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

Two writes that failed at the same time put their reports back in the wrong order when the older one failed first. A mutex is now held across each write, from taking the waiting reports out of the buffer to putting them back if the write fails, so the next write takes them in order. The write Append starts at the 10 MiB threshold waits for one under way, and Append starts no other while it waits. Files written since the start are now deleted to make room in the order they were written. A test that needed two writes at once is removed; new tests cover the one waiting write and the order files are deleted in. Model: opus-5-5
Discarded-check unit tests check nothing was recorded (closes #151 )
2026-10-19 01:59:25 +00:00 · 2026-10-19 03:44:13 +02:00 · 2026-10-19 03:15:35 +02:00 · 2026-10-19 03:10:39 +02:00 · 2026-10-19 02:55:29 +02:00 · 2026-10-19 01:34:45 +02:00
25 changed files with 2283 additions and 420 deletions
@@ -5,11 +5,11 @@
 #
 # The lint and test phases are the gates: `make lint` (script/lint)
 # builds the lint stage alone and `make test` (script/test) the test
-# stage alone, and the builder stage depends on both, so the image
-# cannot be built unless they pass. Each covers the frontend as well,
-# through a copy from a node stage. Inside them each tool is invoked
-# directly, never through make or script/, whose lint and test are
-# themselves docker builds.
+# stage alone, and the two build stages, frontend-build and builder,
+# depend on both, so the image cannot be built unless they pass. Each
+# phase covers the frontend as well, through a copy from a node stage.
+# Inside them each tool is invoked directly, never through make or
+# script/, whose lint and test are themselves docker builds.

 # Frontend lint stage: eslint with the rules in eslint.config.js. The
 # lint phase below runs it.
@@ -35,30 +35,26 @@ RUN golangci-lint run --config .golangci.yml ./...
 # phase run it.
 COPY --from=frontend-lint /app/yarn.lock /dev/null

-# Frontend stage: the unit tests in test/unit/, then the production
-# build into dist/, which the runtime stage serves. The test phase below
-# runs it. The tests print a dot each; if any fails, they run again with
-# every test listed, and the step fails even if that run passes.
-# NODE_OPTIONS chooses the reporter because yarn adds its arguments
-# after the test files, where node would take a reporter option for one
-# more file.
+# Frontend test stage: the unit tests in test/unit/. The test phase
+# below runs it, and the frontend-build stage starts from it. The tests
+# print a dot each; if any fails, they run again with every test listed,
+# and the step fails even if that run passes. NODE_OPTIONS chooses the
+# reporter because yarn adds its arguments after the test files, where
+# node would take a reporter option for one more file.
 # node:22-alpine as of 2026-02-22
-FROM node@sha256:e4bf2a82ad0a4037d28035ae71529873c069b13eb0455466ae0bc13363826e34 AS frontend
+FROM node@sha256:e4bf2a82ad0a4037d28035ae71529873c069b13eb0455466ae0bc13363826e34 AS frontend-test
 WORKDIR /app
 COPY package.json yarn.lock ./
 RUN yarn install --frozen-lockfile
-# vite.config.js reads the commit for the page's footer with git.
-RUN apk add --no-cache git
 COPY . .
 RUN NODE_OPTIONS=--test-reporter=dot timeout 90 yarn --silent run test || \
    { echo "--- Rerunning with every test listed for details ---"; \
      NODE_OPTIONS=--test-reporter=spec timeout 90 yarn --silent run test; \
      exit 1; }
-RUN yarn build

 # Test phase: the backend's tests with the race detector and coverage,
-# and the frontend's through the copy from the frontend stage at the
-# end. -race needs cgo and so a C compiler, which the Debian Go image
+# and the frontend's through the copy from the frontend-test stage at
+# the end. -race needs cgo and so a C compiler, which the Debian Go image
 # ships and the alpine one does not. -timeout 90s is a backstop above
 # the 60-second cap on the suite. The rerun with -v only shows details:
 # the step fails however it ends, because the first run already failed.
@@ -75,9 +71,21 @@ RUN --mount=type=tmpfs,target=/go/pkg/mod \
    go test -timeout 90s -race -cover ./... || \
    { echo "--- Rerunning with -v for details ---"; \
      go test -timeout 90s -race -v ./...; exit 1; }
-# Nothing is wanted from the frontend stage; the copy is what makes
+# Nothing is wanted from the frontend-test stage; the copy is what makes
 # this phase run its tests.
-COPY --from=frontend /app/yarn.lock /dev/null
+COPY --from=frontend-test /app/yarn.lock /dev/null
+
+# Frontend build stage: the production build into dist/, which the
+# runtime stage serves, on the yarn packages and source of the
+# frontend-test stage. Nothing is wanted from the two phases; the copies
+# are what make BuildKit build them first, so this stage cannot run
+# unless lint and test passed.
+FROM frontend-test AS frontend-build
+COPY --from=lint /src/go.sum /dev/null
+COPY --from=test /src/go.sum /dev/null
+# vite.config.js reads the commit for the page's footer with git.
+RUN apk add --no-cache git
+RUN yarn build

