deploy (#365)

Reviewed-on: #365

Dockerfile

@@ -88,7 +88,9 @@ RUN CGO_ENABLED=1 make build VERSION="$VERSION" GO_LDFLAGS='-extldflags "-static
 # alpine:3.21, 2026-03-17
 FROM alpine:3.21@sha256:c3f8e73fdb79deaebaa2037150150191b9dcbfba68b4a46d70103204c53f4709
 
-RUN apk --no-cache add ca-certificates
+# su-exec 0.2-r3 (Alpine 3.21), 2026-09-29: the entrypoint runs the app
+# as webhooker with it.
+RUN apk --no-cache add ca-certificates su-exec=0.2-r3
 
 # Create non-root user
 RUN addgroup -g 1000 -S webhooker && \
@@ -99,13 +101,17 @@ WORKDIR /app
 # Copy binary from builder
 COPY --from=builder /build/bin/webhooker /app/webhooker
 
+# Not under /app, which belongs to webhooker: this script runs as root.
+COPY deploy/docker-entrypoint.sh /usr/local/bin/docker-entrypoint.sh
+
 # Create data directory for all SQLite databases (main app DB +
 # per-webhook event DBs). DATA_DIR defaults to /var/lib/webhooker.
 RUN mkdir -p /var/lib/webhooker
 
 RUN chown -R webhooker:webhooker /app /var/lib/webhooker
 
-USER webhooker
+# No USER: the entrypoint starts as root to make the data directory
+# webhooker's, then runs the app as webhooker.
 
 EXPOSE 8080
 
@@ -124,4 +130,5 @@ ENV BIND_ADDRESS=0.0.0.0
 HEALTHCHECK --interval=30s --timeout=3s --start-period=5s --retries=3 \
     CMD wget --no-verbose --tries=1 --spider http://localhost:8080/.well-known/healthcheck || exit 1
 
+ENTRYPOINT ["/usr/local/bin/docker-entrypoint.sh"]
 CMD ["/app/webhooker"]

README.md

@@ -7,6 +7,13 @@ services, durably stores them, and delivers them to configured targets
 with retry support, logging, and observability. Category: infrastructure
 / web service. License: MIT.
 
+Each entrypoint is a version 4 UUID served at `/webhook/{uuid}`, and
+that UUID is the entrypoint's only credential. webhooker does not use
+shared secrets, HMAC signatures or token headers on the receiver, and
+will not add them — read
+[The entrypoint URL is the authentication secret](#the-entrypoint-url-is-the-authentication-secret)
+before deploying one.
+
 ## Getting Started
 
 ### Prerequisites
@@ -37,9 +44,9 @@ make bootstrap
 # Run all checks (test, lint, format check)
 make check
 
-# Run in development mode. DATA_DIR defaults to /var/lib/webhooker in
+# Run the server from the clone. DATA_DIR defaults to
-# every environment, so set it (in .env or the shell) to a writable
+# /var/lib/webhooker in every environment, so set it (in .env or the
-# directory when running from a clone.
+# shell) to a writable directory.
 DATA_DIR=./data make dev
 
 # Build Docker image
@@ -85,7 +92,8 @@ them at once. A variable already present in the real environment wins
 over the file's value for the same name.
 
 The environment is selected by setting `WEBHOOKER_ENVIRONMENT` to `dev`
-or `prod` (default: `dev`). The setting controls exactly one behavior:
+or `prod` (default: `prod`; `dev` must be set explicitly). The setting
+controls exactly one behavior:
 
 | Behavior | `dev`                   | `prod`           |
 | -------- | ----------------------- | ---------------- |
@@ -127,7 +135,7 @@ TTY detection, and security headers are always applied.
 
 | Variable                | Description                         | Default  |
 | ----------------------- | ----------------------------------- | -------- |
-| `WEBHOOKER_ENVIRONMENT` | `dev` or `prod`                     | `dev`    |
+| `WEBHOOKER_ENVIRONMENT` | `dev` or `prod`                     | `prod`   |
 | `PORT`                  | HTTP listen port                    | `8080`   |
 | `BIND_ADDRESS`          | IP address the HTTP listener binds. Loopback by default, so the cleartext listener is not published on every interface. The Docker image ships `0.0.0.0` instead. See [Bind address](#bind-address) | `127.0.0.1` (image: `0.0.0.0`) |
 | `DATA_DIR`              | Directory for all SQLite databases  | `/var/lib/webhooker` |
@@ -149,6 +157,11 @@ private and reserved ranges — RFC 1918, loopback, CGNAT, link-local and
 the rest — are refused, which stops a target from being used to make
 webhooker probe the network it sits in.
 
+Besides the private and reserved ranges, the default blocklist refuses
+public cloud metadata addresses: currently only `168.63.129.16`, Azure's
+WireServer, which serves an Azure VM its credentials. Because it is a
+public address, listing it in `ALLOWED_EGRESS_CIDRS` reopens it.
+
 That default is also inconvenient for the thing webhooker is mostly
 for: taking a public webhook and forwarding it to something on your own
 network. A container on the same Docker network, a box on `10.x`, a
@@ -187,15 +200,16 @@ Two things this setting cannot do:
   the list is always an allowlist; an empty list (the default) means
   every private and reserved range stays refused. Note that
   `0.0.0.0/0` gets you most of the way there anyway, per above.
-- **It cannot open link-local, or a cloud metadata endpoint that
+- **It cannot open link-local, or a cloud metadata endpoint at a
-  discloses credentials or user data.** An address is on the list below
+  non-public address that discloses credentials or user data.** An
-  when both of these hold: the provider fixes it, so it cannot collide
+  address is on the list below when it is not a public address and both
-  with anything you run; and reaching it hands out credentials, user
+  of these hold: the provider fixes it, so it cannot collide with
-  data or bootstrap material. Those stay blocked no matter what you
+  anything you run; and reaching it hands out credentials, user data or
-  list, including when you list them outright or list a supernet such
+  bootstrap material. Those stay blocked no matter what you list,
-  as `0.0.0.0/0`, `::/0`, `fd00::/8` or `100.64.0.0/10`. Treat this as
+  including when you list them outright or list a supernet such as
-  best effort rather than a guarantee — it is a hand-maintained list
+  `0.0.0.0/0`, `::/0`, `fd00::/8` or `100.64.0.0/10`. Treat this as best
-  and the caveat below the table applies:
+  effort rather than a guarantee — it is a hand-maintained list and the
+  caveat below the table applies:
 
   | Blocked unconditionally | What it is |
   | ----------------------- | ---------- |
@@ -234,7 +248,8 @@ Two things this setting cannot do:
   encodings, which the default blocklist does not match. A publicly
   routable metadata address is not listed here, because nothing on this
   list can be reopened and blocking one that way would leave you no
-  escape hatch at all.
+  escape hatch at all; Azure's `168.63.129.16` is refused by the default
+  blocklist instead, as described above.
 
   This list is not exhaustive of every cloud's metadata address — if
   yours is not here, do not allowlist the block that contains it.
@@ -392,10 +407,9 @@ the bucket is. See [Rate Limiting](#rate-limiting).
 
 The remedy is to set `TRUSTED_PROXIES` to your reverse proxy's
 address, which restores per-client buckets. webhooker logs a warning
-at startup whenever `TRUSTED_PROXIES` is empty, in every environment —
+at startup whenever `TRUSTED_PROXIES` is empty, in every environment,
-not only when `WEBHOOKER_ENVIRONMENT=prod`, because that variable
+because behind a proxy every client shares one bucket in `dev` and
-defaults to `dev` and an operator who never set it is precisely the
+`prod` alike. The warning is informational when nothing proxies to the
-one at risk. The warning is informational when nothing proxies to the
 process: with no proxy in front, the peer address is the client's own
 and the buckets are already per-client. See
 [Rate Limiting](#rate-limiting) for what each limit shares.
@@ -524,6 +538,12 @@ its Argon2id hash. There is no second account and no forgot-password
 flow, so the banner and the reset command below are the only two ways
 in.
 
+A start that finds no `webhooker.db` in `DATA_DIR` also logs
+`created a new, empty database` at `WARN`, with the file's path,
+shortly before the banner. On a deployment that has run before, that
+line means `DATA_DIR` was empty, most often because its volume is not
+mounted.
+
 #### Recovering a lost admin password
 
 `webhooker resetpw` sets an existing account's password from the
@@ -538,8 +558,9 @@ printf '%s' "$NEW_PASSWORD" | \
     DATA_DIR=/var/lib/webhooker webhooker resetpw admin
 ```
 
-In a container it is the same binary, which the image sets as `CMD`
+In a container it is the same binary. The image's `CMD` is
-rather than `ENTRYPOINT`, so the whole command has to be given:
+`/app/webhooker`, and a command given to `docker run` replaces all of
+it, so the whole command has to be given:
 
 ```bash
 docker run --rm -v webhooker-data:/var/lib/webhooker \
@@ -631,7 +652,6 @@ decision:
 docker run -d \
   -p 127.0.0.1:8080:8080 \
   -v /path/to/data:/var/lib/webhooker \
-  -e WEBHOOKER_ENVIRONMENT=prod \
   -e BIND_ADDRESS=0.0.0.0 \
   webhooker:latest
 ```
@@ -677,38 +697,22 @@ those three values rather than trusting the figure. Measured at 65s on
 Docker 29.7.2.) A container `unhealthy` with `connection refused` in
 its health log, or a published port that resets connections, is this.
 
-The container runs as a non-root user (`webhooker`, UID 1000), exposes
+The app runs as a non-root user (`webhooker`, UID 1000), exposes port
-port 8080, and includes a health check against
+8080, and includes a health check against `/.well-known/healthcheck`.
-`/.well-known/healthcheck`. The `/var/lib/webhooker` volume holds all
+The `/var/lib/webhooker` volume holds all SQLite databases: the main
-SQLite databases: the main application database (`webhooker.db`), the
+application database (`webhooker.db`), the per-webhook event databases
-per-webhook event databases (`events-{uuid}.db`), and any archive
+(`events-{uuid}.db`), and any archive databases written by `database`
-databases written by `database` targets (`archive-{uuid}.db`). Mount
+targets (`archive-{uuid}.db`). Mount this as a persistent volume to
-this as a persistent volume to preserve data across container
+preserve data across container restarts.
-restarts.
 
-**The bind-mounted directory must be owned by UID 1000, or the
+**The container sets its data directory's owner and mode itself
-container does not start.** Docker creates a `-v` source path that
+before the app starts**, so a host directory can be mounted as it is,
-does not exist yet as `root:root`, and the process runs as UID 1000,
+whoever owns it. The image's `ENTRYPOINT`,
-so it cannot take its `DATA_DIR` lock:
+`deploy/docker-entrypoint.sh`, starts as root, creates `DATA_DIR` if
-
+it is missing, gives the directory and anything in it that belongs to
-```
+another user to `webhooker`, sets the directory to `0750`, and only
-webhooker: locking data directory /var/lib/webhooker: open
+then runs the app as `webhooker`. Started with `--user`, it changes
-/var/lib/webhooker/webhooker.lock: permission denied
+nothing and runs the app as that user.
-```
-
-It exits non-zero at that point, before opening any database. Create
-the directory ahead of the first `docker run`:
-
-```bash
-mkdir -p /path/to/data
-chown 1000:1000 /path/to/data
-chmod 750 /path/to/data
-```
-
-The same `chown` is what a restore needs — see step 4 of
-[Restore](#restore). A **named volume** does not have this problem:
-Docker copies the image's ownership onto a volume it initializes, and
-the image creates `/var/lib/webhooker` owned by `webhooker`.
 
 **The file modes are not yours to set, and do not depend on the
 directory.** `webhooker.db` holds target configuration in plaintext —
@@ -716,13 +720,59 @@ bearer tokens, API keys, Slack webhook URLs — along with the session
 encryption key, so webhooker creates every SQLite file it owns `0600`:
 each database and both of its `-wal` and `-shm` sidecars, across all
 three tiers. Files an earlier build left `0644` are tightened when
-they are opened. A `DATA_DIR` webhooker creates itself is `0750`, but
+they are opened. The directory's `0750` is defence in depth — it stops
-a bind mount supplies its own directory and Docker's default for one
+other local users listing the directory and learning your webhook
-it creates is `0755`; the `0600` files hold there regardless. The
+UUIDs from the `events-{uuid}.db` filenames — not the barrier
-`chmod 750` above is defence in depth — it stops other local users
+protecting the credentials.
-listing the directory and learning your webhook UUIDs from the
+
-`events-{uuid}.db` filenames — not the barrier protecting the
+### Running under upaas
-credentials.
+
+[upaas](https://git.eeqj.de/sneak/upaas) builds the image from this
+repository's `Dockerfile` and runs it. The app needs:
+
+- **Network and port:** add no port mapping in upaas. upaas publishes
+  every mapped port on all interfaces of the host
+  ([upaas issue 113](https://git.eeqj.de/sneak/upaas/issues/113)),
+  which would put the plain-HTTP admin UI and receiver there. Instead,
+  set the app's Docker Network in upaas to your reverse proxy's Docker
+  network; the proxy then reaches the app at `upaas-` followed by the
+  app name, port `8080`. Leave `PORT` unset: the image's health check
+  probes `8080`.
+- **Volume:** one host directory mounted at `/var/lib/webhooker`.
+- **Environment variables:**
+    - `WEBHOOKER_ENVIRONMENT=prod`
+    - `TRUSTED_PROXIES`: your reverse proxy's address on that Docker
+      network. The `remoteIP` field of the `http request` log line for a
+      request that came through the proxy shows it; the health check's
+      own lines show `::1`. See [Trusted proxies](#trusted-proxies).
+    - Leave `BIND_ADDRESS` and `DATA_DIR` unset: the image sets
+      `BIND_ADDRESS` to `0.0.0.0`, and `DATA_DIR` defaults to
+      `/var/lib/webhooker`.
+    - Everything else is optional; see [Configuration](#configuration).
+- **Health check:** the image's own, which requests
+  `/.well-known/healthcheck`. upaas reads the container's health 60
+  seconds after a deploy and marks the deploy failed unless it is
+  `healthy`.
+- **First run:** the first start prints the `admin` password once, in
+  the banner described under [The admin account](#the-admin-account),
+  to the container's log. upaas names the container `upaas-` followed
+  by the app name, so for an app named `webhooker`:
+
+    ```bash
+    docker logs upaas-webhooker
+    ```
+
+  If the password is lost, stop the container, set a new password with
+  the app's own image and volume, and start it again (see
+  [Recovering a lost admin password](#recovering-a-lost-admin-password)):
+
+    ```bash
+    docker stop upaas-webhooker
+    docker run --rm --volumes-from upaas-webhooker \
+        "$(docker inspect -f '{{.Image}}' upaas-webhooker)" \
+        /app/webhooker resetpw -generate admin
+    docker start upaas-webhooker
+    ```
 
 ## Deployment behind a reverse proxy
 
@@ -745,17 +795,18 @@ reports.
    serves the admin login form and the unauthenticated receiver with
    no TLS at all, and the proxy in front of it changes nothing about
    that.
-2. **Set `WEBHOOKER_ENVIRONMENT=prod`, and make sure the proxy sends
+2. **Make sure the environment is not `dev` (leave
-   `X-Forwarded-Proto`.** These are two requirements, not one. The
+   `WEBHOOKER_ENVIRONMENT` unset or set it to `prod`), and make sure
-   environment setting decides CORS and nothing else: the default
+   the proxy sends `X-Forwarded-Proto`.** These are two requirements,
+   not one. The environment setting decides CORS and nothing else:
    `dev` answers every origin with `Access-Control-Allow-Origin: *`
    (without credentials), which a server-rendered production
-   deployment has no use for. Cookie `Secure` and the strict
+   deployment has no use for, and `prod` — the default — disables it.
-   Origin/Referer mode are **not** tied to it — they are decided per
+   Cookie `Secure` and the strict Origin/Referer mode are **not** tied
-   request from the transport, which behind a proxy means the
+   to it — they are decided per request from the transport, which
-   `X-Forwarded-Proto` header. The block below sets it; without it
+   behind a proxy means the `X-Forwarded-Proto` header. The block below
-   every request is read as plaintext and cookies ship without
+   sets it; without it every request is read as plaintext and cookies
-   `Secure`. See [Configuration](#configuration).
+   ship without `Secure`. See [Configuration](#configuration).
 3. **Set `TRUSTED_PROXIES` to the proxy's address.** Unset, every rate
    limiter keys on the connecting peer, which behind a proxy is the
    proxy on every request: all clients collapse into one global bucket
@@ -847,7 +898,6 @@ sent — `$scheme` above does.
 With that block, webhooker's environment is:
 
