannot name the time zone, updating the clocks throws nothing and leaves them blank", (t) => {
+    t.mock.method(
+        globalThis.document,
+        "getElementById",
+        (id) => (elements[`#${id}`] ??= {}),
+    );
+    const DateTimeFormat = Intl.DateTimeFormat;
+    t.mock.method(Intl, "DateTimeFormat", function (locales, options) {
+        if (options?.timeZoneName === "shortGeneric") {
+            throw new RangeError(
+                "Value shortGeneric out of range for Intl.DateTimeFormat options property timeZoneName",
+            );
+        }
+        return new DateTimeFormat(locales, options);
+    });
+
+    updateClocks();
+
+    assert.equal(elements["#clock-local"].textContent, "");
+    assert.equal(elements["#clock-utc"].textContent, "");
+});
+
 // An app state in which, of the WAN targets, the first timedOut timed out,
 // the next unreachable were found unreachable, the next answered answered
 // after latency ms, and the rest have not been checked yet.

web/src/main.js

@@ -829,7 +829,7 @@ function wanRowsHTML(wanHosts) {
     return rows.join("");
 }
 
-function buildUI(state) {
+export function buildUI(state) {
     const app = document.getElementById("app");
     app.innerHTML = `
     <div class="mx-auto px-[5%] py-8">
@@ -911,7 +911,7 @@ function buildUI(state) {
           <span class="inline-block w-3 h-3 rounded-full bg-lime-500 mr-1 ml-3 align-middle"></span>&lt;100ms
           <span class="inline-block w-3 h-3 rounded-full bg-yellow-500 mr-1 ml-3 align-middle"></span>&lt;200ms
           <span class="inline-block w-3 h-3 rounded-full bg-orange-500 mr-1 ml-3 align-middle"></span>&lt;500ms
-          <span class="inline-block w-3 h-3 rounded-full bg-red-500 mr-1 ml-3 align-middle"></span>&gt;500ms
+          <span class="inline-block w-3 h-3 rounded-full bg-red-500 mr-1 ml-3 align-middle"></span>&gt;=500ms
           <span class="inline-block w-3 h-3 rounded-full bg-gray-500 mr-1 ml-3 align-middle"></span>offline
         </p>
         <p class="mt-2"><a href="https://git.eeqj.de/sneak/netwatch/commit/${__COMMIT_FULL__}" target="_blank" rel="noopener" class="text-gray-600 hover:text-gray-400">${__COMMIT_HASH__}</a></p>
@@ -1114,6 +1114,33 @@ function updateHealthBox(state) {
     else if (health === "healthy") log.notice(`Health: ${labels[health]}`);
 }
 
+// Writes the local time, with its time zone's name, and the UTC time into
+// the header's clocks. The clocks only decorate the page: in a browser
+// whose Intl.DateTimeFormat rejects the "shortGeneric" time zone name
+// (Chrome before 95, Firefox before 91, Safari before 15.4) they are left
+// blank.
+export function updateClocks() {
+    let local = "";
+    let utc = "";
+    try {
+        const now = new Date();
+        const zone = new Intl.DateTimeFormat("en-US", {
+            timeZoneName: "shortGeneric",
+        })
+            .formatToParts(now)
+            .find((p) => p.type === "timeZoneName").value;
+        const pad = (n) => String(n).padStart(2, "0");
+        local = `Local: ${now.toLocaleString("sv-SE", { hour12: false })} ${zone}`;
+        utc =
+            `UTC: ${now.getUTCFullYear()}-${pad(now.getUTCMonth() + 1)}-${pad(now.getUTCDate())} ` +
+            `${pad(now.getUTCHours())}:${pad(now.getUTCMinutes())}:${pad(now.getUTCSeconds())} UTC`;
+    } catch {
+        // Both clocks are left blank.
+    }
+    document.getElementById("clock-local").textContent = local;
+    document.getElementById("clock-utc").textContent = utc;
+}
+
 // --- Debug Log Renderer ------------------------------------------------------
 
 // One line of the debug log panel, its message set as text.
@@ -1509,24 +1536,6 @@ async function init() {
         sortAndRebuildWAN(state);
     });
 
-    function updateClocks() {
-        const now = new Date();
-        const pad = (n) => String(n).padStart(2, "0");
-        const utc =
-            `${now.getUTCFullYear()}-${pad(now.getUTCMonth() + 1)}-${pad(now.getUTCDate())} ` +
-            `${pad(now.getUTCHours())}:${pad(now.getUTCMinutes())}:${pad(now.getUTCSeconds())} UTC`;
-        const local =
-            now.toLocaleString("sv-SE", { hour12: false }) +
-            " " +
-            new Intl.DateTimeFormat("en-US", { timeZoneName: "shortGeneric" })
-                .formatToParts(now)
-                .find((p) => p.type === "timeZoneName").value;
-        document.getElementById("clock-local").textContent = "Local: " + local;
-        document.getElementById("clock-utc").textContent = "UTC: " + utc;
-    }
-    updateClocks();
-    setInterval(updateClocks, 1000);
-
     // Rounds never overlap: a round first gives up the last round's checks
     // if they are still waiting, and the last round then records nothing
     // more. At a steady interval they never are, as they time out at 80% of
@@ -1606,6 +1615,11 @@ async function init() {
 
     window.addEventListener("resize", () => handleResize(state));
     setTimeout(() => handleResize(state), CONFIG.resizeDelay);
+
+    // The clocks start last, after the rounds: they only decorate the page,
+    // so nothing in them may stop the monitor from probing.
+    updateClocks();
+    setInterval(updateClocks, 1000);
 }
 
 // Bootstrap only when loaded as the page: a real DOM containing the #app