 # Backend build stage. Nothing is wanted from the two phases; the copies
 # are what make BuildKit build them first, so this stage cannot run
@@ -131,7 +139,7 @@ RUN addgroup -g 1000 -S netwatch && \
 RUN rm /etc/nginx/conf.d/default.conf
 COPY configs/nginx.conf /etc/nginx/templates/netwatch.conf.template
 COPY configs/security-headers.conf /etc/nginx/security-headers.conf
-COPY --from=frontend /app/dist /usr/share/nginx/html
+COPY --from=frontend-build /app/dist /usr/share/nginx/html
 COPY --from=builder /src/netwatch-server /usr/local/bin/netwatch-server
 COPY bin/entrypoint.sh /usr/local/bin/entrypoint.sh

@@ -59,8 +59,9 @@ either one is broken. We provide:
 - `script/clean` — remove what `script/build` writes
 - `script/projectname` — print the project name (used for the Docker image tag)
 - `script/test` — build the `test` phase of `Dockerfile` without the cache: the
-  frontend's unit tests and production build in its `frontend` stage, and the
-  backend's Go tests with the race detector and coverage
+  frontend's unit tests in its `frontend-test` stage, and the backend's Go tests
+  with the race detector and coverage. The frontend's production build is left
+  to the image build, in the `frontend-build` stage
 - `script/lint` — build the `lint` phase of `Dockerfile` without the cache:
  eslint in its `frontend-lint` stage, and golangci-lint over the Go packages
 - `script/fmt` — format all files (writes): prettier over the JavaScript, CSS,
@@ -78,14 +79,16 @@ either one is broken. We provide:
 - `script/frontend-test` — run the unit tests in `test/unit/` on the host with
  Node's built-in test runner, through the `test` script in `package.json`, and
  if any fails, run them again listing every test, and fail; then the production
-  build. Each test run has a 90-second timeout. `make test` runs the same in
-  Docker
+  build. Each test run has a 90-second timeout. `make test` runs the same tests
+  in Docker, without the production build
 - `script/frontend-fmt` — format the JavaScript, CSS, HTML and Markdown with
  prettier (writes)
 - `script/frontend-fmt-check` — check prettier formatting (read-only)
 - `script/frontend-viewport-test` — responsive-layout verification of the built
  frontend in a containerised headless Chrome (see
-  [test/viewport/README.md](test/viewport/README.md)). Not part of
+  [test/viewport/README.md](test/viewport/README.md)). Before the browser
+  starts, it checks that the page, and a path that falls back to it, are served
+  with `Cache-Control: no-cache` and `Content-Security-Policy`. Not part of
  `script/check`: it takes minutes.
 - `script/docker` — build the image from `Dockerfile` without the build cache,
  tagged `netwatch` via `script/projectname`
@@ -152,8 +155,26 @@ report a delta, so only new samples are sent; the mark advances only on a
 delivered report, and while paused nothing is sent. Changing the interval clears
 each host's history on the page; the samples in it not yet reported still go in
 the next report. Delivery failure is quiet — one debug-log line per outage,
-retried at the next interval, never blocking probing. The report-building step
-is a pure function of host state.
+never blocking probing. The report-building step is a pure function of host
+state.
+
+A sample in the history that no report has delivered is sent only as long as the
+history holds it, so after a refused or failed report the next report carries
+the samples the history still has. Samples an interval change cleared from the
+history are kept apart, the last 100 per host, until they are delivered. The
+history keeps each host's last 100 rounds: 100 seconds at the 1-second interval,
+200 seconds at 2 seconds, 5 minutes at the default of 3 seconds and 100 minutes
+at 60 seconds. At 1 second, one refused or failed report loses the oldest 20
+seconds of each host's samples, as the next delivered report comes 2 minutes
+after the last. At 2 seconds and longer one loses nothing; samples are lost from
+the third refused report in a row at 2 seconds, from the fifth at 3 seconds, and
+later still at the longer intervals.
+
+No report is sent while the page is paused, but each paused round still adds an
+entry to the history, so each report due during a pause counts as a refused one.
+A pause longer than the history spans loses the samples in the history recorded
+between the last delivered report and the pause: up to a minute's worth for each
+host if the report before the pause was delivered.

 ### Backend

@@ -199,23 +220,32 @@ image, `make run` builds it and runs it on port 8080.
 | `DATA_DIR`             | `./data/reports`     | Directory for compressed reports                                                                         |
 | `DATA_DIR_MAX_BYTES`   | `1073741824` (1 GiB) | Most bytes of report files kept in `DATA_DIR`, oldest deleted first; see [Report limits](#report-limits) |
 | `DEBUG`                | `false`              | Enable debug logging                                                                                     |
-| `TRUSTED_PROXIES`      | loopback + RFC1918   | Comma-separated CIDRs whose `X-Forwarded-For` / `X-Real-IP` headers are trusted for client IP resolution |
+| `TRUSTED_PROXIES`      | RFC 1918 ranges      | Comma-separated CIDRs whose `X-Forwarded-For` / `X-Real-IP` headers are trusted for client IP resolution |
 | `REPORTS_PER_MINUTE`   | `60`                 | Reports each client address may send a minute; see [Report limits](#report-limits)                       |
 | `CORS_ALLOWED_ORIGINS` | empty                | Comma-separated origins whose pages may call the API; see [CORS](#cors)                                  |
 | `METRICS_USERNAME`     | empty                | Basic auth user name for `/metrics`; see [Metrics](#metrics)                                             |
 | `METRICS_PASSWORD`     | empty                | Basic auth password for `/metrics`; see [Metrics](#metrics)                                              |
 | `SENTRY_DSN`           | empty                | DSN of the Sentry project to send errors to; see [Sentry](#sentry)                                       |