 ```sh
-WEBHOOKER_ENVIRONMENT=prod
 BIND_ADDRESS=127.0.0.1   # the default; stated here to be explicit
 TRUSTED_PROXIES=127.0.0.1
 ```
@@ -902,15 +952,10 @@ scratch file**: it holds committed transactions that are not yet in the
 have no readable schema at all. `-shm` is regenerable, but there is no
 reason to separate the two — copy the directory and you have them.
 
-A clean shutdown closes `webhooker.db` and every `events-*.db`, which
+A clean shutdown closes every database, which checkpoints and removes
-checkpoints and removes their sidecars; a killed or crashed instance
+its sidecars; a killed or crashed instance leaves them, and they must be
-leaves them, and they must be carried with the `.db`. **Archive
+carried with the `.db`. An archive the service has not opened since a
-databases are different**: their handle is not closed at shutdown, so
+crash keeps that crash's sidecars, even across a later clean stop.
-`archive-*.db-wal` and `-shm` normally survive a clean stop and the
-`-wal` can hold every row the archive has. Measured on a stopped
-instance: `archive-….db` 4096 bytes with no table, its `-wal` 157 KB
-holding all 8 archived events. Copying `DATA_DIR` in full is what makes
-this a non-issue; copying `.db` files out of it by name is not.
 
 Configuration is **not** in `DATA_DIR` — it comes from the environment
 and from a `.env` file read out of the process working directory. Back
@@ -952,12 +997,12 @@ done
 `.backup` reads through the WAL and writes a single consistent file with
 no sidecars of its own, so the destination is complete as it stands.
 Two caveats. First, the runtime image is `alpine:3.21` with only
-`ca-certificates` added — the `sqlite3` CLI is **not** in it, so run
+`ca-certificates` and `su-exec` added — the `sqlite3` CLI is **not** in
-this on the host against the volume path, or from a throwaway container
+it, so run this on the host against the volume path, or from a
-that mounts the volume. Second, each file is captured at its own
+throwaway container that mounts the volume. Second, each file is
-instant, so a webhook created or an event delivered between two files
+captured at its own instant, so a webhook created or an event delivered
-being copied lands in one and not the other. If you need the whole set
+between two files being copied lands in one and not the other. If you
-coherent as of a single moment, stop the service.
+need the whole set coherent as of a single moment, stop the service.
 
 Note that `sqlite3 <db> .dump` is **not** one of these procedures: it is
 an export, it holds a read transaction open for as long as it runs, and
@@ -985,10 +1030,9 @@ The file becomes self-contained again when the handle closes, which
 happens on the next write past the debounce window, when the connection
 pool retires the idle connection (about a minute after the last write),
 or at the idle archive sweep — measured, the same file was a complete
-20 KB `.db` with no sidecars about a minute after its last write.
+20 KB `.db` with no sidecars about a minute after its last write. A
-Shutdown is **not** on that list: the archive handle is not closed when
+clean stop closes it too. So either move `archive-{uuid}.db` together
-the service stops. So either move `archive-{uuid}.db` together with any
+with any `-wal`/`-shm` beside it, or wait until there are none.
-`-wal`/`-shm` beside it, or wait until there are none.
 
 ### Restore
 
@@ -1007,28 +1051,16 @@ the service stops. So either move `archive-{uuid}.db` together with any
    They are part of the database, and dropping a `-wal` silently
    discards every transaction it still holds. An `.backup` set will not
    contain any: it writes a single consolidated file per database. A
-   stop-and-copy set has none for `webhooker.db` or the `events-*.db`,
+   stop-and-copy set normally has none, because a clean stop closes
-   because a clean stop closes those and checkpoints their sidecars
+   every database and checkpoints its sidecars away; the exception is an
-   away — but it will normally have them for `archive-*.db`, whose
+   archive not opened since a crash. A copy salvaged from a crashed
-   handle stays open across shutdown, and those carry the archive's
+   instance has them for everything, and needs all of them.
-   rows. A copy salvaged from a crashed instance has them for
-   everything, and needs all of them.
 
-4. **Fix ownership.** The container runs as the non-root `webhooker`
+4. Start the service. The container gives the directory and the
-   user, UID 1000 / GID 1000. Restored files must be owned by (or
+   restored files to the `webhooker` user before the app starts,
-   writable by) that UID, and so must the directory itself — SQLite
+   whoever restored them (see
-   creates the `-wal` and `-shm` sidecars beside the database, so a
+   [Running with Docker](#running-with-docker)). `AutoMigrate` runs
-   writable file inside a directory it cannot write is not enough:
+   against each restored database as it is opened.
-
-    ```bash
-    chown -R 1000:1000 /path/to/data
-    ```
-
-    Restoring as `root` on the host and forgetting this step is the
-    usual way a restore fails.
-
-5. Start the service. `AutoMigrate` runs against each restored database
-   as it is opened.
 
 ### Upgrades
 
@@ -1149,14 +1181,38 @@ backups at rest and restrict who can read them.
 
 ## The entrypoint URL is the authentication secret
 
-The receiver verifies nothing about an inbound request. The UUID in an
+**The entrypoint UUID is the credential, and it is the only one.**
-entrypoint's URL is its credential: anyone who holds that URL can
+webhooker mints a version 4 UUID per entrypoint and serves it at
-submit events to it, and the receiver checks nothing else about the
+`/webhook/{uuid}`. Possession of that URL is the authentication:
-sender. Treat an entrypoint URL the way you would treat an API token.
+anyone who holds it can submit events to the entrypoint, and the
+receiver verifies nothing else about the sender.
 
-There is no way to rotate the UUID in place. To retire one, delete the
+There is no shared secret, no HMAC signature, no bearer token and no
-entrypoint (or deactivate it, which answers `410`) and create a new
+second factor on the receiver, and none will be added. This was
-one, then point the sender at the new URL.
+considered and rejected; the implementation that existed was removed
+in [PR #279](https://git.eeqj.de/sneak/webhooker/pulls/279), closing
+[issue #67](https://git.eeqj.de/sneak/webhooker/issues/67) and
+[issue #241](https://git.eeqj.de/sneak/webhooker/issues/241). A
+proposal to reintroduce any of them — including as "defence in depth"
+alongside the UUID — is answered by this section. Inbound signature
+headers a sender sends anyway (`X-Hub-Signature` and its
+per-provider equivalents) are stored and forwarded as ordinary
+headers; nothing checks them.
+
+What that means for an operator:
+
+- **The URL is a capability, so treat it as a secret.** Keep it out of
+  logs, ticket bodies, chat messages and screenshots. Anyone who reads
+  it anywhere can post events as that sender.
+- **Rotating means minting a new entrypoint, not changing a key.**
+  There is no way to rotate the UUID in place. To retire one, delete
+  the entrypoint (or deactivate it, which answers `410`) and create a
+  new one, then point the sender at the new URL.
+- **A sender that cannot be given a secret URL is a constraint on that
+  integration, not a reason to change this.** If a service only
+  supports signed payloads to a well-known URL, raise it as its own
+  problem — pick a different integration path, or accept that it
+  cannot be used. It is not grounds to reintroduce shared secrets.
 
 ## Entrypoints
 
@@ -1166,8 +1222,15 @@ standard: normalized scripts in `script/` are the entrypoints for the
 development workflow. Ten of the Makefile's seventeen targets are thin
 shims that call them; `build`, `run`, `dev`, `deps`, `clean`, `css` and
 `version` are inline commands with no script behind them, though
-`build` and `version` both take their value from `script/version`. We
+`build` and `version` both take their value from `script/version`.
-provide:
+
+`make check` needs the third-party browser assets in `static/`, which
+are not committed, so run `make bootstrap` (or just `make assets`) once
+after cloning. Without them the tests fail with a message naming that
+remedy. `make check` does not fetch them itself because it must not
+change any files in the repo.
+
+We provide:
 
 - `script/bootstrap` — install all dependencies (idempotent)
 - `script/setup` — make a fresh clone ready for development
@@ -1234,6 +1297,16 @@ webhooker solves this by acting as a durable intermediary:
    backoff. Every delivery attempt is logged with status codes, response
    bodies, and timing.
 
+   **That guarantee is at-least-once, not exactly-once.** When a send
+   reaches its target but the write recording that outcome fails, the
+   delivery is deliberately left in a recoverable state rather than
+   marked done — losing a delivery is the worse failure — so the
+   pending sweep picks it up about fifteen minutes later, or the next
+   restart does, and the target receives a payload it already got.
+   webhooker adds no delivery identifier of its own to an outbound
+   request, so **make your receiver idempotent** against whatever the
+   payload itself carries.
+
 3. **Observability** — Full request/response logging for every webhook
    received and every delivery attempted. Prometheus metrics expose
    volume, latency, and error rates. The web UI provides real-time
@@ -1466,7 +1539,7 @@ events should be forwarded.
 | `type`           | TargetType | One of: `http`, `slack`, `database`, `log` |
 | `active`         | boolean    | Whether deliveries are enabled (default: true) |
 | `config`         | JSON text  | Type-specific configuration |
-| `max_retries`    | integer    | Maximum retry attempts for `http` and `slack` targets (0 = fire-and-forget, >0 = retries with backoff and a circuit breaker). Ignored by `database` and `log` targets |
+| `max_retries`    | integer    | Total delivery attempts for `http` and `slack` targets, not retries on top of the first: 0 is a single fire-and-forget attempt with no retries and no circuit breaker, and a value of N makes N attempts in all, with exponential backoff and a per-target circuit breaker. Ignored by `database` and `log` targets |
 | `max_queue_size` | integer    | Stored and shown on the target's detail view, but not enforced anywhere yet: nothing in the delivery engine consults it. Queue depth is set by the two fixed 10,000-entry channels |
 
 **Relations:** Belongs to Webhook. Has many Deliveries.
@@ -1474,12 +1547,12 @@ events should be forwarded.
 **Target types:**
 
 - **`http`** — Forward the event as an HTTP POST to a configured URL.
-  Behavior depends on `max_retries`: when `max_retries` is 0 (the
+  `max_retries` is the total number of delivery attempts, not retries on
-  default), the target operates in fire-and-forget mode — a single
+  top of the first: when `max_retries` is 0 (the default), the target
-  attempt with no retries and no circuit breaker. When `max_retries` is
+  operates in fire-and-forget mode, a single attempt with no retries and
-  greater than 0, failed deliveries are retried with exponential backoff
+  no circuit breaker; a value of N makes up to N attempts in all,
-  up to `max_retries` attempts, protected by a per-target circuit
+  retrying failed deliveries with exponential backoff and protecting them
-  breaker.
+  with a per-target circuit breaker.
 - **`slack`** — Post the event as a formatted message to a
   Slack-compatible incoming webhook URL (`webhookUrl` in `config`). It
   is built on the same HTTP core as `http` and honours `max_retries`
@@ -1659,6 +1732,28 @@ retries) is individually logged for full observability.
 
 **Relations:** Belongs to Delivery.
 
+#### Event-tier indexes
+
+These indexes on the per-webhook event databases are declared in the model
+tags, so `AutoMigrate` creates them on a fresh and on an existing database:
+
+| Table              | Columns                     | Serves |
+| ------------------ | --------------------------- | ------ |
+| `deliveries`       | `status`, `deleted_at`      | Startup recovery, the retry and pending sweeps every 60 seconds and the queue-depth sampler every 30 seconds, which select deliveries by status |
+| `deliveries`       | `event_id`, `deleted_at`    | The event log, which loads each event's deliveries, and retention, which selects and deletes the deliveries of expired events |
+| `delivery_results` | `delivery_id`, `deleted_at` | The event log, which loads the attempts of a page's deliveries, and retention, which deletes the attempts of expired events |
+| `events`           | `deleted_at`, `created_at`  | Retention, which selects expired events by age |
+| `events`           | `created_at`                | Retention's delete of the expired events themselves |
+
+GORM's soft delete adds `deleted_at IS NULL` to these queries; retention's
+deletes leave it out, but their lookups of expired rows keep it. SQLite keeps
+no statistics on these tables, and without them it rates the `deleted_at`
+index, which every live row matches, above an index on a column matched
+against several values or compared with `<`. So every index but the last also
+covers `deleted_at`. It comes second, so that retention's deletes can use the
+index without it, except in `events`, where `created_at` is compared with `<`
+and SQLite narrows by a `<` only on the last column it uses.
+
 #### Common Fields
 
 Every entity except `Setting` includes these fields from `BaseModel`.
@@ -1922,7 +2017,7 @@ rescans the database anyway).
 | ----------- | -------- |
 | **Closed**  | Normal operation. Deliveries flow through. Consecutive failures are counted. |
 | **Open**    | Target appears down. Deliveries are skipped and rescheduled for after the cooldown. |
-| **Half-Open** | Cooldown expired. One probe delivery is allowed to test if the target has recovered. |
+| **Half-Open** | Cooldown expired. One probe delivery is allowed to test if the target has recovered. Other deliveries are rescheduled for one whole cooldown later. |
 
 **Transitions:**
 
@@ -1960,7 +2055,9 @@ operations), and log targets (stdout) do not use circuit breakers.
 When a circuit is open and a new delivery arrives, the engine marks the
 delivery as `retrying` and schedules a retry timer for after the
 remaining cooldown period. This ensures no deliveries are lost — they're
-just delayed until the target is healthy again.
+just delayed until the target is healthy again. A delivery already in
+`retrying` keeps that status without another database write each time
+the breaker turns it away.
 
 ### Metrics
 
@@ -1974,7 +2071,7 @@ arriving and being stored, they are just not getting anywhere.
 | Metric | Type | Meaning |
 | ------ | ---- | ------- |
 | `webhooker_events_received_total` | counter | Events received and durably stored. Compare against the delivery counters on one dashboard |
-| `webhooker_delivery_attempts_total` | counter | Delivery attempts actually dispatched to a target. A delivery an open circuit breaker refused is not one: it is counted as a retry instead |
+| `webhooker_delivery_attempts_total` | counter | Delivery attempts actually dispatched to a target. A delivery a circuit breaker refused is not one: it is counted as a retry instead, but only when the refusal moves it into `retrying` |
 | `webhooker_deliveries_succeeded_total` | counter | Deliveries that reached `delivered` |
 | `webhooker_deliveries_failed_total` | counter | Deliveries that failed terminally and will not be retried |
 | `webhooker_delivery_retries_total` | counter | Deliveries put back into `retrying` |
@@ -2591,7 +2688,7 @@ abuse limit later; they are tracked as future work.
 | ------ | --------------------------- | ----------- |
 | `GET`  | `/`                         | Root redirect, 303 (authenticated → `/sources`, unauthenticated → `/pages/login`) |
 | `GET`  | `/.well-known/healthcheck`  | Health check (JSON: `status`, `now`, `uptimeSeconds`, `uptimeHuman`, `version`, `appname`, `maintenanceMode`) |
-| any    | `/s/*`                      | Static file serving (embedded CSS, JS). Mounted for every method, not just `GET`/`HEAD`: chi's `Mount` registers all methods and `http.FileServer` special-cases only `HEAD` (by omitting the body), so a `POST` or `DELETE` to an asset is answered `200` with the file. Pinned by `TestStaticServesEveryMethod` |
+| `GET`, `HEAD` | `/s/*`              | Static file serving (embedded CSS, JS). `GET` and `HEAD` only — `POST`, `PUT`, `PATCH`, `DELETE`, `OPTIONS`, `TRACE` and `CONNECT` are answered `405 Method Not Allowed` with `Allow: GET, HEAD`. Any other method (such as `PROPFIND`) is refused by chi before it reaches this route, and gets `405` without an `Allow` header. Pinned by `TestStaticServesOnlyGetAndHead` |
 | `POST` | `/webhook/{uuid}`           | Webhook receiver endpoint. `POST` only — every other method is answered `405 Method Not Allowed` with `Allow: POST`. Rate limited (see [Rate Limiting](#rate-limiting)) |
 
 #### Authentication Endpoints
@@ -2616,12 +2713,15 @@ abuse limit later; they are tracked as future work.
 | `POST` | `/source/{id}/edit`      | Edit webhook submission |
 | `POST` | `/source/{id}/delete`    | Delete webhook |
 | `GET`  | `/source/{id}/logs`      | Webhook event logs |
+| `GET`  | `/source/{id}/logs/{eventID}/body` | Download an event's full stored body. The log page renders each body only up to its cap, so this is the only route that serves a whole one; it is offered wherever a body is shown truncated |
 | `POST` | `/source/{id}/deliveries/{deliveryID}/replay` | Replay a finished delivery: creates a new delivery for the same event against the target's current configuration (30 per minute per bucket, then `429`) |
 | `POST` | `/source/{id}/events/{eventID}/resubmit` | Resubmit a stored event: creates a new event copying it and fans that out to every currently active target (30 per minute per bucket, then `429`) |
 | `POST` | `/source/{id}/entrypoints` | Add entrypoint to webhook |