-`TRUSTED_PROXIES` defaults to
-`127.0.0.1/32,::1/128,10.0.0.0/8,172.16.0.0/12,192.168.0.0/16`. The loopback
-entries cover a reverse proxy on the same host. A request whose direct peer is
-outside this set has its forwarded headers ignored, and the direct peer is
-logged and rate-limited instead. The container image does not use this default;
-see [Deployment](#deployment).
+`TRUSTED_PROXIES` defaults to `10.0.0.0/8,172.16.0.0/12,192.168.0.0/16`. A list
+replaces the default; set and empty, the server trusts no proxy. Loopback is not
+in the default, so a reverse proxy on the same host has to be named, as in
+`TRUSTED_PROXIES=127.0.0.1/32`. A request whose direct peer is outside this set
+has its forwarded headers ignored, and the direct peer is logged and
+rate-limited instead. From a direct peer inside the set, the client address is
+the right-most `X-Forwarded-For` entry that is not in the set: each proxy
+appends the address it took the request from, so the entries further left were
+written by the client and are not used. An entry written as an address with a
+port, such as `198.51.100.7:5000` or `[2001:db8::1]:443`, is read as that
+address. If the search from the right finds only addresses in the set, or meets
+an entry that is not an address at all, such as `unknown`, the direct peer is
+used. `X-Real-IP` is read only when `X-Forwarded-For` is missing or empty. The
+container image does not use this default; see [Deployment](#deployment).

 A variable set to a value the server cannot use, such as `PORT=abc`,
 `DEBUG=maybe` or a `BIND_ADDRESS` that is not an IP address, stops it from
-starting, with an error naming the variable. An empty variable counts as unset.
+starting, with an error naming the variable. An empty variable counts as unset,
+except `TRUSTED_PROXIES`.

 #### Report storage

@@ -228,12 +258,40 @@ the numbers: its file, if it was created, is removed. The file stays, counted
 toward `DATA_DIR_MAX_BYTES` from the next start, only if removing it fails too.
 Files are created with `O_EXCL` to prevent overwrites.

+The reports of a failed write stay in memory, and the next flush, within a
+minute, tries again to write them. The server writes one file at a time, so a
+file holds its reports in the order they arrived, those of a failed write
+included. From the failed write until a write succeeds, `POST /api/v1/reports`
+refuses every report with 500 and keeps nothing of it. The page sends the
+refused samples again in its next report, as long as it still holds them: at the
+1-second interval one refused report already loses some of those in the history
+(see [Reporting](#reporting)). Reports already answered 200 are lost only if the
+server stops before a write succeeds; the final flush at stop tries them too,
+and if it fails, the process exits non-zero.
+
 #### Report limits

 `POST /api/v1/reports` takes reports from anyone who can reach it, without
-credentials, so it is bounded instead. Both refusals below answer with the same
-`{"status":"error"}` body as any other error.
+credentials, so it is bounded instead. Each refusal below answers with the same
+`{"status":"error"}` body as any other error, except nginx's refusal of a body
+over the limit in the image.

+- **JSON only.** A report must be sent with `Content-Type: application/json`,
+  parameters such as `charset` allowed, as the page sends it; any other is
+  refused with 415, and nothing of it is stored. A page on another origin can
+  send a `text/plain` POST, or a form, without a CORS preflight; refusing them
+  keeps any site from making its visitors' browsers store reports (see
+  [CORS](#cors)).
+- **Body limit.** A report body may be at most 1 MiB, the backend's limit for
+  any request. Run on its own, the backend answers 413 before the route to a
+  request that declares a longer body. Without a declared length, the route
+  answers 413 only when the report it reads runs past 1 MiB; nothing after the
+  end of the report is read, so a longer body whose report ends sooner is
+  stored, and one that stops being valid JSON sooner is answered 400. In the
+  image, nginx refuses a body over 1 MiB first, with or without a declared
+  length, as its `client_max_body_size`, which `configs/nginx.conf` does not
+  set, is 1 MiB by default. nginx answers 413 with its own HTML error page, and
+  the backend never sees the request.
 - **Rate limit.** Each client address, resolved through `TRUSTED_PROXIES`, may
  send `REPORTS_PER_MINUTE` reports a minute; past that it gets 429 with
  `Retry-After: 60`. The minute slides: reports from the minute before still
@@ -246,16 +304,22 @@ credentials, so it is bounded instead. Both refusals below answer with the same
  `X-RateLimit-Reset` headers.
 - **Size cap.** The report files in `DATA_DIR` may total at most
  `DATA_DIR_MAX_BYTES`, counting the files already there at start. Reports
-  waiting in memory count at their uncompressed size until they are written;
-  those lost to a failed write stop counting, and the part of its file written
-  is removed. When a report would take the total past the cap, the oldest report
-  files are deleted to make room, and each deletion is logged with the file's
-  name and size; a file still being written is never deleted. A report is
-  refused with 507, and nothing of it is stored, only when the reports waiting
-  to be written fill the cap on their own, and then no file is deleted. At
-  start, report files past the cap, as after lowering it, are deleted the same
-  way. So the cap is how much of the newest reports is kept: the default of 1
-  GiB is small enough for any host; set it to the space you can give `DATA_DIR`.
+  waiting in memory count at their uncompressed size until they are written,
+  those of a failed write included; the part of its file written is removed.
+  When a report would take the total past the cap, the oldest report files are
+  deleted to make room, and each deletion is logged with the file's name and
+  size; a file still being written is never deleted. A report is refused with
+  507, and nothing of it is stored, only when the reports waiting to be written
+  fill the cap on their own, and then no file is deleted. At start, report files
+  past the cap, as after lowering it, are deleted the same way. So the cap is
+  how much of the newest reports is kept: the default of 1 GiB is small enough
+  for any host; set it to the space you can give `DATA_DIR`.
+
+At the defaults, one address may send 60 reports of up to 1 MiB a minute, and a
+report made up to compress badly takes about as much room in the report files as
+its request body, so in about 17 minutes one address can fill the 1 GiB cap with
+its own reports, every report stored before them deleted to make room. Lowering
+`REPORTS_PER_MINUTE` or raising `DATA_DIR_MAX_BYTES` makes that take longer.