 | `POST` | `/source/{id}/entrypoints/{entrypointID}/delete` | Delete an entrypoint |
 | `POST` | `/source/{id}/entrypoints/{entrypointID}/toggle` | Enable or disable an entrypoint |
 | `POST` | `/source/{id}/targets`   | Add target to webhook |
+| `GET`  | `/source/{id}/targets/{targetID}/edit` | Edit target form. The one page that renders a target's destination URL and header values in full, rather than masked |
+| `POST` | `/source/{id}/targets/{targetID}/edit` | Edit target submission |
 | `POST` | `/source/{id}/targets/{targetID}/delete` | Delete a target |
 | `POST` | `/source/{id}/targets/{targetID}/toggle` | Enable or disable a target |
 
@@ -2660,6 +2760,8 @@ webhooker/
 ├── internal/
 │   ├── banner/
 │   │   └── banner.go          # Ruled block for the one credential shown in the clear
+│   ├── ciscript/
+│   │   └── doc.go             # Tests for the CI shell scripts in script/; no runtime code
 │   ├── resetpw/
 │   │   └── resetpw.go         # `webhooker resetpw`: set an account's password, stopped deployments only
 │   ├── config/
@@ -2729,13 +2831,17 @@ webhooker/
 │   │   ├── ratelimit.go       # Per-IP rate limiting middleware (go-chi/httprate)
 │   │   ├── loginguard.go      # Login failure counters and the Argon2id verification semaphore
 │   │   └── testing.go         # NewForTest: Middleware without the fx lifecycle
+│   ├── reqtls/
+│   │   └── reqtls.go          # IsTLS: the one TLS predicate, r.TLS or X-Forwarded-Proto
 │   ├── server/
 │   │   ├── server.go          # Server struct, fx lifecycle, signal handling
 │   │   ├── http.go            # HTTP server setup with timeouts
 │   │   └── routes.go          # All route definitions
-│   └── session/
+│   ├── session/
-│       ├── session.go         # Cookie-based session management
+│   │   ├── session.go         # Cookie-based session management
-│       └── testing.go         # NewForTest: Session without the fx lifecycle
+│   │   └── testing.go         # NewForTest: Session without the fx lifecycle
+│   └── versionscript/
+│       └── doc.go             # Tests for script/version and the build files that use it
 ├── static/
 │   ├── static.go              # //go:embed directive
 │   ├── css/input.css          # Tailwind input, source for tailwind.css (make css)
@@ -2848,6 +2954,10 @@ check, see [The login endpoint](#the-login-endpoint).
 
 ### Authentication
 
+- **Webhook receiver:** the entrypoint UUID in the URL, and nothing
+  else. No shared secret, no HMAC signature, no token header, and none
+  will be added — see
+  [The entrypoint URL is the authentication secret](#the-entrypoint-url-is-the-authentication-secret).
 - **Web UI:** Cookie-based sessions using gorilla/sessions with
   encrypted cookies. Sessions are configured with HttpOnly, SameSite
   Lax, and Secure whenever the request is on TLS — the flag follows the
@@ -2887,7 +2997,8 @@ check, see [The login endpoint](#the-login-endpoint).
   mode
 - **The entrypoint URL is the receiver's only credential.** Nothing
   about an inbound request is verified; possession of the UUID
-  authorises submission (see
+  authorises submission, and no shared secret or signature check will
+  be added alongside it (see
   [The entrypoint URL is the authentication secret](#the-entrypoint-url-is-the-authentication-secret))
 - **SSRF prevention** for HTTP delivery targets: private/reserved IP
   ranges (RFC 1918, loopback, link-local, cloud metadata) are blocked
@@ -2927,7 +3038,11 @@ check, see [The login endpoint](#the-login-endpoint).
 - Prometheus metrics behind basic auth
 - Static assets embedded in binary (no filesystem access needed at
   runtime)
-- Container runs as non-root user (UID 1000)
+- The app runs as the non-root `webhooker` user (UID 1000) in the
+  container. The image sets no `USER`, so these run as root: the
+  `ENTRYPOINT` script, which sets the data directory's owner and mode
+  before the app starts; the image's health check; and `docker exec`,
+  unless given `--user`
 - GORM soft deletes on every entity that carries `BaseModel`, which is
   all of them but `Setting` (data preserved for audit)
 
@@ -2943,7 +3058,8 @@ each hook. The order, read off the fx stop-hook log:
 3. `server` — the HTTP drain, bounded separately by
    `server.ShutdownTimeout` (**3 seconds**), then a Sentry flush if
    `SENTRY_DSN` is set
-4. `delivery.Engine`
+4. `delivery.Engine` — waits for its workers, then closes the archive
+   databases
 5. `healthcheck`
 6. `WebhookDBManager`
 7. the database close
@@ -3053,10 +3169,13 @@ version is fixed independently of the compiler's:
    `GO_LDFLAGS`, so neither can drop the `-X` that stamps the version.
    The version arrives as the `VERSION` build arg, since the context
    has no `.git` (see [Version stamping](#version-stamping)).
-3. **Runtime stage** (`alpine:3.21`) — copies the static binary,
+3. **Runtime stage** (`alpine:3.21`) — copies the static binary and
-   creates the `/var/lib/webhooker` directory for all SQLite databases,
+   `deploy/docker-entrypoint.sh`, creates the `/var/lib/webhooker`
-   runs as the non-root `webhooker` user (UID 1000), exposes port 8080,
+   directory for all SQLite databases, exposes port 8080, and includes
-   and includes a health check against `/.well-known/healthcheck`.
+   a health check against `/.well-known/healthcheck`. It sets no
+   `USER`: the `ENTRYPOINT` script starts as root, sets the data
+   directory's owner and mode, and runs the app as the non-root
+   `webhooker` user (UID 1000) through `su-exec`.
 
 The lint stage invokes `golangci-lint` directly rather than `make lint`:
 it is already the pinned linter image, and `make lint` builds
@@ -3135,3 +3254,5 @@ MIT
 ## Author
 
 [@sneak](https://sneak.berlin)
+
+

TODO.md

@@ -18,18 +18,27 @@ Issue branches do NOT touch this file — the manager maintains it on
 
 # Status
 
-1.0.0 is open, with work remaining. The milestone
+The milestone (https://git.eeqj.de/sneak/webhooker/milestone/9) is the
-(https://git.eeqj.de/sneak/webhooker/milestone/9) is the authoritative
+authoritative list, and the only place to read a count or a state of
-list, and the only place to read a count or a state of play from. This
+play from. This file records where the project is, not what is in
-file records where the project is, not what is in flight: a sentence
+flight: a sentence whose truth depends on a branch being unmerged is
-whose truth depends on a branch being unmerged is wrong the moment it
+wrong the moment it merges, and this file has been wrong that way
-merges, and this file has been wrong that way before.
+before.
 
-The tag is held on a durability defect
+The durability defect that held the tag has landed
-(https://git.eeqj.de/sneak/webhooker/issues/256): a concurrent reader
+(https://git.eeqj.de/sneak/webhooker/issues/256, commit `8d64259`).
-of a per-webhook event database strands delivered webhooks at
+Every SQLite handle opens with WAL journaling and a busy timeout, a
-`pending`, and the next restart re-delivers them. That issue gates
+bookkeeping write that fails leaves its delivery in a recoverable
-`v1.0.0`, and is where the fix's own state is tracked.
+state rather than a lying one, and recovery skips a delivery that
+already has a successful result row. Final pre-tag verification
+exercised it and confirmed it holds. Whatever the milestone still
+shows open is what remains before `v1.0.0`.
+
+Delivery is at-least-once by design, not by accident: a send whose
+result row does not land is attempted again, so a receiver can see a
+duplicate. That is deliberate — the alternative is a silent lost
+delivery — and the README says so under Rationale. It is not a defect
+to re-file.
 
 One caveat on reading a green check: a docs-only commit deliberately
 replays from the layer cache
@@ -39,11 +48,11 @@ commit invalidates the `COPY` layer and genuinely executes.
 
 # Next Step
 
-Land https://git.eeqj.de/sneak/webhooker/issues/256, then clear the
+Clear the rest of the open 1.0.0 milestone
-rest of the open 1.0.0 milestone and tag `v1.0.0`. Merging `next` into
+(https://git.eeqj.de/sneak/webhooker/milestone/9) and tag `v1.0.0`.
-`main` is a separate act from tagging and waits on neither of those:
+Merging `next` into `main` is a separate act from tagging and waits on
-`next` is kept mergeable at all times, which is the point of the
+neither of those: `next` is kept mergeable at all times, which is the
-branch.
+point of the branch.
 
 # Completed Steps
 

deploy/docker-entrypoint.sh

@@ -0,0 +1,22 @@
+#!/bin/sh
+# deploy/docker-entrypoint.sh: the image's ENTRYPOINT. A bind-mounted
+# data directory keeps its owner from the host, often root, and the app
+# could not write to it. Started as root, this creates DATA_DIR if
+# needed, gives it and everything in it to webhooker, sets its mode, and
+# runs the command as webhooker, so the app never runs as root. Started
+# as another user, it only runs the command.
+set -eu
+
+main() {
+    if [ "$(id -u)" != 0 ]; then
+        exec "$@"
+    fi
+
+    dir="${DATA_DIR:-/var/lib/webhooker}"
+    mkdir -p "$dir"
+    find "$dir" ! -user webhooker -exec chown -h webhooker:webhooker {} +
+    chmod 750 "$dir"
+    exec su-exec webhooker "$@"
+}
+
+main "$@"

internal/config/config.go

@@ -192,9 +192,10 @@ type Config struct {
 	// alwaysBlockedNetworks stays blocked no matter what is listed
 	// here. That set is link-local plus the cloud metadata
 	// endpoints outside it that disclose credentials or user data
-	// at a provider-fixed address; it is not exhaustive of every
+	// at a provider-fixed, non-public address; it is not
-	// cloud's metadata address. See alwaysBlockedNetworks for the
+	// exhaustive of every cloud's metadata address. See
-	// authoritative list and the criterion it is built from.
+	// alwaysBlockedNetworks for the authoritative list and the
+	// criterion it is built from.
 	AllowedEgressCIDRs []netip.Prefix
 
 	params *ConfigParams
@@ -585,12 +586,14 @@ func resolveMetricsAuth() (string, string, error) {
 	)
 }
 
-// resolveEnvironment reads WEBHOOKER_ENVIRONMENT, defaulting to
+// resolveEnvironment reads WEBHOOKER_ENVIRONMENT, defaulting to prod
-// dev, and rejects unrecognised values.
+// when it is unset so a deployment that forgets the variable is not
+// silently permissive; dev must be set explicitly. It rejects
+// unrecognised values.
 func resolveEnvironment() (string, error) {
 	environment := os.Getenv("WEBHOOKER_ENVIRONMENT")
 	if environment == "" {
-		environment = EnvironmentDev
+		environment = EnvironmentProd
 	}
 
 	if environment != EnvironmentDev &&
@@ -744,12 +747,14 @@ func (c *Config) warnEgressAllowlist(log *slog.Logger) {
 
 	log.Warn(
 		"ALLOWED_EGRESS_CIDRS lets delivery targets reach these "+
-			"otherwise-blocked private/reserved networks. Anyone "+
+			"otherwise-blocked networks. Anyone who can create a "+
-			"who can create a delivery target can now make this "+
+			"delivery target can now make this process issue "+
-			"process issue requests into them, and read back the "+
+			"requests into them, and read back the response. Only "+
-			"response. Link-local and the known cloud instance "+
+			"the addresses the README lists as blocked "+
-			"metadata endpoints outside it stay blocked "+
+			"unconditionally stay blocked regardless of what is "+
-			"regardless of what is listed here.",
+			"listed here; a public cloud metadata address such as "+
+			"168.63.129.16 is reachable once it, or a block "+
+			"covering it, is listed.",
 		"allowedEgressCIDRs",
 		strings.Join(PrefixStrings(c.AllowedEgressCIDRs), ","),
 	)
@@ -772,10 +777,8 @@ func (c *Config) warnEgressAllowlist(log *slog.Logger) {
 // everyone else's wrong passwords, and the receiver's limits become
 // service-wide ceilings.
 //
-// The warning is deliberately not gated on WEBHOOKER_ENVIRONMENT. That
+// The warning is deliberately not gated on WEBHOOKER_ENVIRONMENT:
-// variable defaults to dev, so gating on it would silence the warning
+// behind a proxy every client shares one bucket in dev and prod alike.
-// for exactly the operator who forgot to configure the deployment —
-// the case it exists to catch.
 //
 // The default of trusting nobody is deliberate — trusting forwarded
 // headers from arbitrary peers lets any client choose its own bucket —

internal/config/config_test.go

@@ -44,9 +44,9 @@ func TestEnvironmentConfig(t *testing.T) {
 		isProd      bool
 	}{
 		{
-			name:   "default is dev",
+			name:   "default is prod",
-			isDev:  true,
+			isDev:  false,
-			isProd: false,
+			isProd: true,
 		},
 		{
 			name:     "explicit dev",
@@ -834,12 +834,13 @@ func TestEgressAllowlistWarning(t *testing.T) {
 			// to be able to read back which networks are open.
 			assert.Contains(t, logged, "10.0.0.0/8")
 			assert.Contains(t, logged, "127.0.0.0/8")
-			// What stays shut. Asserted on the clause naming the
+			// What stays shut is the whole unconditional set, not
-			// wider set rather than on "Link-local" alone, so the
+			// link-local alone; a public metadata address is not in
-			// string cannot narrow back to link-local only while
+			// it, so a listed block covering it opens it.
-			// the always-blocked set covers ULA, CGNAT and two
+			assert.Contains(t, logged, "blocked unconditionally")
-			// public metadata addresses as well.
+			assert.Contains(t, logged, "168.63.129.16 is reachable")
-			assert.Contains(t, logged, "metadata endpoints outside it")
+			// The listed blocks need not be private or reserved.
+			assert.NotContains(t, logged, "private/reserved")
 		})
 	}
 }
@@ -848,10 +849,9 @@ func TestEgressAllowlistWarning(t *testing.T) {
 // tells an operator a deployment behind a reverse proxy shares one
 // rate-limit bucket between every client, which turns the receiver
 // limits into service-wide ceilings and collapses login failure
-// counting. It must fire whenever TRUSTED_PROXIES is empty,
+// counting. It must fire whenever TRUSTED_PROXIES is empty, in any
-// in any environment: WEBHOOKER_ENVIRONMENT defaults to dev, so gating
+// environment, because behind a proxy every client shares one bucket
-// on it would silence the warning for exactly the operator who never
+// in dev and prod alike. It stays quiet once proxies are named.
-// configured the deployment. It stays quiet once proxies are named.
 func TestSharedRateLimitBucketWarning(t *testing.T) {
 	tests := []struct {
 		name           string
@@ -871,10 +871,6 @@ func TestSharedRateLimitBucketWarning(t *testing.T) {
 			expectWarning:  false,
 		},
 		{
-			// The default environment. An internet-exposed
-			// deployment whose operator never set
-			// WEBHOOKER_ENVIRONMENT lands here and has exactly
-			// the exposure the warning announces.
 			name:          "dev without trusted proxies warns",
 			environment:   config.EnvironmentDev,
 			expectWarning: true,

internal/database/bootstrap_banner_test.go

@@ -3,6 +3,8 @@ package database_test
 import (
 	"bytes"
 	"context"
+	"log/slog"
+	"path/filepath"
 	"strings"
 	"testing"
 
@@ -83,3 +85,37 @@ func TestFirstBoot_PrintsTheAdminPasswordAsABanner(t *testing.T) {
 		t, ok, "the printed password must open the seeded account",
 	)
 }
+
+// TestNewDatabase_IsLoggedWithItsPath is the log half of
+// https://git.eeqj.de/sneak/webhooker/issues/359. A DATA_DIR that is
+// unexpectedly empty boots exactly like a first start, so the start
+// that creates the database must say so, and where. Opening that
+// database again must not.
+func TestNewDatabase_IsLoggedWithItsPath(t *testing.T) {
+	t.Parallel()
+
+	dir := t.TempDir()
+
+	open := func() string {
+		var out bytes.Buffer
+
+		db, err := database.Open(dir, slog.New(slog.NewTextHandler(&out, nil)))
+		require.NoError(t, err)
+		require.NoError(t, db.Close())
+
+		return out.String()
+	}
+
+	const created = `level=WARN msg="created a new, empty database"`
+
+	first := open()
+	second := open()
+
+	assert.Contains(
+		t, first,
+		created+" path="+filepath.Join(dir, database.MainDBFileName),
+	)
+	assert.NotContains(
+		t, second, created, "an existing database is not new",
+	)
+}

internal/database/database.go

@@ -8,6 +8,7 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"io/fs"
 	"log/slog"
 	"os"
 	"path/filepath"
@@ -17,12 +18,12 @@ import (
 	"gorm.io/gorm"
 	"sneak.berlin/go/webhooker/internal/banner"
 	"sneak.berlin/go/webhooker/internal/config"
+	"sneak.berlin/go/webhooker/internal/datadir"
 	"sneak.berlin/go/webhooker/internal/gormlog"
 	"sneak.berlin/go/webhooker/internal/logger"
 )
 
 const (
-	dataDirPerm       = 0750
 	randomPasswordLen = 16
 	sessionKeyLen     = 32
 )
@@ -185,7 +186,9 @@ func (d *Database) connect() error {
 // caller's decision.
 func (d *Database) connectTo(dataDir string) error {
 	// Ensure the data directory exists before opening the database.
-	err := os.MkdirAll(dataDir, dataDirPerm)
+	// datadir.DirPerm is the single source of the directory mode; this
+	// package creates the directory too, since either may run first.
+	err := os.MkdirAll(dataDir, datadir.DirPerm)
 	if err != nil {
 		return fmt.Errorf(
 			"creating data directory %s: %w",
@@ -197,6 +200,12 @@ func (d *Database) connectTo(dataDir string) error {
 	// Construct the main application database path inside DATA_DIR.
 	dbPath := filepath.Join(dataDir, MainDBFileName)
 
+	// Checked before opening, which creates the file. A DATA_DIR that
+	// is unexpectedly empty -- its volume not mounted, say -- looks
+	// exactly like a first start, so a new database is a warning.
+	_, statErr := os.Stat(dbPath)
+	created := errors.Is(statErr, fs.ErrNotExist)
+
 	// Opened through OpenSQLite so this handle carries the same WAL
 	// journaling, busy timeout, immediate-transaction locking, and pool
 	// bounds as every other database file. See sqlite_open.go.
@@ -227,7 +236,12 @@ func (d *Database) connectTo(dataDir string) error {
 	}
 
 	d.db = db
+
+	if created {
+		d.log.Warn("created a new, empty database", "path", dbPath)
+	} else {
 		d.log.Info("connected to database", "path", dbPath)
+	}
 
 	// Run migrations
 	return d.migrate()

internal/database/event_tier_indexes_test.go

@@ -0,0 +1,166 @@
+package database_test
+
+import (
+	"context"
+	"fmt"
+	"testing"
+	"time"
+
+	"github.com/google/uuid"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"gorm.io/gorm"