 #### CORS

@@ -278,12 +342,16 @@ the metrics record its duration and response size, labelled with its path,
 method and status; they also count those requests in progress, and include Go's
 runtime and process metrics. No other request is recorded: not those to
 `/metrics` itself, and not those answered before they reach either route, such
-as a CORS preflight, or a request refused with 404 for a path no route has, 405
-for a method its route does not take, or 413 for declaring a body length over
-the 1 MiB limit. A report whose body goes over the limit without declaring its
-length reaches the route, is answered 413 there, and is recorded with that
-status. Clients can make up any number of paths and methods, and each would add
-labels to the metrics for as long as the server runs.
+as a CORS preflight, or a request refused with 404 for a path no route has or
+405 for a method its route does not take. Run on its own, the backend also
+answers 413 before the route to a request declaring a body length over the 1 MiB
+limit, which is not recorded. A body over the limit sent without a declared
+length reaches the route and is recorded with the status the route answers,
+which is 413 only when the report it reads runs past the limit: nothing after
+the end of the report is read. In the image neither reaches the backend: nginx
+answers both with 413 itself (see [Report limits](#report-limits)), and nothing
+of them is recorded. Clients can make up any number of paths and methods, and
+each would add labels to the metrics for as long as the server runs.

 `/metrics` is rate limited, so that its password cannot be guessed quickly: each
 client address, resolved through `TRUSTED_PROXIES`, may make 60 requests to it a
@@ -406,11 +474,15 @@ address nginx passes on and no other. The image:

 - Listens on port 8080 by default (override with `PORT` env var)
 - Takes the client address from `X-Forwarded-For` only on requests from the
-  reverse proxies named in `TRUSTED_PROXIES`, and by default from none.
-  `bin/entrypoint.sh` checks each entry with `netwatch-server check-cidr`, which
-  parses it as the backend parses its own `TRUSTED_PROXIES`
+  reverse proxies named in `TRUSTED_PROXIES`, by default from any address in
+  `10.0.0.0/8`, `172.16.0.0/12` or `192.168.0.0/16`. `bin/entrypoint.sh` checks
+  each entry with `netwatch-server check-cidr`, which parses it as the backend
+  parses its own `TRUSTED_PROXIES`
 - Sends access logs to stdout
- Caches static assets with immutable headers
+- Sends the page with `Cache-Control: no-cache`, so a browser that keeps a copy
+  checks with the server that it is current before using it again; only the
+  files under `/assets/`, whose names change whenever their contents do, are
+  cached as immutable
 - Sends the security headers `REPO_POLICIES.md` requires on every response, as
  `configs/security-headers.conf` sets them, in place of the backend's own
 - Stores reports in `DATA_DIR`, `/data/reports` by default, on the `/data`
@@ -427,9 +499,9 @@ What the [upaas](https://git.eeqj.de/sneak/upaas) app for netwatch needs:

 - **Port:** container port `8080`.
 - **Volume:** container path `/data`; the reports are kept in `/data/reports`.
- **Environment variables:** none is required. An empty one counts as unset, and
-  one set to a value netwatch cannot use stops the container at start, with the
-  reason in its log.
+- **Environment variables:** none is required. An empty one counts as unset,
+  except `TRUSTED_PROXIES`, and one set to a value netwatch cannot use stops the
+  container at start, with the reason in its log.
    - `PORT`, default `8080`: the container port, from 1 to 65535. `8081` cannot
      be used: the backend listens on it inside the container
    - `REPORTS_PER_MINUTE`, default `60`: reports each client address may send a
@@ -443,16 +515,17 @@ What the [upaas](https://git.eeqj.de/sneak/upaas) app for netwatch needs:
      in: `/data` or a path below it, with no `.` or `..` part and no extra `/`.
      The container also stops if the path goes through a symbolic link that
      leads out of `/data` or is written as a full path
-    - `TRUSTED_PROXIES`, default empty: set it to the address the reverse proxy
-      in front of the container connects from, as an IP address or CIDR; several
-      are separated by commas. nginx takes the client address from
-      `X-Forwarded-For` only on a request from one of them, and the rate limit
-      counts that address. Unset, `X-Forwarded-For` is ignored and every client
-      behind the proxy shares the proxy's one allowance of `REPORTS_PER_MINUTE`.
-      Name only addresses nothing but the proxy connects from: any client that
-      connects from one can write its own `X-Forwarded-For`, and through a port
-      Docker publishes, every client may connect from the Docker network's
-      gateway, such as `172.17.0.1`.
+    - `TRUSTED_PROXIES`, default `10.0.0.0/8,172.16.0.0/12,192.168.0.0/16`: the
+      addresses the reverse proxy in front of the container connects from, as IP
+      addresses or CIDRs separated by commas. nginx takes the client address
+      from `X-Forwarded-For` only on a request from one of them, and the rate
+      limit counts that address. A list replaces the default. Set and empty,
+      `X-Forwarded-For` is ignored and every client behind the proxy shares the
+      proxy's one allowance of `REPORTS_PER_MINUTE`. Any client that connects
+      from one of these addresses can write its own `X-Forwarded-For`, and
+      through a port Docker publishes, every client may connect from the Docker
+      network's gateway, such as `172.17.0.1`, which the default includes; where
+      that matters, name only the addresses nothing but the proxy connects from.
    - `METRICS_USERNAME` and `METRICS_PASSWORD`, default empty: with both set,
      the backend records Prometheus metrics of its requests and serves them at
      `/metrics` on the container port, to requests with this user name and
@@ -21,6 +21,188 @@ Run `make frontend-viewport-test` in CI as its own step; it is not part of

 # Completed Steps

+- 2026-10-19: report file writes no longer overlap
+  ([#150](https://git.eeqj.de/sneak/netwatch/issues/150)), so the reports of
+  failed writes are written in the order they arrived. Two writes that failed at
+  the same time put their reports back in the wrong order when the older failed
+  first. One write runs at a time, and each takes the waiting reports only when
+  it starts, after a failed write has put its own back. The write `Append`
+  starts when 10 MiB are waiting waits for any write under way, and `Append`
+  starts no other while it waits. Files written since the start are deleted to
+  make room in the order they were written, after those found at start. A unit
+  test holds a write open, starts a second at 10 MiB, fails both and checks the
+  order in the next file. Others check that `Append` starts no second write at
+  10 MiB while one waits, that it starts one again the next time 10 MiB are
+  waiting, and that the first file written since the start is the first deleted
+- 2026-10-19: the three unit tests of a check whose result is discarded (its
+  round given up, the user paused, or the first round) also check that the
+  host's history is still empty, its status still `pending`, and nothing was
+  written to the console
+  ([#151](https://git.eeqj.de/sneak/netwatch/issues/151)). They checked only the
+  row, so a change that recorded the result and skipped only the row would have
+  kept them passing
+- 2026-10-18: when the backend runs on its own, a `DATA_DIR` that cannot be
+  created or read, or where no file can be created, stops the start with an
+  error naming `DATA_DIR` and its value, as the other settings' errors do
+  ([#148](https://git.eeqj.de/sneak/netwatch/issues/148)); the first two named
+  only the path, and the third was found only at the first write of a report
+  file. At start the server creates a file in `DATA_DIR` and removes it. Unit
+  tests start the buffer with `DATA_DIR` below a regular file and with
+  `DATA_DIR` set to `/proc`, and check the error. The report file number is an
+  `atomic.Int64`, as the Go style guide asks for int types
+  ([#161](https://git.eeqj.de/sneak/netwatch/issues/161)); file names are
+  unchanged
+- 2026-10-18: the image's nginx sends the page with `Cache-Control: no-cache`
+  ([#141](https://git.eeqj.de/sneak/netwatch/issues/141)), so a browser that
+  keeps a copy checks with the server that it is current before using it again.
+  The page went out with no `Cache-Control`, so a browser could keep it after a
+  deploy and ask for files under `/assets/` that the new build no longer has,
+  and show an empty page. A path that `location /` answers with the page through
+  its `try_files` fallback gets the same header. `location /` now includes the
+  security headers again, as `/assets/` does, since its own `add_header` drops
+  the server's. `/assets/` keeps its immutable caching.
+  `make frontend-viewport-test` fails unless `/` and a path that falls back to
+  the page carry `Cache-Control: no-cache` and `Content-Security-Policy`, which
+  comes from the same file as the other security headers
+- 2026-10-18: `README.md` says what the page loses while reports are refused or
+  the page is paused ([#145](https://git.eeqj.de/sneak/netwatch/issues/145)),
+  how fast one address can replace the stored reports
+  ([#146](https://git.eeqj.de/sneak/netwatch/issues/146)), and that in the image
+  nginx answers a body over 1 MiB itself
+  ([#147](https://git.eeqj.de/sneak/netwatch/issues/147)). A sample in the
+  history that no report has delivered is sent only as long as the history holds
+  it: at the 1-second interval one refused or failed report loses the oldest 20
+  seconds of each host's samples, and a pause longer than the history loses
+  every sample between the last delivered report and the pause. Samples an
+  interval change cleared are kept apart, the last 100 per host, until they are
+  delivered. At the defaults, one address sending 1 MiB reports made up to
+  compress badly can fill the 1 GiB cap in about 17 minutes. nginx's
+  `client_max_body_size` is also 1 MiB by default, so in the image it answers a
+  larger body with 413 and its own HTML page, and the backend never sees or
+  records it; the backend's own 413 sentences now say they apply to it run on
+  its own. The reporter, the body limit and nginx are unchanged
+- 2026-10-18: a browser that cannot format the header's clocks still probes
+  ([#149](https://git.eeqj.de/sneak/netwatch/issues/149)). The clocks name the
+  local time zone through `Intl.DateTimeFormat`'s `shortGeneric` option, which
+  Chrome before 95, Firefox before 91 and Safari before 15.4 reject with an
+  error; it came before the first round started, so such a browser never checked
+  a target. The clocks now start last, after the rounds, and an error formatting
+  them leaves both blank. The footer's color legend reads `>=500ms` for red,
+  where it read `>500ms`, so 500 ms, which is red, is in one of its colors
+  ([#152](https://git.eeqj.de/sneak/netwatch/issues/152)). Unit tests check the
+  clocks with a working `Intl.DateTimeFormat` and with one that rejects
+  `shortGeneric`, and draw the page and check the legend against every entry of
+  `CONFIG.latencyColors`
+- 2026-10-18: `netwatch-server prepare-data-dir` no longer changes the owner of
+  a file outside `/data` through a hard link in the volume
+  ([#143](https://git.eeqj.de/sneak/netwatch/issues/143)). A file with a second
+  name (hard link) keeps its owner, and the start logs its path. Owners are now
+  set on the opened directory, file or link, never by name, so a hard link
+  swapped in after the check is caught too; a symbolic link is opened without
+  following it and given to the netwatch user itself, so the server can still
+  follow one in a sticky directory that anyone can write to. Devices, named
+  pipes and sockets keep their owner. Setting the owner on an open file that is
+  not a directory is built only for Linux: `prepare-data-dir` runs only in the
+  Linux image, and macOS, where the backend is also built for development, has
+  neither `O_PATH` nor `AT_EMPTY_PATH`. On any other system the package still
+  builds, and `prepare-data-dir` fails at the first file that is not a
+  directory. The test that runs as root hard-links a file, a symbolic link and a
+  named pipe from outside the volume into it and finds their owners unchanged
+- 2026-10-18: from a trusted proxy, the backend takes the client address from
+  the right-most `X-Forwarded-For` entry that is not in `TRUSTED_PROXIES`
+  ([#144](https://git.eeqj.de/sneak/netwatch/issues/144)). It took the left-most
+  entry, which the client writes itself behind a proxy that appends to the
+  header, so a client could choose the address the rate limit counts and the
+  request log records. An entry written as an address with a port is read as
+  that address. With only trusted addresses in the header, or an entry that is
+  not an address at all (such as `unknown`) met on the way, the direct peer is
+  used. `X-Real-IP` is read only without `X-Forwarded-For`, and every