+	"sneak.berlin/go/webhooker/internal/database"
+)
+
+// TestWebhookDBManager_OpenAddsEventTierIndexes verifies that opening a
+// per-webhook database that predates these indexes creates them. It
+// stands in for an older database file by dropping the indexes
+// AutoMigrate just created, then reopening the same file.
+func TestWebhookDBManager_OpenAddsEventTierIndexes(t *testing.T) {
+	t.Parallel()
+
+	indexes := []struct {
+		model any
+		name  string
+	}{
+		{&database.Delivery{}, "idx_deliveries_status"},
+		{&database.Delivery{}, "idx_deliveries_event_id"},
+		{&database.DeliveryResult{}, "idx_delivery_results_delivery_id"},
+		{&database.Event{}, "idx_events_deleted_at_created_at"},
+		{&database.Event{}, "idx_events_created_at"},
+	}
+
+	mgr, lc := setupTestWebhookDBManager(t)
+	ctx := context.Background()
+	require.NoError(t, lc.Start(ctx))
+
+	defer func() { require.NoError(t, lc.Stop(ctx)) }()
+
+	webhookID := uuid.New().String()
+
+	db, err := mgr.GetDB(webhookID)
+	require.NoError(t, err)
+
+	// A fresh database has them.
+	for _, ix := range indexes {
+		require.True(t, db.Migrator().HasIndex(ix.model, ix.name))
+	}
+
+	// Stand in for a database file created before the indexes existed.
+	for _, ix := range indexes {
+		require.NoError(t, db.Migrator().DropIndex(ix.model, ix.name))
+		require.False(t, db.Migrator().HasIndex(ix.model, ix.name))
+	}
+
+	// Drop the cached connection so the next open reopens the file and
+	// runs AutoMigrate against it, as a restart would.
+	require.NoError(t, mgr.CloseAll())
+
+	db, err = mgr.GetDB(webhookID)
+	require.NoError(t, err)
+
+	for _, ix := range indexes {
+		assert.True(t, db.Migrator().HasIndex(ix.model, ix.name),
+			"opening the existing database should create %s", ix.name)
+	}
+}
+
+// TestEventTierQueriesUseTheirIndexes verifies that the statements the
+// indexes are for use them. GORM builds each statement in a dry run as
+// the code named above it does, soft-delete condition included, and
+// SQLite, which keeps no statistics on these tables, must plan to seek
+// on each index listed by the columns in parentheses.
+func TestEventTierQueriesUseTheirIndexes(t *testing.T) {
+	t.Parallel()
+
+	mgr, lc := setupTestWebhookDBManager(t)
+	ctx := context.Background()
+	require.NoError(t, lc.Start(ctx))
+
+	defer func() { require.NoError(t, lc.Stop(ctx)) }()
+
+	db, err := mgr.GetDB(uuid.New().String())
+	require.NoError(t, err)
+
+	dry := db.Session(&gorm.Session{DryRun: true})
+	ids := []string{
+		uuid.New().String(), uuid.New().String(), uuid.New().String(),
+	}
+	cutoff := time.Now()
+
+	var (
+		deliveries []database.Delivery
+		results    []database.DeliveryResult
+		depths     []struct{ Depth int }
+	)
+
+	byStatus := "idx_deliveries_status (status=? AND deleted_at=?)"
+	byEvent := "idx_deliveries_event_id (event_id=? AND deleted_at=?)"
+	byAge := "idx_events_deleted_at_created_at (deleted_at=? AND created_at<?)"
+
+	// The delivery engine: recovery and the retry sweep, the sweep for
+	// stranded pending deliveries, and the queue depth count.
+	assertPlanUses(t, db, dry.Where(
+		"status = ?", database.DeliveryStatusRetrying,
+	).Find(&deliveries), byStatus)
+	assertPlanUses(t, db, dry.Where(
+		"status = ? AND updated_at < ?",
+		database.DeliveryStatusPending, cutoff,
+	).Limit(500).Find(&deliveries), byStatus)
+	assertPlanUses(t, db, dry.Model(&database.Delivery{}).
+		Select("target_id", "status", "count(*) as depth").
+		Where("status IN ?", []database.DeliveryStatus{
+			database.DeliveryStatusPending,
+			database.DeliveryStatusRetrying,
+		}).Group("target_id, status").Find(&depths), byStatus)
+
+	// The event log: each event's deliveries, then their attempts
+	// (loadEventsWithDeliveries, loadDeliveryResults).
+	assertPlanUses(t, db, dry.Where("event_id = ?", ids[0]).
+		Find(&deliveries), byEvent)
+	assertPlanUses(t, db, dry.Where("delivery_id IN ?", ids).
+		Order("attempt_num ASC").Find(&results),
+		"idx_delivery_results_delivery_id (delivery_id=? AND deleted_at=?)")
+
+	// Retention's three deletes (reapExpired), whose subqueries are built
+	// afresh for each statement as it builds them.
+	expiredEventIDs := func() *gorm.DB {
+		return dry.Model(&database.Event{}).Select("id").
+			Where("created_at < ?", cutoff)
+	}
+
+	assertPlanUses(t, db, dry.Unscoped().Where(
+		"delivery_id IN (?)", dry.Model(&database.Delivery{}).
+			Select("id").Where("event_id IN (?)", expiredEventIDs()),
+	).Delete(&database.DeliveryResult{}),
+		"idx_delivery_results_delivery_id (delivery_id=?)", byEvent, byAge)
+	assertPlanUses(t, db, dry.Unscoped().Where(
+		"event_id IN (?)", expiredEventIDs(),
+	).Delete(&database.Delivery{}),
+		"idx_deliveries_event_id (event_id=?)", byAge)
+	assertPlanUses(t, db, dry.Unscoped().Where(
+		"created_at < ?", cutoff,
+	).Delete(&database.Event{}), "idx_events_created_at (created_at<?)")
+}
+
+// assertPlanUses asserts that SQLite's plan for a statement GORM built
+// in a dry run, run with the same SQL and arguments GORM would send,
+// names each of the given indexes.
+func assertPlanUses(
+	t *testing.T, db, built *gorm.DB, indexes ...string,
+) {
+	t.Helper()
+
+	var plan []struct{ Detail string }
+
+	require.NoError(t, db.Raw(
+		"EXPLAIN QUERY PLAN "+built.Statement.SQL.String(),
+		built.Statement.Vars...,
+	).Scan(&plan).Error)
+
+	for _, index := range indexes {
+		assert.Contains(t, fmt.Sprint(plan), index,
+			built.Statement.SQL.String())
+	}
+}

internal/database/model_delivery.go

@@ -1,5 +1,7 @@
 package database
 
+import "gorm.io/gorm"
+
 // DeliveryStatus represents the status of a delivery
 type DeliveryStatus string
 
@@ -29,12 +31,19 @@ func (s DeliveryStatus) Terminal() bool {
 }
 
 // Delivery represents a delivery attempt for an event to a target
+//
+//nolint:lll // a struct tag cannot wrap
 type Delivery struct {
 	BaseModel
 
-	EventID  string         `gorm:"type:uuid;not null"         json:"eventId"`
+	EventID  string         `gorm:"type:uuid;not null;index:idx_deliveries_event_id,priority:1"       json:"eventId"`
 	TargetID string         `gorm:"type:uuid;not null"                                                json:"targetId"`
-	Status   DeliveryStatus `gorm:"not null;default:'pending'" json:"status"`
+	Status   DeliveryStatus `gorm:"not null;default:'pending';index:idx_deliveries_status,priority:1" json:"status"`
+
+	// DeletedAt repeats the BaseModel field only to be the second column
+	// of the event_id and status indexes, for the reason DeliveryResult
+	// gives.
+	DeletedAt gorm.DeletedAt `gorm:"index:idx_deliveries_event_id,priority:2;index:idx_deliveries_status,priority:2" json:"deletedAt,omitzero"`
 
 	// Relations
 	Event           Event            `json:"event,omitzero"`

internal/database/model_delivery_result.go

@@ -1,10 +1,21 @@
 package database
 
+import "gorm.io/gorm"
+
 // DeliveryResult represents the result of a delivery attempt
+//
+//nolint:lll // a struct tag cannot wrap
 type DeliveryResult struct {
 	BaseModel
 
-	DeliveryID   string `gorm:"type:uuid;not null"   json:"deliveryId"`
+	// DeliveryID and DeletedAt make up one index, in that order.
+	// DeletedAt repeats the BaseModel field only to join it: GORM adds
+	// "deleted_at IS NULL" to almost every query, and where a column is
+	// matched against several values SQLite otherwise reads through the
+	// deleted_at index, which every live row matches.
+	DeliveryID string         `gorm:"type:uuid;not null;index:idx_delivery_results_delivery_id,priority:1" json:"deliveryId"`
+	DeletedAt  gorm.DeletedAt `gorm:"index:idx_delivery_results_delivery_id,priority:2"                    json:"deletedAt,omitzero"`
+
 	AttemptNum   int    `gorm:"not null"             json:"attemptNum"`
 	Success      bool   `json:"success"`
 	StatusCode   int    `json:"statusCode,omitempty"`

internal/database/model_event.go

@@ -1,9 +1,27 @@
 package database
 
+import (
+	"time"
+
+	"gorm.io/gorm"
+)
+
 // Event represents a captured webhook event
+//
+//nolint:lll // a struct tag cannot wrap
 type Event struct {
 	BaseModel
 
+	// CreatedAt and DeletedAt repeat the BaseModel fields only to index
+	// them for retention, which finds events by age. Its lookups carry
+	// GORM's "deleted_at IS NULL" (see DeliveryResult) and compare
+	// created_at with <, so their index has deleted_at first: SQLite
+	// narrows by a < only on the last column it uses. Its final delete
+	// has no deleted_at condition and uses the index on created_at
+	// alone. The other tables keep the unindexed BaseModel created_at.
+	CreatedAt time.Time      `gorm:"index;index:idx_events_deleted_at_created_at,priority:2" json:"createdAt"`
+	DeletedAt gorm.DeletedAt `gorm:"index:idx_events_deleted_at_created_at,priority:1"       json:"deletedAt,omitzero"`
+
 	WebhookID    string `gorm:"type:uuid;not null" json:"webhookId"`
 	EntrypointID string `gorm:"type:uuid;not null" json:"entrypointId"`
 

internal/database/webhook_db_manager.go

@@ -13,6 +13,7 @@ import (
 	"gorm.io/driver/sqlite"
 	"gorm.io/gorm"
 	"sneak.berlin/go/webhooker/internal/config"
+	"sneak.berlin/go/webhooker/internal/datadir"
 	"sneak.berlin/go/webhooker/internal/gormlog"
 	"sneak.berlin/go/webhooker/internal/logger"
 )
@@ -40,6 +41,11 @@ type WebhookDBManager struct {
 	dataDir string
 	dbs     sync.Map // map[webhookID]*gorm.DB
 	log     *slog.Logger
+
+	// mu is held while a database is opened, deleted, or closed, so
+	// each file has at most one open handle. Reading an already cached
+	// handle does not take it.
+	mu sync.Mutex
 }
 
 // NewWebhookDBManager creates a new WebhookDBManager and
@@ -53,8 +59,9 @@ func NewWebhookDBManager(
 		log:     params.Logger.Get(),
 	}
 
-	// Create data directory if it doesn't exist
+	// Create data directory if it doesn't exist. datadir.DirPerm is the
-	err := os.MkdirAll(m.dataDir, dataDirPerm)
+	// single source of the directory mode; either package may run first.
+	err := os.MkdirAll(m.dataDir, datadir.DirPerm)
 	if err != nil {
 		return nil, fmt.Errorf(
 			"creating data directory %s: %w",
@@ -84,35 +91,34 @@ func (m *WebhookDBManager) GetDB(
 ) (*gorm.DB, error) {
 	// Fast path: already open
 	if val, ok := m.dbs.Load(webhookID); ok {
-		cachedDB, castOK := val.(*gorm.DB)
+		return asGormDB(val, webhookID)
-		if !castOK {
-			return nil, fmt.Errorf(
-				"%w for webhook %s",
-				errInvalidCachedDBType,
-				webhookID,
-			)
 	}
 
-		return cachedDB, nil
+	// Slow path: open the database under the lock, looking in the
+	// cache again first. A caller that raced another one here then
+	// waits for its handle instead of opening a second one.
+	m.mu.Lock()
+	defer m.mu.Unlock()
+
+	if val, ok := m.dbs.Load(webhookID); ok {
+		return asGormDB(val, webhookID)
 	}
 
-	// Slow path: open/create the database
 	db, err := m.openDB(webhookID)
 	if err != nil {
 		return nil, err
 	}
 
-	// Store it; if another goroutine beat us, close ours
+	m.dbs.Store(webhookID, db)
-	actual, loaded := m.dbs.LoadOrStore(webhookID, db)
+
-	if loaded {
+	return db, nil
-		// Another goroutine created it first; close our duplicate
-		sqlDB, closeErr := db.DB()
-		if closeErr == nil {
-			_ = sqlDB.Close()
 }
 
-		existingDB, castOK := actual.(*gorm.DB)
+// asGormDB returns a value read from the cache as the database
-		if !castOK {
+// handle it is.
+func asGormDB(val any, webhookID string) (*gorm.DB, error) {
+	db, ok := val.(*gorm.DB)
+	if !ok {
 		return nil, fmt.Errorf(
 			"%w for webhook %s",
 			errInvalidCachedDBType,
@@ -120,9 +126,6 @@ func (m *WebhookDBManager) GetDB(
 		)
 	}
 
-		return existingDB, nil
-	}
-
 	return db, nil
 }
 
@@ -151,6 +154,11 @@ func (m *WebhookDBManager) DBExists(
 func (m *WebhookDBManager) DeleteDB(
 	webhookID string,
 ) error {
+	// Held until the files are gone, so GetDB cannot open the file
+	// again between the close and the removal.
+	m.mu.Lock()
+	defer m.mu.Unlock()
+
 	// Close and remove from cache
 	if val, ok := m.dbs.LoadAndDelete(webhookID); ok {
 		if gormDB, castOK := val.(*gorm.DB); castOK {
@@ -184,6 +192,11 @@ func (m *WebhookDBManager) DeleteDB(
 // CloseAll closes all open per-webhook database connections.
 // Called during application shutdown.
 func (m *WebhookDBManager) CloseAll() error {
+	// An open already under way finishes and is cached first, so it
+	// is closed here rather than cached after this loop has passed.
+	m.mu.Lock()
+	defer m.mu.Unlock()
+
 	var lastErr error
 
 	m.dbs.Range(func(key, value any) bool {

internal/database/webhook_db_manager_test.go

@@ -1,10 +1,14 @@
 package database_test
 
 import (
+	"bytes"
 	"context"
+	"log/slog"
 	"net/http"
 	"os"
 	"path/filepath"
+	"strings"
+	"sync"
 	"testing"
 
 	"github.com/google/uuid"
@@ -104,6 +108,54 @@ func TestWebhookDBManager_CreateAndGetDB(t *testing.T) {
 	assert.Equal(t, `{"test": true}`, readEvent.Body)
 }
 
+// Many callers ask for one webhook's database at the same moment,
+// before it is cached. Only one of them may open the file; the others
+// must wait for its handle. openDB logs one "opened per-webhook
+// database" line per open, and those lines are what is counted.
+func TestWebhookDBManager_ConcurrentFirstTouchOpensOnce(t *testing.T) {
+	t.Parallel()
+
+	var logs bytes.Buffer
+
+	mgr := database.NewTestWebhookDBManagerWithLogger(
+		t.TempDir(),
+		slog.New(slog.NewTextHandler(&logs, nil)),
+	)
+
+	t.Cleanup(func() { assert.NoError(t, mgr.CloseAll()) })
+
+	webhookID := uuid.New().String()
+
+	const callers = 16
+
+	start := make(chan struct{})
+	handles := make([]*gorm.DB, callers)
+	errs := make([]error, callers)
+
+	var wg sync.WaitGroup
+
+	for i := range callers {
+		wg.Go(func() {
+			<-start
+
+			handles[i], errs[i] = mgr.GetDB(webhookID)
+		})
+	}
+
+	close(start)
+	wg.Wait()
+
+	for i := range callers {
+		require.NoError(t, errs[i])
+		assert.Same(t, handles[0], handles[i])
+	}
+
+	assert.Equal(
+		t, 1,
+		strings.Count(logs.String(), "opened per-webhook database"),
+	)
+}
+
 func TestWebhookDBManager_DeleteDB(t *testing.T) {
 	t.Parallel()
 

internal/datadir/lock.go

@@ -29,9 +29,11 @@ import (
 // process that was killed with SIGKILL blocks nothing.
 const LockFileName = "webhooker.lock"
 
-// dirPerm is the mode Acquire creates DATA_DIR with. It matches what
+// DirPerm is the mode DATA_DIR is created with. It is the single
-// internal/database uses, since whichever runs first creates it.
+// source of that mode: internal/database consumes it rather than
-const dirPerm = 0o750
+// keeping its own copy, so the two packages that both create the
+// directory cannot drift into disagreeing about its permissions.
+const DirPerm = 0o750
 
 // ErrLocked reports that another live process holds the data
 // directory. Callers that need to know whether a deployment is running
@@ -64,7 +66,7 @@ func Acquire(dir string) (*Lock, error) {
 		return nil, ErrNoDir
 	}
 
-	err := os.MkdirAll(dir, dirPerm)
+	err := os.MkdirAll(dir, DirPerm)
 	if err != nil {
 		return nil, fmt.Errorf(
 			"creating data directory %s: %w", dir, err,

internal/delivery/circuit_breaker.go

@@ -76,12 +76,20 @@ func (cb *CircuitBreaker) Allow() bool {
 	}
 }
 
-// CooldownRemaining returns how much time is left before
+// CooldownRemaining returns how long a delivery that Allow refused
-// an open circuit transitions to half-open.
+// should wait before it is tried again. Closed, it returns zero.
+// Open, it returns what is left of the cooldown, or zero once that
+// has passed. Half-open, it returns the whole cooldown: the one
+// probe delivery is still in flight, and if it fails the circuit
+// reopens for that long.
 func (cb *CircuitBreaker) CooldownRemaining() time.Duration {
 	cb.mu.Lock()
 	defer cb.mu.Unlock()
 
+	if cb.state == CircuitHalfOpen {
+		return cb.cooldown
+	}
+
 	if cb.state != CircuitOpen {
 		return 0
 	}

internal/delivery/circuit_breaker_test.go

@@ -267,7 +267,7 @@ func TestCircuitBreaker_CooldownRemaining_ClosedReturnsZero(
 	)
 }
 
-func TestCircuitBreaker_CooldownRemaining_HalfOpenReturnsZero(
+func TestCircuitBreaker_CooldownRemaining_HalfOpenReturnsCooldown(
 	t *testing.T,
 ) {
 	t.Parallel()
@@ -282,9 +282,11 @@ func TestCircuitBreaker_CooldownRemaining_HalfOpenReturnsZero(
 
 	require.True(t, cb.Allow())
 
-	assert.Equal(t, time.Duration(0),
+	// The cooldown newShortCooldownCB gives the breaker.
+	assert.Equal(t, 50*time.Millisecond,
 		cb.CooldownRemaining(),
-		"half-open circuit should have zero cooldown remaining",
+		"a delivery refused while half-open should wait "+
+			"a whole cooldown",
 	)
 }
 

internal/delivery/engine.go

@@ -362,6 +362,15 @@ func (e *Engine) start() {
 // stop cancels the worker pool's context and waits for the pool
 // to drain, bounded by the stop hook's context: a wedged worker
 // must not hang the process past fx's stop timeout.
+//
+// Once the pool has drained it closes the archive writers, so a
+// clean stop leaves no archive -wal behind. Nothing else holds a
+// writer for long by then: the archive sweeper stops before the
+// engine, and deleting a webhook only closes one. If the pool did
+// not drain in time, the writers are left open, as a kill would
+// leave them. Closing them would wait for any write in progress,
+// and a worker still running would then open new writers that
+// nothing closes, so it gains nothing over a kill.
 func (e *Engine) stop(ctx context.Context) error {
 	e.log.Info("delivery engine stopping")
 
@@ -376,6 +385,8 @@ func (e *Engine) stop(ctx context.Context) error {
 		return err
 	}
 
+	e.dbTarget.evictAll()
+
 	e.log.Info("delivery engine stopped")
 
 	return nil
@@ -438,9 +449,34 @@ func (e *Engine) processNewTask(
 		return
 	}
 
+	// Restart recovery can send and release this delivery before the
+	// receiver's Notify queues it. Ownership cannot refuse a delivery
+	// nobody holds, so the row decides whether it still needs sending.
+	row, err := e.loadDelivery(webhookDB, task.DeliveryID)
+	if err != nil {
+		e.log.Error(
+			"failed to load delivery",
+			"delivery_id", task.DeliveryID,
+			"error", err,
+		)
+
+		return
+	}
+
+	if row.Status != database.DeliveryStatusPending {
+		e.log.Info(
+			"delivery already handled, not sent again",
+			"delivery_id", task.DeliveryID,
+			"event_id", task.EventID,
+			"status", row.Status,
+		)
+
+		return
+	}
+
 	event := buildEventFromTask(task)
 
-	event, err = e.resolveEventBody(
+	event, err = e.hydrateEvent(
 		webhookDB, event, task,
 	)
 	if err != nil {
@@ -482,7 +518,7 @@ func (e *Engine) processRetryTask(
 		return
 	}
 
-	d, err := e.loadRetryDelivery(
+	d, err := e.loadDelivery(
 		webhookDB, task.DeliveryID,
 	)
 	if err != nil {
@@ -512,7 +548,7 @@ func (e *Engine) processRetryTask(
 
 	event := buildEventFromTask(task)
 
-	event, err = e.resolveEventBody(
+	event, err = e.hydrateEvent(
 		webhookDB, event, task,
 	)
 	if err != nil {
@@ -703,9 +739,7 @@ func (e *Engine) recoverSingleRetry(
 		// webhook on one bad read would be a far larger fault than
 		// the strand it is meant to clear.
 		if errors.Is(err, gorm.ErrRecordNotFound) {
-			e.failMissingTargetRetry(
+			e.failMissingTarget(webhookDB, webhookID, d)
-				webhookDB, webhookID, d,
-			)
 
 			return
 		}
@@ -1108,9 +1142,7 @@ func (e *Engine) sweepSingleRetry(
 		// Deleted is terminal, unreadable is not; see
 		// recoverSingleRetry.
 		if errors.Is(err, gorm.ErrRecordNotFound) {
-			e.failMissingTargetRetry(
+			e.failMissingTarget(webhookDB, webhookID, d)
-				webhookDB, webhookID, d,
-			)
 
 			return
 		}
@@ -1224,19 +1256,19 @@ func (e *Engine) failUnretryableRetry(
 	e.failDelivery(webhookDB, d, target.Type, reason)
 }
 
-// failMissingTargetRetry terminally fails an orphaned retrying
+// failMissingTarget terminally fails a recovered delivery, pending or
-// delivery whose target row is gone. Both restart recovery and the
+// retrying, whose target row is gone. Restart recovery and the periodic
-// periodic sweep call it, so the transition exists once.
+// sweep call it for both statuses, so the transition exists once.
 //
-// Until it existed both paths logged the failed lookup and returned,
+// Until it existed those paths logged the failed lookup and moved on,
-// which left the delivery retrying for the life of the database and
+// which left the delivery where it was for the life of the database and
 // the sweep repeating the same error every minute forever. Failing it
 // with a recorded reason is the treatment the other orphaned-retry
 // cases already get, so all of them read alike in the event log.
 //
 // Logged at warn rather than error: a deleted target is an operator
 // action, not a system fault.
-func (e *Engine) failMissingTargetRetry(
+func (e *Engine) failMissingTarget(
 	webhookDB *gorm.DB,
 	webhookID string,
 	d *database.Delivery,
@@ -1249,13 +1281,37 @@ func (e *Engine) failMissingTargetRetry(
 
 	defer e.inflight.release(d.ID)
 
+	// The batch was read before ownership was taken, and a worker may
+	// have settled the delivery and let it go in between. Only a row
+	// still in the status the batch read is failed.
+	row, err := e.loadDelivery(webhookDB, d.ID)
+	if err != nil {
+		e.log.Error(
+			"failed to load delivery",
+			"delivery_id", d.ID,
+			"error", err,
+		)
+
+		return
+	}
+
+	if row.Status != d.Status {
+		e.log.Debug(
+			"delivery already handled, not failed",
+			"delivery_id", d.ID,
+			"status", row.Status,
+		)
+
+		return
+	}
+
 	targetType, reason := e.missingTargetReason(d.TargetID)
 
 	e.log.Warn(
-		"failing orphaned retrying delivery: "+
+		"failing recovered delivery: its target no longer exists",
-			"its target no longer exists",
 		"webhook_id", webhookID,
 		"delivery_id", d.ID,
+		"status", d.Status,
 		"target_id", d.TargetID,
 		"target_type", targetType,
 	)
@@ -1289,15 +1345,14 @@ func (e *Engine) missingTargetReason(
 	if err != nil {
 		return "", fmt.Sprintf(
 			"target %s no longer exists; the delivery "+
-				"cannot be retried and has been failed "+
+				"has been failed terminally",
-				"terminally",
 			targetID,
 		)
 	}
 
 	return target.Type, fmt.Sprintf(
 		"target %q (type %s) was deleted; the delivery "+
-			"cannot be retried and has been failed terminally",
+			"has been failed terminally",
 		target.Name, target.Type,
 	)
 }
@@ -1547,6 +1602,11 @@ func truncate(s string, maxLen int) string {
 
 // --- Helper functions ---
 
+// buildEventFromTask reconstructs the event a Task describes, as far
+// as the Task itself goes. The fields it cannot fill — the body when
+// it was too large to inline, and the receipt time, which no Task
+// carries — come from the stored row in hydrateEvent, which every
+// caller of this function runs next.
 func buildEventFromTask(task *Task) database.Event {
 	event := database.Event{
 		EntrypointID: task.EntrypointID,
@@ -1574,33 +1634,71 @@ func buildTargetFromTask(task *Task) database.Target {
 	return target
 }
 
-func (e *Engine) resolveEventBody(
+// hydrateEvent fills in the event fields a Task does not carry, by
+// reading the stored event row.
+//
+// CreatedAt is the event's receipt time and lives only in that row.
+// The Slack target renders it into every message it sends, so an
+// unhydrated event puts the zero time in front of a human on every
+// notification the product delivers. See
+// https://git.eeqj.de/sneak/webhooker/issues/257.
+//
+// The body comes from the same row when the Task did not inline it,
+// which is the case for a body at or above MaxInlineBodySize.
+//
+// A read failure is fatal to the delivery only when the body depended
+// on it. When the Task inlined the body, the delivery has everything
+// it needs to be sent and goes ahead with the timestamp unset: the row
+// can be gone under a retention reap while a queued delivery still
+// holds its body, and dropping a deliverable event to protect one
+// metadata field would be a worse failure than the one it prevents.
+func (e *Engine) hydrateEvent(
 	webhookDB *gorm.DB,
 	event database.Event,
 	task *Task,
 ) (database.Event, error) {
-	if task.Body != nil {
+	columns := []string{"created_at"}
-		event.Body = *task.Body
 
-		return event, nil
+	if task.Body == nil {
+		columns = append(columns, "body")
 	}
 
 	var dbEvent database.Event
 
-	err := webhookDB.Select("body").
+	err := webhookDB.Select(columns).
 		First(&dbEvent, "id = ?", task.EventID).Error
 	if err != nil {
+		if task.Body == nil {
 			return event, fmt.Errorf(
 				"fetching event body: %w", err,
 			)
 		}
 
-	event.Body = dbEvent.Body
+		e.log.Warn(
+			"could not read the stored event; delivering "+
+				"the inlined body without its receipt time",
+			"event_id", task.EventID,
+			"delivery_id", task.DeliveryID,
+			"error", err,
+		)
+
+		event.Body = *task.Body
 
 		return event, nil
 	}
 
-func (e *Engine) loadRetryDelivery(
+	event.CreatedAt = dbEvent.CreatedAt
+
+	if task.Body != nil {
+		event.Body = *task.Body
+	} else {
+		event.Body = dbEvent.Body
+	}
+
+	return event, nil
+}
+
+func (e *Engine) loadDelivery(
 	webhookDB *gorm.DB, deliveryID string,
 ) (*database.Delivery, error) {
 	var d database.Delivery
@@ -1953,14 +2051,31 @@ func (e *Engine) sendRecoveredDeliveries(
 
 		target, ok := targetMap[deliveries[i].TargetID]
 		if !ok {
+			// A missing entry does not mean the target is gone: the
+			// map is also empty when its query failed. Only a lookup
+			// that finds no row ends the delivery; any other error
+			// leaves it pending for the next sweep. See
+			// recoverSingleRetry.
+			var err error
+
+			target, err = e.loadTarget(deliveries[i].TargetID)
+			if errors.Is(err, gorm.ErrRecordNotFound) {
+				e.failMissingTarget(webhookDB, webhookID, &deliveries[i])
+
+				continue
+			}
+
+			if err != nil {
 				e.log.Error(
-				"target not found for delivery",
+					"failed to load target for recovered delivery",
 					"delivery_id", deliveries[i].ID,
 					"target_id", deliveries[i].TargetID,
+					"error", err,
 				)
 
 				continue
 			}
+		}
 
 		if !e.takeForRedispatch(
 			webhookDB, deliveries[i].ID,

internal/delivery/engine_lifecycle_test.go

@@ -2,6 +2,8 @@ package delivery_test
 
 import (
 	"context"
+	"fmt"
+	"path/filepath"
 	"testing"
 	"time"
 
@@ -269,3 +271,88 @@ func TestEngine_StopHookHonoursStopTimeout(t *testing.T) {
 
 	requireStopHookExpires(t, lc.hooks[0], "delivery engine")
 }
+
+// deliverToArchive runs one delivery to a database target through
+// the running engine and returns the webhook's archive file path.
+// The archive writer holds the file open afterwards.
+func deliverToArchive(t *testing.T, s iSetup) string {
+	t.Helper()
+
+	deliveryID, task := seedLogTask(t, s)
+	task.TargetType = database.TargetTypeDatabase
+
+	s.Engine.Notify([]delivery.Task{task})
+
+	iWaitForDelivered(t, s.WebhookDB, deliveryID)
+
+	return filepath.Join(
+		filepath.Dir(s.DBMgr.DBPath(s.WebhookID)),
+		fmt.Sprintf("archive-%s.db", s.WebhookID),
+	)
+}
+
+// TestEngine_StopHookClosesArchives is the regression test for an
+// archive split across two files by a clean stop. The engine never
+// closed its archive writers, so after a stop the archived rows
+// could sit in archive-{id}.db-wal while archive-{id}.db held no
+// table at all, and copying the .db on its own gave an empty
+// database.
+func TestEngine_StopHookClosesArchives(t *testing.T) {
+	t.Parallel()
+
+	s := newISetup(t)
+
+	lc := startEngineViaHook(t, s.Engine)
+
+	path := deliverToArchive(t, s)
+	require.FileExists(
+		t, path+"-wal",
+		"an open archive should have a -wal for the stop to remove",
+	)
+
+	require.NoError(t, lc.hooks[0].OnStop(context.Background()))
+
+	wals, err := filepath.Glob(
+		filepath.Join(filepath.Dir(path), "archive-*.db-wal"),
+	)
+	require.NoError(t, err)
+	require.Empty(
+		t, wals, "a clean stop must leave no archive -wal behind",
+	)
+
+	// With no -wal beside it, the row can only be in the .db.
+	count, err := countArchivedRows(path)
+	require.NoError(t, err)
+	require.Equal(t, int64(1), count)
+}
+
+// TestEngine_StopHookTimeoutLeavesArchivesOpen covers a stop whose
+// budget runs out while a worker is still running. The archive
+// writers are left open, as a kill would leave them: closing them
+// would wait for any write in progress, and that worker would then
+// open new writers that nothing closes.
+func TestEngine_StopHookTimeoutLeavesArchivesOpen(t *testing.T) {
+	t.Parallel()
+
+	s := newISetup(t)
+
+	lc := startEngineViaHook(t, s.Engine)
+
+	deliverToArchive(t, s)
+
+	release := make(chan struct{})
+
+	t.Cleanup(func() {
+		close(release)
+		s.Engine.EvictWebhook(s.WebhookID)
+	})
+
+	s.Engine.ExportWedgeWorker(release)
+
+	requireStopHookExpires(t, lc.hooks[0], "delivery engine")
+
+	require.True(
+		t, s.Engine.ExportArchiveHandleOpen(s.WebhookID),
+		"a stop that timed out must not close archive writers",
+	)
+}

internal/delivery/engine_test.go

@@ -17,6 +17,7 @@ import (
 	"time"
 
 	"github.com/google/uuid"
+	"github.com/prometheus/client_golang/prometheus"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 	"gorm.io/driver/sqlite"
@@ -24,6 +25,7 @@ import (
 	_ "modernc.org/sqlite"
 	"sneak.berlin/go/webhooker/internal/database"
 	"sneak.berlin/go/webhooker/internal/delivery"
+	"sneak.berlin/go/webhooker/internal/metrics"
 )
 
 // testContentType is the event content type used in tests.
@@ -894,6 +896,100 @@ func TestDeliverHTTP_CircuitBreakerBlocks(t *testing.T) {
 	)
 }
 
+// recordingScheduler keeps the delay of every retry it is asked to
+// schedule, and schedules nothing.
+type recordingScheduler struct {
+	delays []time.Duration
+}
+
+func (s *recordingScheduler) ScheduleRetry(
+	_ delivery.Task, delay time.Duration,
+) {
+	s.delays = append(s.delays, delay)
+}
+
+// TestDeliverHTTP_HalfOpenBreakerDelaysQueuedTasks proves that while a
+// half-open breaker's one probe delivery is in flight, every other task
+// for the target is put back with a whole cooldown as its delay rather
+// than none, and that its status is written the first time the breaker
+// turns it away and not on each pass after that.
+func TestDeliverHTTP_HalfOpenBreakerDelaysQueuedTasks(t *testing.T) {
+	t.Parallel()
+
+	db := testWebhookDB(t)
+	e := testEngine(t, 1)
+
+	// Every write of retrying moves the retry counter, so on a registry
+	// this test owns the counter is the number of those writes.
+	reg := prometheus.NewRegistry()
+	e.ExportSetMetrics(metrics.New(reg))
+
+	targetID := uuid.New().String()
+	cb := newShortCooldownCB(t)
+	e.ExportSetCircuitBreaker(targetID, cb)
+
+	for range delivery.ExportDefaultFailureThreshold {
+		cb.RecordFailure()
+	}
+
+	time.Sleep(60 * time.Millisecond)
+
+	require.True(t, cb.Allow(), "the probe delivery should go through")
+	require.Equal(t, delivery.CircuitHalfOpen, cb.State())
+
+	cfg := newHTTPTargetConfig(
+		"http://will-not-be-called.invalid",
+	)
+	sched := &recordingScheduler{}
+
+	const queued, passes = 3, 4
+
+	for range queued {
+		event := seedEvent(t, db, `{"cb":"half-open"}`)
+		dlv := seedDelivery(
+			t, db, event.ID, targetID,
+			database.DeliveryStatusPending,
+		)
+
+		for range passes {
+			// Each pass starts from the stored row, as a retry does.
+			var row database.Delivery
+
+			require.NoError(t, db.First(
+				&row, "id = ?", dlv.ID,
+			).Error)
+
+			fix := buildHTTPFixture(
+				row, event, targetID,
+				"test-cb-half-open", cfg, 5, 1,
+			)
+
+			e.ExportDeliverHTTPWithScheduler(
+				context.TODO(), db, fix.Delivery, fix.Task, sched,
+			)
+		}
+
+		assertDeliveryStatus(t, db, dlv.ID,
+			database.DeliveryStatusRetrying,
+		)
+	}
+
+	require.Len(t, sched.delays, queued*passes)
+
+	for _, delay := range sched.delays {
+		// The cooldown newShortCooldownCB gives the breaker.
+		assert.Equal(t, 50*time.Millisecond, delay,
+			"a task turned away while half-open should wait "+
+				"a whole cooldown",
+		)
+	}
+
+	assert.InDelta(t, float64(queued),
+		mCounter(t, reg, mRetries, mTypeHTTP), 0,
+		"status should be written once per task, not once per pass",
+	)
+}
+
 func TestGetCircuitBreaker_CreatesOnDemand(t *testing.T) {
 	t.Parallel()
 
@@ -1070,6 +1166,10 @@ func TestIsForwardableHeader(t *testing.T) {
 	assert.False(t,
 		delivery.ExportIsForwardableHeader("Content-Length"),
 	)
+
+	assert.False(t,
+		delivery.ExportIsForwardableHeader("Content-Type"),
+	)
 }
 
 func TestTruncate(t *testing.T) {
@@ -1151,6 +1251,81 @@ func TestDoHTTPRequest_ForwardsHeaders(t *testing.T) {
 	)
 }
 
+// The event's stored inbound headers carry the same Content-Type the
+// receiver saved as the event's ContentType, so a delivery could send
+// it twice. It must go out exactly once, with a Content-Type configured
+// on the target winning, then the event's ContentType.
+func TestApplyRequestHeaders_SendsOneContentType(t *testing.T) {
+	t.Parallel()
+
+	cases := map[string]struct {
+		inbound    string
+		event      string
+		configured string
+		want       []string
+	}{
+		"inbound and event agree": {
+			inbound: testContentType,
+			event:   testContentType,
+			want:    []string{testContentType},
+		},
+		"inbound and event disagree": {
+			inbound: "text/plain",
+			event:   testContentType,
+			want:    []string{testContentType},
+		},
+		"event has none": {
+			inbound: testContentType,
+			want:    nil,
+		},
+		"target configures its own": {
+			inbound:    testContentType,
+			event:      testContentType,
+			configured: "application/xml",
+			want:       []string{"application/xml"},
+		},
+	}
+
+	for name, tc := range cases {
+		t.Run(name, func(t *testing.T) {
+			t.Parallel()
+
+			inbound, err := json.Marshal(map[string][]string{
+				headerContentType: {tc.inbound},
+			})
+			require.NoError(t, err)
+
+			cfg := &delivery.HTTPTargetConfig{}
+			if tc.configured != "" {
+				cfg.Headers = map[string]string{
+					headerContentType: tc.configured,
+				}
+			}
+
+			req, err := http.NewRequestWithContext(
+				context.Background(),
+				http.MethodPost,
+				"https://target.example.com/hook",
+				http.NoBody,
+			)
+			require.NoError(t, err)
+
+			delivery.ExportApplyRequestHeaders(
+				req,
+				&database.Event{
+					Headers:     string(inbound),
+					ContentType: tc.event,
+				},
+				cfg,
+			)
+
+			assert.Equal(t,
+				tc.want, req.Header.Values(headerContentType),
+			)
+		})
+	}
+}
+
 func TestProcessDelivery_RoutesToCorrectHandler(
 	t *testing.T,
 ) {

internal/delivery/event_timestamp_test.go

@@ -0,0 +1,442 @@
+package delivery_test
+
+import (
+	"context"
+	"encoding/json"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+	"time"
+
+	"github.com/google/uuid"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"gorm.io/gorm"
+	"sneak.berlin/go/webhooker/internal/database"
+	"sneak.berlin/go/webhooker/internal/delivery"
+)
+
+// tsEventCreatedAt is the receipt time seeded on the events these
+// tests deliver. It is far enough from both the zero time and from
+// now that neither can be mistaken for it.
+func tsEventCreatedAt() time.Time {
+	return time.Date(
+		2026, time.March, 4, 5, 6, 7, 0, time.UTC,
+	)
+}
+
+// tsZeroStamp is what a Slack message renders when the event handed
+// to FormatSlackMessage carries no CreatedAt.
+const tsZeroStamp = "*Timestamp:* `0001-01-01T00:00:00Z`"
+
+// tsEventBody is the body seeded on every event in this file. It is
+// small enough that a Task can inline it.
+const tsEventBody = `{"hello":"world"}`
+
+// tsUndeliverableHook stands in for a Slack incoming webhook on the
+// tests that never send: the config parser requires a URL, but no
+// request is made.
+const tsUndeliverableHook = "https://hooks.slack.com/services/T/B/x"
+
+// tsSink is a stand-in Slack incoming webhook that records the raw
+// body posted to it.
+type tsSink struct {
+	*httptest.Server
+
+	bodies chan []byte
+}
+
+func newTSSink(t *testing.T) *tsSink {
+	t.Helper()
+
+	s := &tsSink{bodies: make(chan []byte, 8)}
+
+	s.Server = httptest.NewServer(http.HandlerFunc(
+		func(w http.ResponseWriter, r *http.Request) {
+			body, _ := io.ReadAll(r.Body)
+
+			select {
+			case s.bodies <- body:
+			default:
+			}
+
+			w.WriteHeader(http.StatusOK)
+		},
+	))
+
+	t.Cleanup(s.Close)
+
+	return s
+}
+
+// text returns the Slack message text from the single payload the
+// sink received.
+func (s *tsSink) text(t *testing.T) string {
+	t.Helper()
+
+	select {
+	case raw := <-s.bodies:
+		t.Logf("raw slack payload: %s", raw)
+
+		var payload struct {
+			Text string `json:"text"`
+		}
+
+		require.NoError(t, json.Unmarshal(raw, &payload))
+
+		return payload.Text
+	case <-time.After(5 * time.Second):
+		t.Fatal("slack sink received no payload")
+
+		return ""
+	}
+}
+