+  `X-Forwarded-For` header line is read, not only the first. The image's nginx,
+  and the `TRUSTED_PROXIES=127.0.0.1/32` it starts the backend with, are
+  unchanged; the comment in `configs/nginx.conf` that said the backend takes the
+  first entry now says it takes the one address nginx passes on. Unit tests
+  resolve the address from chains ending in an untrusted address, in trusted
+  ones, of only trusted ones, with ports, with an entry that is not an address,
+  and split over two header lines
+- 2026-10-18: after an interval change, the summary's Reachable slot reads
+  `--/--` in white, as at page load, until the second round since the change
+  ends ([#140](https://git.eeqj.de/sneak/netwatch/issues/140)). It used to read
+  0/26 in red from the change on, as no WAN host had a check recorded, for up to
+  two intervals with the network fine. The summary update now writes `--/--`
+  while every WAN host still waits for its first check. A unit test resets the
+  hosts as the interval menu does and checks the slot; another checks that it
+  reads 0/26 in red once every WAN host's check has failed
+- 2026-10-14: stopping the server waits for a report file write already under
+  way, the periodic flush's or the one `Append` starts when 10 MiB are waiting,
+  before the final flush
+  ([#122](https://git.eeqj.de/sneak/netwatch/issues/122)). It used to return at
+  once, and the process exiting cut the write off, losing its reports. A write
+  under way that fails puts its reports back in memory, so the final flush
+  writes them; the stop fails, and the process exits non-zero, only if that
+  flush fails too. Two tests hold the periodic write and the one at 10 MiB open
+  and find the stop waiting until each is released; a third makes the held write
+  fail and finds its report written by the final flush. They run in a
+  `testing/synctest` bubble, so the periodic flush comes without waiting a
+  minute
+- 2026-10-14: `make test` no longer runs the frontend's production build, which
+  the tests do not need ([#118](https://git.eeqj.de/sneak/netwatch/issues/118)).
+  `yarn build` moved out of the stage that runs the frontend's unit tests, now
+  named `frontend-test`, into a new `frontend-build` stage, which starts from it
+  and, like the `builder` stage, waits on the lint and test phases, so the image
+  still cannot be built unless they pass. The runtime stage copies `dist/` from
+  it, and the `git` that `vite.config.js` reads the commit with is installed
+  there. No build cache was added
+- 2026-10-14: opening the debug log no longer raises the measured latencies
+  ([#131](https://git.eeqj.de/sneak/netwatch/issues/131)). With the panel open,
+  each line written drew the panel again with every line before it, up to 1000,
+  and a check that ends while the page draws is timed only when the drawing is
+  done. Each line is now added at the end of the open panel, its first line is
+  dropped once it holds more than 1000, and it scrolls to its newest line once a
+  frame rather than once a line; the whole panel is drawn only when it is
+  opened. A unit test writes more lines than the panel keeps and checks that the
+  lines already shown are left as they are
+- 2026-10-14: unset, the backend's own `TRUSTED_PROXIES` trusts `10.0.0.0/8`,
+  `172.16.0.0/12` and `192.168.0.0/16`, and no longer loopback; a list replaces
+  that; set and empty, it trusts no proxy, where it used to get the default, as
+  viper reads an empty environment variable as unset
+  ([#129](https://git.eeqj.de/sneak/netwatch/issues/129)). A value it cannot
+  parse still stops the start, and the image still starts the backend with
+  `TRUSTED_PROXIES=127.0.0.1/32`. Tests load the config with it unset, empty and
+  set to a list
+- 2026-10-14: while report files cannot be written, `POST /api/v1/reports`
+  answers 500, and storing recovers once writing works again, with no restart
+  ([#123](https://git.eeqj.de/sneak/netwatch/issues/123)). A failed write puts
+  its reports back in the report buffer, still counted against
+  `DATA_DIR_MAX_BYTES`, where they used to be lost; the next flush, within a
+  minute, tries them again. Until a write succeeds, every report is refused and
+  nothing of it is kept, so the buffer does not grow while writes fail. A test
+  posts reports through the report handler before, during and after a failure;
+  others check that a report appended while a write runs is kept, behind that
+  write's own, when the write fails, and that the cap still holds exactly once a
+  failed write's reports are written
+- 2026-10-14: a request whose handler panics is logged with status 500 and
+  counted with `code` 500 in the metrics
+  ([#125](https://git.eeqj.de/sneak/netwatch/issues/125)), not 200. `Recoverer`
+  is still the first middleware; `Write500OnPanic`, after the request log and
+  after the metrics, writes the 500 as the panic passes and lets it go on to
+  Sentry and `Recoverer`, which now writes a 500 only when no status was written
+- 2026-10-14: with `TRUSTED_PROXIES` unset, the container's nginx trusts the
+  three private ranges of RFC 1918, `10.0.0.0/8`, `172.16.0.0/12` and
+  `192.168.0.0/16` ([#119](https://git.eeqj.de/sneak/netwatch/issues/119)), as
+  sneak ruled on 2026-09-22 for every app; it trusted no proxy before. A list
+  replaces them, and set and empty it trusts none. The backend inside the
+  container still trusts only `127.0.0.1/32`
+- 2026-10-14: `POST /api/v1/reports` takes only a report sent with
+  `Content-Type: application/json`, parameters such as `charset` allowed, and
+  refuses any other with 415 and the usual error body, storing nothing
+  ([#126](https://git.eeqj.de/sneak/netwatch/issues/126)). A page on another
+  origin can send a `text/plain` POST without a CORS preflight, so any site
+  could make its visitors' browsers store reports; a JSON POST waits for the
+  preflight, which only an origin in `CORS_ALLOWED_ORIGINS` passes. The page
+  already sent `application/json`. A test posts a report as `text/plain`, as
+  `application/x-www-form-urlencoded`, as `multipart/form-data`, with no
+  `Content-Type`, and as `application/json` with and without a charset, and
+  checks which are stored
 - 2026-10-14: changing the interval no longer drops the samples not yet reported
  to the backend ([#124](https://git.eeqj.de/sneak/netwatch/issues/124)). The
  interval menu resets each host with the new `HostState.reset`, which moves the
@@ -452,3 +634,9 @@ Run `make frontend-viewport-test` in CI as its own step; it is not part of
 - A configurable host list (an environment variable or a config file)
 - Export of the latency history (CSV or JSON)
 - A notification when the health status changes to DEGRADED
+- `make test` within the 20-second target
+  ([#118](https://git.eeqj.de/sneak/netwatch/issues/118)): every run compiles
+  the Go standard library and every dependency with the race detector from
+  nothing. Whether test builds may keep Go's compiled packages between runs
+  waits on sneak's ruling in
+  [prompts#113](https://git.eeqj.de/sneak/prompts/issues/113)
@@ -34,9 +34,11 @@ fi