+func tsSlackConfig(t *testing.T, url string) string {
+	t.Helper()
+
+	data, err := json.Marshal(
+		delivery.SlackTargetConfig{WebhookURL: url},
+	)
+	require.NoError(t, err)
+
+	return string(data)
+}
+
+// tsSeedEvent writes an event whose CreatedAt is tsEventCreatedAt
+// rather than the write time, so an assertion on the rendered
+// timestamp cannot pass by accident against "roughly now".
+func tsSeedEvent(
+	t *testing.T, db *gorm.DB, webhookID string,
+) database.Event {
+	t.Helper()
+
+	event := database.Event{
+		WebhookID:    webhookID,
+		EntrypointID: uuid.New().String(),
+		Method:       http.MethodPost,
+		Headers:      `{}`,
+		Body:         tsEventBody,
+		ContentType:  "application/json",
+	}
+	event.ID = uuid.New().String()
+	event.CreatedAt = tsEventCreatedAt()
+	event.UpdatedAt = tsEventCreatedAt()
+
+	require.NoError(t, db.Create(&event).Error)
+
+	var stored database.Event
+
+	require.NoError(t,
+		db.First(&stored, "id = ?", event.ID).Error,
+	)
+	require.Equal(t,
+		tsEventCreatedAt().UTC(), stored.CreatedAt.UTC(),
+		"seeded created_at did not round-trip",
+	)
+
+	return event
+}
+
+// tsSeedTarget writes the slack target row into the main database.
+// The retry path confirms the target still exists before sending.
+func tsSeedTarget(
+	t *testing.T, mainDB *gorm.DB, webhookID, config string,
+) database.Target {
+	t.Helper()
+
+	target := database.Target{
+		WebhookID: webhookID,
+		Name:      "slack-sink",
+		Type:      database.TargetTypeSlack,
+		Config:    config,
+		Active:    true,
+	}
+
+	require.NoError(t, mainDB.Create(&target).Error)
+
+	return target
+}
+
+func tsTask(
+	d database.Delivery,
+	event database.Event,
+	webhookID string,
+	target database.Target,
+	attemptNum int,
+	body *string,
+) delivery.Task {
+	return delivery.Task{
+		DeliveryID:   d.ID,
+		EventID:      event.ID,
+		WebhookID:    webhookID,
+		EntrypointID: event.EntrypointID,
+		TargetID:     target.ID,
+		TargetName:   target.Name,
+		TargetType:   database.TargetTypeSlack,
+		TargetConfig: target.Config,
+		MaxRetries:   0,
+		Method:       event.Method,
+		Headers:      event.Headers,
+		ContentType:  event.ContentType,
+		Body:         body,
+		AttemptNum:   attemptNum,
+	}
+}
+
+func tsAssertRealTimestamp(t *testing.T, text string) {
+	t.Helper()
+
+	assert.NotContains(t, text, tsZeroStamp,
+		"slack message carries the zero timestamp",
+	)
+	assert.Contains(t, text,
+		"*Timestamp:* `"+
+			tsEventCreatedAt().UTC().Format(time.RFC3339)+"`",
+		"slack message does not carry the event's receipt time",
+	)
+}
+
+// tsCase is one end-to-end delivery of a seeded event to a slack
+// sink, over whichever engine path `process` names.
+type tsCase struct {
+	// status is the delivery row's status before the engine runs.
+	// The retry path refuses a delivery that is not retrying.
+	status database.DeliveryStatus
+
+	// inlineBody mirrors a Task built for a body under
+	// MaxInlineBodySize. When false the engine reads the body back
+	// from the stored row.
+	inlineBody bool
+
+	attemptNum int
+
+	process func(
+		ctx context.Context, e *delivery.Engine, task *delivery.Task,
+	)
+}
+
+// run delivers one event through the named path and returns the
+// Slack message text the sink received.
+func (c tsCase) run(t *testing.T) (iSetup, database.Delivery, string) {
+	t.Helper()
+
+	s := newISetup(t)
+	sink := newTSSink(t)
+
+	cfg := tsSlackConfig(t, sink.URL)
+	target := tsSeedTarget(t, s.MainDB, s.WebhookID, cfg)
+	event := tsSeedEvent(t, s.WebhookDB, s.WebhookID)
+
+	d := iSeedDelivery(
+		t, s.WebhookDB, event.ID, target.ID, c.status,
+	)
+
+	var body *string
+
+	if c.inlineBody {
+		bodyStr := event.Body
+		body = &bodyStr
+	}
+
+	task := tsTask(
+		d, event, s.WebhookID, target, c.attemptNum, body,
+	)
+
+	c.process(context.TODO(), s.Engine, &task)
+
+	return s, d, sink.text(t)
+}
+
+// TestSlackFirstAttemptCarriesEventTimestamp covers the path an
+// event takes on its first delivery: the task comes from the
+// receiver and the engine reconstructs the event from it.
+func TestSlackFirstAttemptCarriesEventTimestamp(t *testing.T) {
+	t.Parallel()
+
+	s, d, text := tsCase{
+		status:     database.DeliveryStatusPending,
+		inlineBody: true,
+		attemptNum: 1,
+		process: func(
+			ctx context.Context,
+			e *delivery.Engine,
+			task *delivery.Task,
+		) {
+			e.ExportProcessNewTask(ctx, task)
+		},
+	}.run(t)
+
+	tsAssertRealTimestamp(t, text)
+
+	iAssertStatus(t, s.WebhookDB, d.ID,
+		database.DeliveryStatusDelivered,
+	)
+}
+
+// TestSlackFirstAttemptLargeBodyCarriesEventTimestamp covers the
+// first-attempt path for an event whose body exceeded
+// MaxInlineBodySize, so the task carries no body and the engine
+// reads it back from the stored row.
+func TestSlackFirstAttemptLargeBodyCarriesEventTimestamp(
+	t *testing.T,
+) {
+	t.Parallel()
+
+	_, _, text := tsCase{
+		status:     database.DeliveryStatusPending,
+		inlineBody: false,
+		attemptNum: 1,
+		process: func(
+			ctx context.Context,
+			e *delivery.Engine,
+			task *delivery.Task,
+		) {
+			e.ExportProcessNewTask(ctx, task)
+		},
+	}.run(t)
+
+	tsAssertRealTimestamp(t, text)
+}
+
+// TestSlackRetryCarriesEventTimestamp covers the retry path, which
+// reconstructs the event from the same task the first attempt used.
+func TestSlackRetryCarriesEventTimestamp(t *testing.T) {
+	t.Parallel()
+
+	s, d, text := tsCase{
+		status:     database.DeliveryStatusRetrying,
+		inlineBody: true,
+		attemptNum: 2,
+		process: func(
+			ctx context.Context,
+			e *delivery.Engine,
+			task *delivery.Task,
+		) {
+			e.ExportProcessRetryTask(ctx, task)
+		},
+	}.run(t)
+
+	tsAssertRealTimestamp(t, text)
+
+	iAssertStatus(t, s.WebhookDB, d.ID,
+		database.DeliveryStatusDelivered,
+	)
+}
+
+// TestFormatSlackMessageOverTaskReconstructedEvent asserts on the
+// formatted message directly, over the event the delivery paths
+// reconstruct from a Task. It is the unit-level guard under the
+// end-to-end tests: revert the CreatedAt population in hydrateEvent
+// and this fails on the zero timestamp.
+func TestFormatSlackMessageOverTaskReconstructedEvent(
+	t *testing.T,
+) {
+	t.Parallel()
+
+	s := newISetup(t)
+
+	cfg := tsSlackConfig(t, tsUndeliverableHook)
+	target := tsSeedTarget(t, s.MainDB, s.WebhookID, cfg)
+	event := tsSeedEvent(t, s.WebhookDB, s.WebhookID)
+
+	d := iSeedDelivery(
+		t, s.WebhookDB, event.ID, target.ID,
+		database.DeliveryStatusPending,
+	)
+
+	bodyStr := event.Body
+	task := tsTask(d, event, s.WebhookID, target, 1, &bodyStr)
+
+	rebuilt, err := s.Engine.ExportEventForTask(
+		s.WebhookDB, &task,
+	)
+	require.NoError(t, err)
+	assert.False(t, rebuilt.CreatedAt.IsZero(),
+		"reconstructed event carries the zero time",
+	)
+	assert.Equal(t,
+		tsEventCreatedAt().UTC(), rebuilt.CreatedAt.UTC(),
+	)
+
+	tsAssertRealTimestamp(
+		t, delivery.FormatSlackMessage(&rebuilt),
+	)
+}
+
+// TestFormatSlackMessageZeroTimestamp asserts the rendering choice
+// directly, without going through the engine: a zero CreatedAt (the
+// shape a reaped-row fallback produces) renders as "unknown" rather
+// than the year-1 zero time, while a real CreatedAt still renders as
+// RFC3339.
+func TestFormatSlackMessageZeroTimestamp(t *testing.T) {
+	t.Parallel()
+
+	zeroEvent := database.Event{
+		Method:      http.MethodPost,
+		ContentType: testContentType,
+		Body:        tsEventBody,
+	}
+
+	zeroText := delivery.FormatSlackMessage(&zeroEvent)
+
+	assert.NotContains(t, zeroText, "0001-01-01",
+		"slack message carries the zero-time year",
+	)
+	assert.Contains(t, zeroText, "*Timestamp:* `unknown`",
+		"slack message does not mark an unset receipt time as unknown",
+	)
+
+	nonZeroEvent := zeroEvent
+	nonZeroEvent.CreatedAt = tsEventCreatedAt()
+
+	nonZeroText := delivery.FormatSlackMessage(&nonZeroEvent)
+
+	assert.Contains(t, nonZeroText,
+		"*Timestamp:* `"+
+			tsEventCreatedAt().UTC().Format(time.RFC3339)+"`",
+		"slack message does not render a real receipt time as RFC3339",
+	)
+}
+
+// TestEventReconstructionSurvivesAReapedRow pins the fallback: an
+// event row reaped by retention while its delivery still holds the
+// body inline is still delivered, with the receipt time unset,
+// rather than dropped.
+func TestEventReconstructionSurvivesAReapedRow(t *testing.T) {
+	t.Parallel()
+
+	s := newISetup(t)
+
+	cfg := tsSlackConfig(t, tsUndeliverableHook)
+	target := tsSeedTarget(t, s.MainDB, s.WebhookID, cfg)
+	event := tsSeedEvent(t, s.WebhookDB, s.WebhookID)
+
+	d := iSeedDelivery(
+		t, s.WebhookDB, event.ID, target.ID,
+		database.DeliveryStatusPending,
+	)
+
+	bodyStr := event.Body
+	task := tsTask(d, event, s.WebhookID, target, 1, &bodyStr)
+
+	require.NoError(t, s.WebhookDB.Unscoped().Delete(
+		&database.Event{}, "id = ?", event.ID,
+	).Error)
+
+	rebuilt, err := s.Engine.ExportEventForTask(
+		s.WebhookDB, &task,
+	)
+	require.NoError(t, err)
+	assert.Equal(t, bodyStr, rebuilt.Body)
+	assert.True(t, rebuilt.CreatedAt.IsZero())
+
+	// A task with no inlined body has nothing left to deliver, so
+	// the same reaped row is an error there.
+	noBody := task
+	noBody.Body = nil
+
+	_, err = s.Engine.ExportEventForTask(s.WebhookDB, &noBody)
+	require.Error(t, err)
+}

internal/delivery/export_test.go

@@ -101,6 +101,19 @@ func (e *Engine) ExportDeliverHTTP(
 	e.httpTarget.Deliver(ctx, webhookDB, d, task, e)
 }
 
+// ExportDeliverHTTPWithScheduler delivers via the http target, handing
+// any retry to sched instead of the engine, so a test can see the
+// delay each retry is given.
+func (e *Engine) ExportDeliverHTTPWithScheduler(
+	ctx context.Context,
+	webhookDB *gorm.DB,
+	d *database.Delivery,
+	task *Task,
+	sched Scheduler,
+) {
+	e.httpTarget.Deliver(ctx, webhookDB, d, task, sched)
+}
+
 // ExportDeliverDatabase delivers via the database target.
 func (e *Engine) ExportDeliverDatabase(
 	webhookDB *gorm.DB, d *database.Delivery,
@@ -151,6 +164,16 @@ func (e *Engine) ExportProcessRetryTask(
 	e.processRetryTask(ctx, task)
 }
 
+// ExportEventForTask exposes the event reconstruction the delivery
+// paths run: buildEventFromTask followed by hydrateEvent.
+func (e *Engine) ExportEventForTask(
+	webhookDB *gorm.DB, task *Task,
+) (database.Event, error) {
+	return e.hydrateEvent(
+		webhookDB, buildEventFromTask(task), task,
+	)
+}
+
 // ExportProcessDelivery exposes processDelivery.
 func (e *Engine) ExportProcessDelivery(
 	ctx context.Context,
@@ -169,6 +192,14 @@ func (e *Engine) ExportGetCircuitBreaker(
 	return e.httpTarget.getCircuitBreaker(targetID)
 }
 
+// ExportSetCircuitBreaker makes cb the http target's circuit breaker
+// for targetID, so a test can use one with a short cooldown.
+func (e *Engine) ExportSetCircuitBreaker(
+	targetID string, cb *CircuitBreaker,
+) {
+	e.httpTarget.circuitBreakers.Store(targetID, cb)
+}
+
 // ExportParseHTTPConfig exposes parseHTTPConfig.
 func (e *Engine) ExportParseHTTPConfig(
 	configJSON string,
@@ -311,6 +342,31 @@ func (e *Engine) ExportRecoverRetryingDeliveries(
 	e.recoverRetryingDeliveries(webhookDB, webhookID)
 }
 
+// ExportFailMissingTarget exposes failMissingTarget, so a test can hand
+// it a delivery as a batch read it earlier.
+func (e *Engine) ExportFailMissingTarget(
+	webhookDB *gorm.DB,
+	webhookID string,
+	d *database.Delivery,
+) {
+	e.failMissingTarget(webhookDB, webhookID, d)
+}
+
+// ExportSendRecoveredDeliveries exposes sendRecoveredDeliveries, so a
+// test can hand it a target map that lacks a delivery's target.
+func (e *Engine) ExportSendRecoveredDeliveries(
+	ctx context.Context,
+	webhookDB *gorm.DB,
+	deliveries []database.Delivery,
+	webhookID string,
+	targetMap map[string]database.Target,
+	settled map[string]struct{},
+) {
+	e.sendRecoveredDeliveries(
+		ctx, webhookDB, deliveries, webhookID, targetMap, settled,
+	)
+}
+
 // ExportDeliveryCh returns the delivery channel.
 func (e *Engine) ExportDeliveryCh() chan Task {
 	return e.deliveryCh

internal/delivery/inflight_test.go

@@ -273,6 +273,73 @@ func TestOwnershipIsReleasedAfterDelivery(t *testing.T) {
 	)
 }
 
+// TestNotifyAfterRecoveryDoesNotSendAgain is the startup race of
+// https://git.eeqj.de/sneak/webhooker/issues/299. The receiver has
+// written a delivery, restart recovery finds it pending, sends it and
+// releases it, and only then does the receiver's Notify for it arrive.
+// Nothing owns the delivery by then, so Notify takes it.
+func TestNotifyAfterRecoveryDoesNotSendAgain(t *testing.T) {
+	t.Parallel()
+
+	targetID := uuid.New().String()
+	s := fSweepSetup(t, targetID, "recovered")
+
+	event := iSeedEvent(
+		t, s.WebhookDB, s.WebhookID, `{"recovered":true}`,
+	)
+
+	d := iSeedDelivery(
+		t, s.WebhookDB, event.ID, targetID,
+		database.DeliveryStatusPending,
+	)
+
+	s.Engine.ExportStart()
+
+	defer func() {
+		require.NoError(
+			t, s.Engine.ExportStop(context.Background()),
+		)
+	}()
+
+	// Restart recovery sends the delivery and lets it go.
+	iWaitForDelivered(t, s.WebhookDB, d.ID)
+	require.Eventually(
+		t,
+		func() bool {
+			return s.Engine.ExportInflightHeld() == 0
+		},
+		5*time.Second, 20*time.Millisecond,
+	)
+
+	body := event.Body
+
+	s.Engine.Notify([]delivery.Task{{
+		DeliveryID:   d.ID,
+		EventID:      event.ID,
+		WebhookID:    s.WebhookID,
+		TargetID:     targetID,
+		TargetName:   "recovered",
+		TargetType:   database.TargetTypeLog,
+		Body:         &body,
+		EntrypointID: event.EntrypointID,
+	}})
+
+	// Notify took the delivery, and a worker releases it once it has
+	// run the task.
+	require.Eventually(
+		t,
+		func() bool {
+			return s.Engine.ExportInflightHeld() == 0
+		},
+		5*time.Second, 20*time.Millisecond,
+	)
+
+	assert.Len(
+		t, iResults(t, s.WebhookDB, d.ID), 1,
+		"the delivery was sent a second time",
+	)
+}
+
 // TestRetryingRecoverySkipsASuccessfulResult is the retrying-side twin
 // of the pending reconcile. A second attempt that reached the receiver
 // and whose status write then failed sits at retrying holding a

internal/delivery/metrics_test.go

@@ -412,9 +412,10 @@ func TestDeliveryMetrics_BreakerBlockedIsNotAnAttempt(
 
 	s.Engine.ExportProcessRetryTask(context.TODO(), &blocked)
 
-	// The breaker refused it: rescheduled, so the retry counter
+	// The breaker refused it: rescheduled without rewriting the
-	// moved, but nothing was attempted or timed.
+	// retrying status it already had, so the retry counter did not
-	assert.InDelta(t, retriesBefore+1,
+	// move, and nothing was attempted or timed.
+	assert.InDelta(t, retriesBefore,
 		mCounter(t, reg, mRetries, mTypeHTTP), 0)
 	assert.InDelta(t, threshold,
 		mCounter(t, reg, mAttempts, mTypeHTTP), 0)

internal/delivery/redirect_test.go

@@ -170,7 +170,8 @@ func TestDelivery_CrossOriginRedirectDropsOriginScopedHeaders(
 // Stripping must not fire within the configured origin, or every
 // destination that redirects its own path would lose its
 // credential and start answering 401 — and would lose the inbound
-// signature the receiver verifies.
+// signature header the target endpoint verifies. webhooker's own
+// receiver verifies no signature; it only forwards the header.
 func TestDelivery_SameOriginRedirectKeepsOriginScopedHeaders(
 	t *testing.T,
 ) {
@@ -338,10 +339,11 @@ func TestRedirectPolicy_StopsAtHopCap(t *testing.T) {
 // The set the redirect policy strips is whatever the delivery path
 // actually put on the wire, so a header added to the forward set is
 // covered without a second edit. A header the event never carried
-// is not in the set, and the delivery path's own two are deliberately
+// is not in the set, and neither is the inbound Content-Type, because
-// excluded: Content-Type describes the body, which a 307 carries
+// it is not forwarded. Two more are deliberately excluded: a
-// across hosts, and the inbound User-Agent every real sender supplies
+// Content-Type configured on the target describes the body, which a
-// is overwritten before the request goes out.
+// 307 carries across hosts, and the inbound User-Agent every real
+// sender supplies is overwritten before the request goes out.
 func TestApplyRequestHeaders_ReportsOriginScopedNames(t *testing.T) {
 	t.Parallel()
 
@@ -370,6 +372,7 @@ func TestApplyRequestHeaders_ReportsOriginScopedNames(t *testing.T) {
 		&delivery.HTTPTargetConfig{
 			Headers: map[string]string{
 				probeHeaderName: probeHeaderValue,
+				"Content-Type":  testContentType,
 			},
 		},
 	)
@@ -377,7 +380,11 @@ func TestApplyRequestHeaders_ReportsOriginScopedNames(t *testing.T) {
 	assert.Equal(t,
 		[]string{probeHeaderName, inboundHeaderName}, names,
 		"both header classes are reported, and only those: "+
-			"Host is never forwarded, Content-Type and "+
+			"Host and the inbound Content-Type are never "+
-			"User-Agent are the delivery path's own",
+			"forwarded, User-Agent is the delivery path's own",
+	)
+	assert.NotContains(t, names, "Content-Type",
+		"a Content-Type configured on the target must survive "+
+			"a cross-origin 307/308 with the body it describes",
 	)
 }

internal/delivery/ssrf.go

@@ -26,7 +26,7 @@ var (
 		"hostname resolved to no IP addresses",
 	)
 	errBlockedIP = errors.New(
-		"blocked private/reserved IP range",
+		"blocked private, reserved or cloud metadata address",
 	)
 	errBlockedMetadata = errors.New(
 		"blocked link-local or cloud instance metadata " +
@@ -37,9 +37,10 @@ var (
 	)
 )
 
-// blockedNetworks contains all private/reserved IP ranges
+// blockedNetworks is the default blocklist: the private and
-// that should be blocked to prevent SSRF attacks. An operator
+// reserved IP ranges, plus the public cloud metadata addresses,
-// can permit specific blocks out of this set with
+// that are blocked to prevent SSRF attacks. An operator can
+// permit specific blocks out of this set with
 // ALLOWED_EGRESS_CIDRS; see Guard.
 //
 //nolint:gochecknoglobals // package-level network list is appropriate here
@@ -122,6 +123,8 @@ func init() {
 		"::1/128",
 		"fc00::/7",
 		"fe80::/10",
+		// Azure WireServer, a public address that serves VM credentials.
+		"168.63.129.16/32",
 	})
 
 	// Every entry is named. The set must not grow or shrink
@@ -216,8 +219,8 @@ func matchesAny(networks []*net.IPNet, ip net.IP) bool {
 }
 
 // isBlockedIP checks whether an IP address falls within
-// any blocked private/reserved network range, before any