 # TRUSTED_PROXIES names the reverse proxies in front of the container,
 # as IP addresses or CIDRs separated by commas. nginx takes the client
-# address from X-Forwarded-For only on a request from one of them, so
-# unset or empty, it trusts no one. configs/nginx.conf includes the file
-# written here, one set_real_ip_from line per entry.
+# address from X-Forwarded-For only on a request from one of them.
+# Unset, they are the three private ranges of RFC 1918; set and empty,
+# nginx trusts no one, which is why the default below follows - and not
+# :-. configs/nginx.conf includes the file written here, one
+# set_real_ip_from line per entry.
 #
 # nginx looks up an entry it cannot read as an address as a hostname,
 # and trusts what it finds (1.2.3 is found as 1.2.0.3). So each entry
@@ -45,7 +47,7 @@ fi
 # TRUSTED_PROXIES. Its error, naming the CIDR, is dropped for the one
 # below, naming the entry as written. set -f keeps a * in an entry from
 # becoming a list of file names.
-TRUSTED_PROXIES="${TRUSTED_PROXIES:-}"
+TRUSTED_PROXIES="${TRUSTED_PROXIES-10.0.0.0/8,172.16.0.0/12,192.168.0.0/16}"
 set -f
 for proxy in $(printf '%s' "$TRUSTED_PROXIES" | tr ',' ' '); do
    case "$proxy" in
@@ -64,10 +66,13 @@ done > /etc/nginx/trusted-proxies.conf
 # netwatch-server keeps its report files in DATA_DIR, on the /data
 # volume, which may be a host directory owned by root or by another
 # uid. Here, as root, netwatch-server prepare-data-dir creates DATA_DIR
-# and gives /data and everything in it to the netwatch user, so the
-# host directory needs no preparing. It stops the start, naming
-# DATA_DIR, unless DATA_DIR is /data or a path below it, and it acts on
-# nothing outside /data, whatever symbolic links it meets there.
+# and gives /data and the directories, regular files and symbolic links
+# in it to the netwatch user, so the host directory needs no preparing.
+# A file with a second name (hard link) keeps its owner, as that name
+# could be outside /data, and its path is logged. It stops the start,
+# naming DATA_DIR, unless DATA_DIR is /data or a path below it, and it
+# acts on nothing outside /data, whatever symbolic or hard links it
+# meets there.
 export DATA_DIR="${DATA_DIR:-/data/reports}"
 netwatch-server prepare-data-dir "$DATA_DIR" || exit 1