+// the default blocklist, before any operator allowlist is
-// operator allowlist is considered.
+// considered.
 func isBlockedIP(ip net.IP) bool {
 	return matchesAny(blockedNetworks, ip)
 }
@@ -320,7 +323,7 @@ func (g *Guard) allows(ip net.IP) bool {
 //
 //  1. alwaysBlockedNetworks is refused before the allowlist is
 //     consulted, so no configured CIDR reaches link-local or a
-//     cloud instance metadata endpoint.
+//     cloud metadata endpoint at a non-public address.
 //  2. The allowlist is consulted next, so a listed private
 //     network becomes reachable.
 //  3. Everything else keeps the default blocklist's answer.

internal/delivery/ssrf_allowlist_test.go

@@ -390,6 +390,41 @@ func TestGuardAllowlist_PublicUnaffected(t *testing.T) {
 	}
 }
 
+// TestGuardAllowlist_AzureWireServerReopenable covers Azure's
+// WireServer, a public address that serves VM credentials. The
+// default guard refuses it, but because it is public it sits in
+// the default blocklist rather than the unconditional set, so an
+// operator who lists it can reach it.
+func TestGuardAllowlist_AzureWireServerReopenable(t *testing.T) {
+	t.Parallel()
+
+	const wireServerIP = "168.63.129.16"
+
+	target := "http://" + wireServerIP + "/?comp=versions"
+
+	defaultGuard := delivery.NewTestGuard()
+
+	err := defaultGuard.ValidateTargetURL(context.Background(), target)
+	require.Error(t, err,
+		"WireServer must be refused with no allowlist set",
+	)
+	assert.NotContains(t, err.Error(), metadataRefusalClause,
+		"WireServer must be refused by the default blocklist, "+
+			"which an allowlist can override",
+	)
+
+	assertDialRefused(t, defaultGuard, target)
+
+	listed := delivery.NewTestGuard(
+		netip.MustParsePrefix(wireServerIP + "/32"),
+	)
+
+	assert.NoError(t,
+		listed.ValidateTargetURL(context.Background(), target),
+		"an operator who lists WireServer must be able to reach it",
+	)
+}
+
 // TestGuardCheckIP_BothPathsShareOneDecision asserts that the
 // validator and the dialer are not two policies that happen to
 // agree: both are defined in terms of checkIP, so the exported

internal/delivery/target_database.go

@@ -277,6 +277,24 @@ func (t *databaseTarget) evict(webhookID string) {
 	)
 }
 
+// evictAll evicts every cached archive writer, exactly as evict
+// does for one webhook. The engine calls it at shutdown, once its
+// workers have returned. Closing the last handle on an archive
+// moves the contents of its -wal into the .db and removes the
+// -wal, so a clean stop leaves each archive as a single file.
+func (t *databaseTarget) evictAll() {
+	t.mu.Lock()
+
+	writers := t.writers
+	t.writers = nil
+
+	t.mu.Unlock()
+
+	for _, w := range writers {
+		w.evict()
+	}
+}
+
 // sweepWebhook prunes one webhook's archive of rows older than
 // expiry, without requiring a write. It returns nil (nothing to
 // do) when the archive file does not exist, so a sweep never

internal/delivery/target_database_evict_test.go

@@ -1,6 +1,7 @@
 package delivery_test
 
 import (
+	"context"
 	"errors"
 	"fmt"
 	"net/http"
@@ -361,3 +362,46 @@ func TestEvictWebhook_LaterDeliveryRecreatesWriter(t *testing.T) {
 		"a later delivery should recreate the writer",
 	)
 }
+
+// TestEngineStop_WriteAfterStopIsRefused proves the engine's stop
+// closes each archive writer the way deleting its webhook does: a
+// write that reaches a writer after the stop is refused, reopens
+// nothing and adds no row.
+func TestEngineStop_WriteAfterStopIsRefused(t *testing.T) {
+	t.Parallel()
+
+	eng, _ := evictTestEngine(t)
+
+	webhookDB := testWebhookDB(t)
+	event := seedEvent(t, webhookDB, `{"archived":true}`)
+	d := seedDatabaseTargetDelivery(t, webhookDB, event, "")
+
+	eng.ExportDeliverDatabase(webhookDB, d)
+
+	w := eng.ExportArchiveWriterFor(event.WebhookID)
+	require.NotNil(t, w)
+	require.True(t, w.HandleOpen())
+
+	require.NoError(t, eng.ExportStop(context.Background()))
+
+	err := w.Write(evictTestRow("ev-after-stop"), 0)
+
+	require.ErrorIs(
+		t, err, delivery.ErrExportArchiveWriterEvicted,
+		"a write after the stop must be refused",
+	)
+	assert.False(
+		t, w.HandleOpen(),
+		"a refused write must not reopen the archive",
+	)
+	assert.False(
+		t, eng.ExportHasArchiveWriter(event.WebhookID),
+		"the stop should empty the registry",
+	)
+
+	count, err := countArchivedRows(w.Path())
+	require.NoError(t, err)
+	assert.Equal(
+		t, int64(1), count, "the refused row must not be written",
+	)
+}

internal/delivery/target_headers_test.go

@@ -11,10 +11,11 @@ import (
 	"sneak.berlin/go/webhooker/internal/delivery"
 )
 
-// Literals these tests repeat, named so that the header name and the
+// Literals these tests repeat, named so that the header names and the
 // keep-forever archive config each have one definition.
 const (
 	headerAuthorization = "Authorization"
+	headerContentType   = "Content-Type"
 	bearerValue         = "Bearer abc"
 	archiveConfigNever  = "{\"expiry\":\"never\"}"
 )

internal/delivery/target_http.go

@@ -197,10 +197,14 @@ func (c *httpCore) circuitBreakerBlock(
 		"cooldown_remaining", remaining,
 	)
 
+	// A delivery already at retrying is left as it is, so a task
+	// the breaker keeps turning away writes nothing each time.
+	if d.Status != database.DeliveryStatusRetrying {
 		c.eng.settleStatus(
 			webhookDB, d, d.Target.Type,
 			database.DeliveryStatusRetrying,
 		)
+	}
 
 	retryTask := *task
 	sched.ScheduleRetry(retryTask, remaining)
@@ -537,6 +541,11 @@ func isForwardableHeader(name string) bool {
 		"Upgrade", "Proxy-Authorization",
 		"Proxy-Connection", "Content-Length":
 		return false
+	case "Content-Type":
+		// applyRequestHeaders sets Content-Type itself. The receiver
+		// already stored this inbound value as the event's
+		// ContentType, so forwarding it too would send it twice.
+		return false
 	default:
 		return true
 	}
@@ -549,6 +558,10 @@ func isForwardableHeader(name string) bool {
 // policy strips exactly that set on a hop that leaves the origin,
 // so the forward set is decided here and only here — a header added
 // to it is covered off-origin without a second edit elsewhere.
+//
+// Content-Type goes out once: a Content-Type configured on the target
+// wins, otherwise the event's ContentType, otherwise none. The inbound
+// Content-Type in the event's headers is never forwarded.
 func applyRequestHeaders(
 	req *http.Request,
 	event *database.Event,
@@ -569,10 +582,10 @@ func applyRequestHeaders(
 
 	req.Header.Set("User-Agent", "webhooker/1.0")
 
-	// Content-Type describes the body being sent rather than the
+	// A Content-Type configured on the target describes the body
-	// sender, and the delivery path sets it from the event itself.
+	// being sent rather than the sender. A 307/308 preserves the
-	// A 307/308 preserves the body across hosts, so stripping it
+	// body across hosts, so stripping it would send that body
-	// would send that body untyped.
+	// untyped.
 	delete(originScoped, "Content-Type")
 
 	// User-Agent is overwritten just above, so an inbound one never

internal/delivery/target_slack.go

@@ -231,10 +231,15 @@ func FormatSlackMessage(
 		event.ContentType,
 	)
 
+	timestamp := "unknown"
+	if !event.CreatedAt.IsZero() {
+		timestamp = event.CreatedAt.UTC().Format(time.RFC3339)
+	}
+
 	fmt.Fprintf(
 		&b,
 		"*Timestamp:* `%s`\n",
-		event.CreatedAt.UTC().Format(time.RFC3339),
+		timestamp,
 	)
 
 	fmt.Fprintf(

internal/delivery/terminal_state_test.go

@@ -18,16 +18,17 @@ import (
 // https://git.eeqj.de/sneak/webhooker/issues/107: a delivery failed
 // with nothing in its event log to say why, and a retrying delivery
 // whose target was deleted, which used to keep sending and then never
-// terminalise.
+// terminalise. Section 4 is the same deleted-target gap for a pending
+// delivery: https://git.eeqj.de/sneak/webhooker/issues/293.
 
 // tUnknownType is a target type no build implements. It stands in for
 // a target whose type was written by a build that knew a type this one
 // does not.
 const tUnknownType = database.TargetType("pubsub")
 
-// tSeedDeletedTarget creates a target, a retrying delivery against it
+// tSeedDeletedTarget creates a target, a delivery against it at the
-// with one recorded failed attempt, and then deletes the target the
+// given status with one recorded failed attempt, and then deletes the
-// way the source page does.
+// target the way the source page does.
 //
 // It asserts the delete is soft, because that is the whole reason the
 // engine could not tell a deleted target from a target id that never
@@ -36,6 +37,7 @@ func tSeedDeletedTarget(
 	t *testing.T,
 	s iSetup,
 	name, url string,
+	status database.DeliveryStatus,
 ) string {
 	t.Helper()
 
@@ -51,8 +53,7 @@ func tSeedDeletedTarget(
 	)
 
 	d := iSeedDelivery(
-		t, s.WebhookDB, event.ID, targetID,
+		t, s.WebhookDB, event.ID, targetID, status,
-		database.DeliveryStatusRetrying,
 	)
 
 	iSeedFailedResult(t, s.WebhookDB, d.ID)
@@ -173,6 +174,7 @@ func TestRecoverSingleRetry_TargetDeleted(t *testing.T) {
 
 	deliveryID := tSeedDeletedTarget(
 		t, s, "gone-on-recovery", "http://example.com/hook",
+		database.DeliveryStatusRetrying,
 	)
 
 	s.Engine.ExportRecoverWebhookDeliveries(
@@ -210,6 +212,7 @@ func TestSweepSingleRetry_TargetDeleted(t *testing.T) {
 
 	deliveryID := tSeedDeletedTarget(
 		t, s, "gone-on-sweep", "http://example.com/hook",
+		database.DeliveryStatusRetrying,
 	)
 
 	// Twice, because the bug was an error the sweep repeated every
@@ -278,11 +281,11 @@ func TestSweepSingleRetry_TargetNeverExisted(t *testing.T) {
 	)
 }
 
-// TestFailMissingTargetRetry_WritesNoTargetRow holds the new terminal
+// TestFailMissingTarget_WritesNoTargetRow holds the new terminal path
-// path to the same rule as the existing one: no target row, and so no
+// to the same rule as the existing one: no target row, and so no
 // plaintext target config, may be written into the per-webhook event
 // database. See https://git.eeqj.de/sneak/webhooker/issues/206.
-func TestFailMissingTargetRetry_WritesNoTargetRow(
+func TestFailMissingTarget_WritesNoTargetRow(
 	t *testing.T,
 ) {
 	t.Parallel()
@@ -297,6 +300,7 @@ func TestFailMissingTargetRetry_WritesNoTargetRow(
 
 	deliveryID := tSeedDeletedTarget(
 		t, s, "credential-bearing", hookURL,
+		database.DeliveryStatusRetrying,
 	)
 
 	s.Engine.ExportSweepWebhookRetries(
@@ -529,3 +533,260 @@ func TestRecoverSingleRetry_TargetUnreadable_LeavesDeliveryAlone(
 
 	assert.Zero(t, s.Engine.ExportInflightHeld())
 }
+
+// --- 4. A pending delivery whose target is gone ---
+
+func TestRecoverPending_TargetDeleted(t *testing.T) {
+	t.Parallel()
+
+	s := newISetup(t)
+
+	deliveryID := tSeedDeletedTarget(
+		t, s, "gone-while-pending", "http://example.com/hook",
+		database.DeliveryStatusPending,
+	)
+
+	s.Engine.ExportRecoverWebhookDeliveries(
+		context.Background(), s.WebhookID,
+	)
+
+	iAssertStatus(
+		t, s.WebhookDB, deliveryID,
+		database.DeliveryStatusFailed,
+	)
+
+	last := tLastResult(t, s, deliveryID, 2)
+
+	assert.False(t, last.Success)
+	assert.Equal(t, 2, last.AttemptNum)
+	assert.Contains(t, last.Error, "gone-while-pending")
+	assert.Contains(t, last.Error, "was deleted")
+
+	assert.Empty(t, fDrain(s.Engine),
+		"a delivery whose target is gone was sent",
+	)
+	assert.Zero(t, s.Engine.ExportInflightHeld(),
+		"the terminal path leaked its ownership reference",
+	)
+}
+
+// TestRecoverPending_TargetDeleted_LeavesAnOwnedDeliveryAlone: the
+// terminal write takes ownership like every other recovery write, so a
+// delivery the engine still holds is not failed underneath its worker.
+func TestRecoverPending_TargetDeleted_LeavesAnOwnedDeliveryAlone(
+	t *testing.T,
+) {
+	t.Parallel()
+
+	s := newISetup(t)
+
+	deliveryID := tSeedDeletedTarget(
+		t, s, "gone-but-owned", "http://example.com/hook",
+		database.DeliveryStatusPending,
+	)
+
+	require.True(t, s.Engine.ExportRetainDelivery(deliveryID))
+
+	s.Engine.ExportRecoverWebhookDeliveries(
+		context.Background(), s.WebhookID,
+	)
+
+	iAssertStatus(
+		t, s.WebhookDB, deliveryID,
+		database.DeliveryStatusPending,
+	)
+
+	assert.Len(t, iResults(t, s.WebhookDB, deliveryID), 1,
+		"a delivery the engine owns was failed underneath it",
+	)
+}
+
+// TestFailMissingTarget_LeavesASettledDeliveryAlone: the recovery paths
+// read their batch before taking ownership, and a worker may send a
+// delivery and let it go in between. The terminal write goes by the row
+// as it is now, not as the batch read it.
+func TestFailMissingTarget_LeavesASettledDeliveryAlone(
+	t *testing.T,
+) {
+	t.Parallel()
+
+	s := newISetup(t)
+
+	deliveryID := tSeedDeletedTarget(
+		t, s, "gone-after-sending", "http://example.com/hook",
+		database.DeliveryStatusPending,
+	)
+
+	var batch database.Delivery
+
+	require.NoError(t, s.WebhookDB.First(
+		&batch, "id = ?", deliveryID,
+	).Error)
+
+	// A worker settles the delivery after the batch was read.
+	require.NoError(t, s.WebhookDB.Model(&database.Delivery{}).
+		Where("id = ?", deliveryID).
+		Update("status", database.DeliveryStatusDelivered).Error)
+
+	s.Engine.ExportFailMissingTarget(
+		s.WebhookDB, s.WebhookID, &batch,
+	)
+
+	iAssertStatus(
+		t, s.WebhookDB, deliveryID,
+		database.DeliveryStatusDelivered,
+	)
+
+	assert.Len(t, iResults(t, s.WebhookDB, deliveryID), 1,
+		"a delivery settled after the batch read was then failed",
+	)
+	assert.Zero(t, s.Engine.ExportInflightHeld())
+}
+
+// TestSweepPending_TargetDeleted sweeps twice over a batch that also
+// holds a healthy stranded delivery. The one whose target is gone is
+// failed once and then left alone; the healthy one is queued by the
+// first sweep and not again by the second.
+func TestSweepPending_TargetDeleted(t *testing.T) {
+	t.Parallel()
+
+	liveTargetID := uuid.New().String()
+	s := fSweepSetup(t, liveTargetID, "still-there")
+
+	deliveryID := tSeedDeletedTarget(
+		t, s, "gone-on-pending-sweep", "http://example.com/hook",
+		database.DeliveryStatusPending,
+	)
+	rAgePending(t, s.WebhookDB, deliveryID)
+
+	event := iSeedEvent(
+		t, s.WebhookDB, s.WebhookID, `{"target":"live"}`,
+	)
+
+	healthy := iSeedDelivery(
+		t, s.WebhookDB, event.ID, liveTargetID,
+		database.DeliveryStatusPending,
+	)
+	rAgePending(t, s.WebhookDB, healthy.ID)
+
+	ctx := context.Background()
+
+	s.Engine.ExportSweepWebhookRetries(ctx, s.WebhookID)
+
+	tasks := fDrain(s.Engine)
+	require.Len(t, tasks, 1,
+		"the first sweep did not queue the healthy delivery",
+	)
+	assert.Equal(t, healthy.ID, tasks[0].DeliveryID)
+
+	s.Engine.ExportSweepWebhookRetries(ctx, s.WebhookID)
+
+	assert.Empty(t, fDrain(s.Engine),
+		"the second sweep queued a delivery again",
+	)
+
+	iAssertStatus(
+		t, s.WebhookDB, deliveryID,
+		database.DeliveryStatusFailed,
+	)
+
+	last := tLastResult(t, s, deliveryID, 2)
+
+	assert.Contains(t, last.Error, "gone-on-pending-sweep")
+	assert.Contains(t, last.Error, "was deleted")
+}
+
+// TestSendRecoveredDeliveries_TargetMissingFromMap: the batch's target
+// map is empty when its query failed, so every delivery in the batch is
+// looked up on its own. A healthy one is sent to the target that lookup
+// finds.
+func TestSendRecoveredDeliveries_TargetMissingFromMap(
+	t *testing.T,
+) {
+	t.Parallel()
+
+	s := newISetup(t)
+
+	targetID := uuid.New().String()
+
+	iCreateTarget(
+		t, s.MainDB, targetID, s.WebhookID, "found-on-lookup",
+		database.TargetTypeLog, "", 0,
+	)
+
+	event := iSeedEvent(
+		t, s.WebhookDB, s.WebhookID, `{"map":"empty"}`,
+	)
+
+	d := iSeedDelivery(
+		t, s.WebhookDB, event.ID, targetID,
+		database.DeliveryStatusPending,
+	)
+
+	s.Engine.ExportSendRecoveredDeliveries(
+		context.Background(), s.WebhookDB,
+		[]database.Delivery{d}, s.WebhookID,
+		map[string]database.Target{}, nil,
+	)
+
+	tasks := fDrain(s.Engine)
+	require.Len(t, tasks, 1,
+		"the healthy delivery was not queued exactly once",
+	)
+	assert.Equal(t, d.ID, tasks[0].DeliveryID)
+	assert.Equal(t, targetID, tasks[0].TargetID)
+	assert.Equal(t, database.TargetTypeLog, tasks[0].TargetType)
+
+	iAssertStatus(
+		t, s.WebhookDB, d.ID,
+		database.DeliveryStatusPending,
+	)
+}
+
+// TestRecoverPending_TargetUnreadable_LeavesDeliveryAlone: a failed
+// read of the main database is not a deleted target. Restart recovery
+// holds every pending delivery of the webhook in one batch, so failing
+// on this would fail all of them.
+func TestRecoverPending_TargetUnreadable_LeavesDeliveryAlone(
+	t *testing.T,
+) {
+	t.Parallel()
+
+	s := newISetup(t)
+
+	targetID := uuid.New().String()
+
+	iCreateTarget(
+		t, s.MainDB, targetID, s.WebhookID, "healthy",
+		database.TargetTypeLog, "", 0,
+	)
+
+	event := iSeedEvent(
+		t, s.WebhookDB, s.WebhookID, `{"still":"pending"}`,
+	)
+
+	d := iSeedDelivery(
+		t, s.WebhookDB, event.ID, targetID,
+		database.DeliveryStatusPending,
+	)
+
+	sqlDB, err := s.MainDB.DB()
+	require.NoError(t, err)
+	require.NoError(t, sqlDB.Close())
+
+	s.Engine.ExportRecoverPendingDeliveries(
+		context.Background(), s.WebhookDB, s.WebhookID,
+	)
+
+	iAssertStatus(
+		t, s.WebhookDB, d.ID,
+		database.DeliveryStatusPending,
+	)
+
+	assert.Empty(t, iResults(t, s.WebhookDB, d.ID),
+		"an unreadable main database produced a terminal "+
+			"failure row",
+	)
+	assert.Empty(t, fDrain(s.Engine))
+	assert.Zero(t, s.Engine.ExportInflightHeld())
+}

internal/handlers/ui_copy_test.go

@@ -300,3 +300,80 @@ func TestEntrypointCopyButtonIsProgressiveEnhancement(t *testing.T) {
 		"the page must render to completion, not abort partway",
 	)
 }
+
+// maxRetriesHelp is the wording both target forms must carry. The
+// delivery core makes max_retries attempts in total, not that many
+// retries on top of a first try (a fresh delivery starts at attempt 1
+// and target_http gives up once the attempt number reaches
+// max_retries), and 0 is special-cased to a single fire-and-forget
+// attempt with no circuit breaker.
+const maxRetriesHelp = "This is the total number of delivery attempts, " +
+	"not retries on top of the first: a value of 3 makes three attempts " +
+	"in all. 0 means a single attempt with no retries and no circuit " +
+	"breaker."
+
+// TestTargetFormMaxRetriesCopyMatchesBehaviour pins the max_retries
+// help text on both the create form (the add-target form on the webhook
+// detail page) and the edit form, so the copy cannot drift back to
+// calling the number a retry count.
+func TestTargetFormMaxRetriesCopyMatchesBehaviour(t *testing.T) {
+	t.Parallel()
+
+	var h *handlers.Handlers
+
+	var sess *session.Session
+
+	app := newTestApp(t, &h, &sess)
+	app.RequireStart()
+
+	t.Cleanup(app.RequireStop)
+
+	webhook := &database.Webhook{Name: "wh", RetentionDays: 14}
+	webhook.ID = testWebhookID
+
+	entrypoint := database.Entrypoint{Path: "abc123"}
+	entrypoint.ID = "ep-1"
+
+	createBody := renderPage(
+		t, h, sess, "source_detail.html", map[string]any{
+			dataKeyWebhook: webhook,
+			"Entrypoints": handlers.NewEntrypointViews(
+				[]database.Entrypoint{entrypoint},
+			),
+			"Targets": delivery.NewTargetViews(nil),
+			"Events":  []database.Event{},
+			"BaseURL": "https://hooks.example.com",
+		},
+	)
+
+	assert.Contains(
+		t, createBody, maxRetriesHelp,
+		"the add-target form must explain max_retries as total attempts",
+	)
+
+	// A slack target exercises the same max_retries field while needing
+	// only Config.URL from the edit template, so the test data stays
+	// minimal. The Target key mirrors the field names the template reads
+	// off the handler's view value.
+	editBody := renderPage(
+		t, h, sess, "target_edit.html", map[string]any{
+			dataKeyWebhook: webhook,
+			"Target": map[string]any{
+				"ID":         "tg-1",
+				"Name":       "t",
+				"Type":       "slack",
+				"Active":     true,
+				"MaxRetries": 3,
+				"Config": map[string]any{
+					"URL": "https://hooks.slack.com/services/x",
+				},
+			},
+			dataKeyError: "",
+		},
+	)
+
+	assert.Contains(
+		t, editBody, maxRetriesHelp,
+		"the target edit form must explain max_retries as total attempts",
+	)
+}

internal/middleware/csrf_test.go

@@ -380,9 +380,8 @@ func csrfTookStrictPath(
 
 // TestCSRF_ForwardedProtoSpellingsTakeStrictPath runs the header
 // spellings a real proxy emits through the middleware. The environment
-// is dev -- the DEFAULT when WEBHOOKER_ENVIRONMENT is unset -- to pin
+// is set to dev -- the permissive setting -- to pin that the routing is
-// that the routing is a per-request transport decision and owes
+// a per-request transport decision and owes nothing to configuration.
-// nothing to configuration.
 func TestCSRF_ForwardedProtoSpellingsTakeStrictPath(t *testing.T) {
 	t.Parallel()
 

internal/middleware/recoverer.go

@@ -133,6 +133,11 @@ func (w *recoverResponseWriter) Unwrap() http.ResponseWriter {
 // what the access log records and the metrics count, and outside the
 // sentryhttp handler, whose Repanic option depends on something
 // further out recovering what it re-raises.
+//
+// Unlike http.Error on its own, it deletes any Set-Cookie the handler
+// set before panicking, because a request that failed must not hand
+// the client a credential; every other header is left to http.Error.
+// See https://git.eeqj.de/sneak/webhooker/issues/193.
 func (s *Middleware) Recoverer() func(http.Handler) http.Handler {
 	return func(next http.Handler) http.Handler {
 		return http.HandlerFunc(func(
@@ -164,6 +169,8 @@ func (s *Middleware) Recoverer() func(http.Handler) http.Handler {
 					return
 				}
 
+				rw.Header().Del("Set-Cookie")
+
 				http.Error(
 					rw,
 					http.StatusText(

internal/middleware/recoverer_test.go

@@ -304,16 +304,44 @@ func TestRecovererRepanicsErrAbortHandler(t *testing.T) {
 	)
 }
 
+// TestRecovererDropsSetCookieFromTheRecovered500 covers a handler that
+// sets a cookie and a redirect target and then panics before sending
+// anything. A request that failed must not hand the client a
+// credential, so the 500 carries no cookie; Location is left alone.
+func TestRecovererDropsSetCookieFromTheRecovered500(t *testing.T) {
+	t.Parallel()
+
+	probe := newRecovererProbe(
+		t, false,
+		func(w http.ResponseWriter, _ *http.Request) {
+			w.Header().Set("Set-Cookie", "session=x")
+			w.Header().Set("Location", "/after")
+
+			panic(panicMarker)
+		},
+	)
+
+	resp, err := probe.get(t)
+	require.NoError(t, err)
+	require.NoError(t, resp.Body.Close())
+
+	assert.Equal(t, http.StatusInternalServerError, resp.StatusCode)
+	assert.Empty(t, resp.Cookies())
+	assert.Equal(t, "/after", resp.Header.Get("Location"))
+}
+
 // TestRecovererKeepsAnAlreadyCommittedResponse covers a handler that
 // panics after sending its status. The bytes are already on the wire,
-// so a second WriteHeader would change nothing the client sees and
+// cookie included, so a second WriteHeader would change nothing the
-// would draw net/http's "superfluous response.WriteHeader" report.
+// client sees and would draw net/http's "superfluous
+// response.WriteHeader" report.
 func TestRecovererKeepsAnAlreadyCommittedResponse(t *testing.T) {
 	t.Parallel()
 
 	probe := newRecovererProbe(
 		t, false,
 		func(w http.ResponseWriter, _ *http.Request) {
+			w.Header().Set("Set-Cookie", "session=x")
 			w.WriteHeader(committedStatus)
 			_, _ = w.Write([]byte("partial"))
 
@@ -331,6 +359,7 @@ func TestRecovererKeepsAnAlreadyCommittedResponse(t *testing.T) {
 
 	assert.Equal(t, committedStatus, resp.StatusCode)
 	assert.Equal(t, "partial", string(body))
+	assert.Len(t, resp.Cookies(), 1)
 
 	record := probe.panicRecord(t)
 	assert.Equal(t, panicMarker, record["panic"])

internal/reqtls/reqtls.go

@@ -5,7 +5,7 @@
 // several packages, by hand, and the answers disagreed. The session
 // cookie's Secure attribute was decided at startup from the configured
 // environment while the CSRF cookie's was decided per-request, so a
-// deployment behind a TLS proxy in the default environment emitted one
+// deployment behind a TLS proxy in the dev environment emitted one
 // Secure cookie and one non-Secure cookie on the same response.
 // Everything kept working, which is exactly why nobody noticed.
 //

internal/server/routes.go

@@ -92,11 +92,25 @@ func (s *Server) setupGlobalMiddleware() {
 func (s *Server) setupRoutes() {
 	s.router.Get("/", s.h.HandleIndex())
 
-	s.router.Mount(
+	// Static assets answer GET and HEAD only. chi's default 405
-		"/s",
+	// carries no Allow header, so this group supplies its own.
-		http.StripPrefix("/s", http.FileServer(http.FS(static.Static))),
+	staticFiles := http.StripPrefix(
+		"/s", http.FileServer(http.FS(static.Static)),
 	)
 
+	s.router.Route("/s", func(r chi.Router) {
+		r.MethodNotAllowed(func(w http.ResponseWriter, _ *http.Request) {
+			w.Header().Set("Allow", "GET, HEAD")
+			http.Error(
+				w,
+				"Method Not Allowed",
+				http.StatusMethodNotAllowed,
+			)
+		})
+		r.Method(http.MethodGet, "/*", staticFiles)
+		r.Method(http.MethodHead, "/*", staticFiles)
+	})
+
 	s.router.Route("/api/v1", func(_ chi.Router) {
 		// API routes will be added here.
 	})

internal/server/routes_test.go

@@ -7,6 +7,7 @@ import (
 	"net/http/httptest"
 	"net/url"
 	"regexp"
+	"slices"
 	"strconv"
 	"strings"
 	"testing"
@@ -220,9 +221,21 @@ func (e *testEnv) csrfFrom(
 	// out of the markup has to be unescaped before it is submitted.
 	token := html.UnescapeString(match[1])
 
-	combined := make([]*http.Cookie, 0, len(cookies))
+	// A cookie the page sets replaces the one of the same name, as in
-	combined = append(combined, cookies...)
+	// a browser. Sent both, the server would read the first, older one.
-	combined = append(combined, w.Result().Cookies()...)
+	set := w.Result().Cookies()
+	combined := make([]*http.Cookie, 0, len(cookies)+len(set))
+
+	for _, c := range cookies {
+		replaced := slices.ContainsFunc(set, func(n *http.Cookie) bool {
+			return n.Name == c.Name
+		})
+		if !replaced {
+			combined = append(combined, c)
+		}
+	}
+
+	combined = append(combined, set...)
 
 	return token, combined
 }
@@ -396,13 +409,15 @@ func (e *testEnv) storedHash(t *testing.T, username string) string {
 
 // --- /s static group ---
 
-// TestStaticServesEveryMethod pins what the static mount actually
+// TestStaticServesOnlyGetAndHead pins the methods the static group
-// answers. chi's Mount registers the handler for all methods and
+// answers: GET and HEAD are served the asset, and the other methods
-// http.FileServer only special-cases HEAD (by suppressing the body),
+// chi routes (POST, PUT, DELETE and the rest) are refused with 405
-// so a POST or a DELETE to an asset is served the file rather than
+// and an Allow header naming those two. A method chi does not route,
-// refused. The README documents this; the test is what keeps the two
+// such as PROPFIND, is refused with 405 by the top-level router
-// from drifting.
+// before it reaches the static group, so it gets no Allow header.
-func TestStaticServesEveryMethod(t *testing.T) {
+// The README documents this; the test is what keeps the two from
+// drifting.
+func TestStaticServesOnlyGetAndHead(t *testing.T) {
 	t.Parallel()
 
 	env := newTestEnv(t)
@@ -417,6 +432,7 @@ func TestStaticServesEveryMethod(t *testing.T) {
 		http.MethodPost,
 		http.MethodPut,
 		http.MethodDelete,
+		"PROPFIND",
 	} {
 		t.Run(method, func(t *testing.T) {
 			t.Parallel()
@@ -428,18 +444,38 @@ func TestStaticServesEveryMethod(t *testing.T) {
 			w := httptest.NewRecorder()
 			env.router.ServeHTTP(w, req)
 
-			assert.Equal(t, http.StatusOK, w.Code,
+			switch method {
-				"static mount answers every method")
+			case http.MethodGet:
-
+				assert.Equal(t, http.StatusOK, w.Code)
-			if method == http.MethodHead {
-				assert.Empty(t, w.Body.Bytes(),
-					"HEAD must not carry a body")
-
-				return
-			}
-
 				assert.Equal(t, body, w.Body.Bytes(),
 					"the asset itself is returned")
+			case http.MethodHead:
+				assert.Equal(t, http.StatusOK, w.Code)
+				assert.Empty(t, w.Body.Bytes(),
+					"HEAD must not carry a body")
+			case "PROPFIND":
+				assert.Equal(
+					t, http.StatusMethodNotAllowed, w.Code,
+				)
+				assert.Empty(t, w.Header().Get("Allow"),
+					"chi refuses a method it does not route "+
+						"before the static group runs")
+				assert.NotContains(
+					t, w.Body.String(), string(body),
+					"a refused method must not get the asset",
+				)
+			default:
+				assert.Equal(
+					t, http.StatusMethodNotAllowed, w.Code,
+				)
+				assert.Equal(
+					t, "GET, HEAD", w.Header().Get("Allow"),
+				)
+				assert.NotContains(
+					t, w.Body.String(), string(body),
+					"a refused method must not get the asset",
+				)
+			}
 		})
 	}
 }
@@ -591,6 +627,59 @@ func TestPagesLogin_CorrectPasswordSurvivesASpentBudget(
 	)
 }
 
+// TestPagesLogin_CookiesFromAnEarlierDatabase is
+// https://git.eeqj.de/sneak/webhooker/issues/359. A new database
+// brings a new session key, and the operator's browser still holds
+// the session and CSRF cookies signed with the old one. Logging in
+// must work as from a fresh browser and leave cookies the new key
+// accepts.
+func TestPagesLogin_CookiesFromAnEarlierDatabase(t *testing.T) {
+	t.Parallel()
+
+	const (
+		username = "operator"
+		password = "correct-horse-battery-staple"
+	)
+
+	earlier := newTestEnv(t)
+	earlierID, _ := earlier.seedUser(t, username, password)
+	_, stale := earlier.csrfFrom(t, "/pages/login", nil)
+	stale = append(stale, earlier.authCookies(t, earlierID, username)...)
+
+	env := newTestEnv(t)
+	env.seedUser(t, username, password)
+
+	token, cookies := env.csrfFrom(t, "/pages/login", stale)
+
+	form := url.Values{}
+	form.Set("csrf_token", token)
+	form.Set("username", username)
+	form.Set("password", password)
+
+	w := env.post("/pages/login", form, cookies)
+	require.Equal(
+		t, http.StatusSeeOther, w.Code,
+		"a session cookie from another key must not fail the login",
+	)
+
+	// The response deletes the old session cookie and then sets the
+	// new one; a browser keeps the last.
+	var fresh *http.Cookie
+
+	for _, c := range w.Result().Cookies() {
+		if c.Name == session.SessionName {
+			fresh = c
+		}
+	}
+
+	require.NotNil(t, fresh, "login must set a session cookie")
+	assert.Equal(
+		t, "/sources",
+		env.get("/", []*http.Cookie{fresh}).Header().Get("Location"),
+		"the new session cookie must authenticate",
+	)
+}
+
 // --- /user/{username} group ---
 
 // TestPasswordChange_OversizeBody_RejectedAndPasswordUnchanged

internal/session/codec_test.go

@@ -19,8 +19,8 @@ import (
 )
 
 // The tests below exercise the securecookie codecs underneath the
-// store and nothing else: Session.Get only decodes, so no server-side
+// store and nothing else: they decode through the store itself, so no
-// expiry check takes part in the result. They exist because
+// server-side expiry check takes part in the result. They exist because
 // NewCookieStore gives its codecs a 30-day max age that assigning
 // store.Options does not override, which would let the codec accept a
 // cookie weeks past the cap the cookie attribute advertises.
@@ -75,10 +75,11 @@ func restamp(
 	return base64.URLEncoding.EncodeToString(payload)
 }
 
-// decodeCookie feeds value back through the store's decode path.
+// decodeCookie feeds value back through the store's decode path. It
+// asks the store rather than Session.Get, which treats a cookie that
+// does not decode as absent and so hides the codec's reason.
 func decodeCookie(
 	t *testing.T,
-	s *session.Session,
 	value string,
 ) (*sessions.Session, error) {
 	t.Helper()
@@ -94,7 +95,7 @@ func decodeCookie(
 		SameSite: http.SameSiteLaxMode,
 	})
 
-	sess, err := s.Get(req)
+	sess, err := session.NewStore(testKey()).Get(req, session.SessionName)
 	require.NotNil(t, sess)
 
 	return sess, err
@@ -105,7 +106,7 @@ func TestCodec_AcceptsCookieInsideAbsoluteCap(t *testing.T) {
 
 	s := testSession(t)
 
-	sess, err := decodeCookie(t, s, restamp(
+	sess, err := decodeCookie(t, restamp(
 		t,
 		issuedCookie(t, s),
 		time.Now().Add(-(testAbsoluteMaxAge-time.Hour)),
@@ -126,7 +127,7 @@ func TestCodec_RejectsCookiePastAbsoluteCap(t *testing.T) {
 
 	s := testSession(t)
 
-	sess, err := decodeCookie(t, s, restamp(
+	sess, err := decodeCookie(t, restamp(
 		t,
 		issuedCookie(t, s),
 		time.Now().Add(-(testAbsoluteMaxAge+time.Hour)),

internal/session/session.go

@@ -146,8 +146,8 @@ func newStore(key []byte) *sessions.CookieStore {
 //
 // This is decided per-request, not once at startup. Deciding it at
 // startup from the configured environment is what this replaces, and
-// it got the DEFAULT posture wrong: "dev" is the environment when
+// it got the DEFAULT posture wrong: "dev" was then the environment when
-// WEBHOOKER_ENVIRONMENT is unset, so a deployment terminating TLS at a
+// WEBHOOKER_ENVIRONMENT was unset, so a deployment terminating TLS at a
 // proxy without also setting the environment emitted the
 // authentication cookie with no Secure attribute -- silently, and on
 // the same response as a CSRF cookie that did have one.
@@ -224,10 +224,22 @@ func New(
 }
 
 // Get retrieves a session for the request.
+//
+// A session cookie that does not decode -- one signed with an earlier
+// session key, say, because the database was made anew -- is treated
+// as absent: the caller gets a new, empty session and no error, and
+// the next save replaces the cookie.
 func (s *Session) Get(
 	r *http.Request,
 ) (*sessions.Session, error) {
-	return s.store.Get(r, SessionName)
+	sess, err := s.store.Get(r, SessionName)
+	if sess == nil {
+		return nil, err
+	}
+
+	// For a cookie that does not decode, gorilla/sessions returns a
+	// new, empty session alongside the error that is dropped here.
+	return sess, nil
 }
 
 // GetKey returns the raw 32-byte authentication key used for

internal/session/session_test.go

@@ -990,8 +990,8 @@ func sessionCookieFrom(
 
 // TestSave_SecureFollowsRequestTransport is the regression test for
 // the defect this replaces: Secure was fixed at startup from the
-// configured environment, and "dev" is the environment when
+// configured environment, and "dev" was then the environment when
-// WEBHOOKER_ENVIRONMENT is unset. A deployment behind a TLS proxy in
+// WEBHOOKER_ENVIRONMENT was unset. A deployment behind a TLS proxy in
 // that DEFAULT posture shipped the authentication cookie with no
 // Secure attribute and said nothing about it.
 //

templates/source_detail.html

@@ -120,10 +120,13 @@
                         <label class="text-sm text-gray-700">Timeout (seconds, blank = default):</label>
                         <input type="number" name="timeout" min="0" max="300" :disabled="targetType !== 'http'" class="input text-sm w-24">
                     </div>
-                    <div x-show="targetType === 'http'" class="flex gap-2 items-center">
+                    <div x-show="targetType === 'http'">
-                        <label class="text-sm text-gray-700">Max retries (0 = fire-and-forget):</label>
+                        <div class="flex gap-2 items-center">
+                            <label class="text-sm text-gray-700">Max retries:</label>
                             <input type="number" name="max_retries" value="0" min="0" max="20" class="input text-sm w-24">
                         </div>
+                        <p class="text-xs text-gray-500 mt-1">This is the total number of delivery attempts, not retries on top of the first: a value of 3 makes three attempts in all. 0 means a single attempt with no retries and no circuit breaker.</p>
+                    </div>
                     <div x-show="targetType === 'slack'">
                         <input type="url" name="url" placeholder="https://hooks.slack.com/services/..." :disabled="targetType !== 'slack'" class="input text-sm">
                         <p class="text-xs text-gray-500 mt-1">Slack or Mattermost incoming webhook URL. Payloads are pretty-printed in code blocks.</p>

templates/target_edit.html

@@ -69,7 +69,7 @@
             <div class="form-group">
                 <label for="max_retries" class="label">Max retries</label>
                 <input type="number" id="max_retries" name="max_retries" value="{{.Target.MaxRetries}}" min="0" max="20" class="input">
-                <p class="text-xs text-gray-500 mt-1">0 is fire-and-forget: one attempt, no circuit breaker.</p>
+                <p class="text-xs text-gray-500 mt-1">This is the total number of delivery attempts, not retries on top of the first: a value of 3 makes three attempts in all. 0 means a single attempt with no retries and no circuit breaker.</p>
             </div>
             {{end}}