@@ -18,9 +18,10 @@ server {

    # The client address comes from X-Forwarded-For only on a request
    # from the reverse proxies in TRUSTED_PROXIES: bin/entrypoint.sh
-    # writes one set_real_ip_from line for each into this file, and
-    # leaves it empty when TRUSTED_PROXIES is unset, so that by default
-    # the client address is the one each request comes from.
+    # writes one set_real_ip_from line for each into this file. Unset,
+    # TRUSTED_PROXIES is the three private ranges of RFC 1918; set and
+    # empty, the file is empty and the client address is the one each
+    # request comes from.
    include /etc/nginx/trusted-proxies.conf;
    real_ip_header X-Forwarded-For;
    real_ip_recursive on;
@@ -29,8 +30,13 @@ server {
    access_log /dev/stdout combined;
    error_log  /dev/stderr warn;

+    # With no-cache, a browser never uses a copy of the page that names
+    # assets a newer build no longer has without first checking with the
+    # server that it is current.
    location / {
        try_files $uri $uri/ /index.html;
+        add_header Cache-Control no-cache;
+        include /etc/nginx/security-headers.conf;
    }

    # Cache static assets aggressively
@@ -45,7 +51,9 @@ server {
    # 127.0.0.1:8081. These headers go with every request passed to it.
    # X-Forwarded-For carries only the client address, as resolved by
    # the real IP settings above, and not the chain the request came
-    # with: the backend takes the first entry, which a client can write.
+    # with. bin/entrypoint.sh starts the backend with
+    # TRUSTED_PROXIES=127.0.0.1/32, the address nginx connects from, so
+    # the backend takes that one entry as the client address.
    proxy_set_header Host $host;
    proxy_set_header X-Real-IP $remote_addr;
    proxy_set_header X-Forwarded-For $remote_addr;
@@ -17,6 +17,7 @@ require (
 	github.com/slok/go-http-metrics v0.13.0
 	github.com/spf13/viper v1.21.0
 	go.uber.org/fx v1.24.0
+	golang.org/x/sys v0.47.0
 )

 require (
@@ -41,7 +42,6 @@ require (
 	go.uber.org/multierr v1.10.0 // indirect
 	go.uber.org/zap v1.26.0 // indirect
 	go.yaml.in/yaml/v3 v3.0.4 // indirect
-	golang.org/x/sys v0.47.0 // indirect
 	golang.org/x/text v0.40.0 // indirect
 	google.golang.org/protobuf v1.36.11 // indirect
 )
@@ -9,6 +9,7 @@ import (
 	"math"
 	"net/netip"
 	"net/url"
+	"os"
 	"strconv"
 	"strings"

@@ -21,13 +22,11 @@ import (
 )

 // defaultTrustedProxies lists the networks whose forwarded
-// headers are honoured by default: IPv4 and IPv6 loopback,
-// for a reverse proxy on the same host, and the RFC1918
-// ranges. The container image does not use it:
+// headers are honoured when TRUSTED_PROXIES is unset: the
+// RFC 1918 ranges. The container image does not use it:
 // bin/entrypoint.sh gives the server 127.0.0.1/32, since
 // nginx is its only client there.
-const defaultTrustedProxies = "127.0.0.1/32,::1/128," +
-	"10.0.0.0/8,172.16.0.0/12,192.168.0.0/16"
+const defaultTrustedProxies = "10.0.0.0/8,172.16.0.0/12,192.168.0.0/16"

 // Default limits on stored reports; README.md gives the reasons for
 // these values.
@@ -135,6 +134,14 @@ func New(
 			viper.GetString("DEBUG"), errNotBool)
 	}

+	// An empty TRUSTED_PROXIES trusts no proxy, but viper reads an
+	// empty environment variable as unset and would give the default,
+	// so the environment is read first.
+	trustedProxies, set := os.LookupEnv("TRUSTED_PROXIES")
+	if !set {
+		trustedProxies = viper.GetString("TRUSTED_PROXIES")
+	}
+
 	s := &Config{
 		BindAddress:        viper.GetString("BIND_ADDRESS"),
 		CORSAllowedOrigins: splitList(viper.GetString("CORS_ALLOWED_ORIGINS")),
@@ -146,7 +153,7 @@ func New(
 		Port:               port,
 		ReportsPerMinute:   viper.GetInt("REPORTS_PER_MINUTE"),
 		SentryDSN:          viper.GetString("SENTRY_DSN"),
-		TrustedProxies:     splitList(viper.GetString("TRUSTED_PROXIES")),
+		TrustedProxies:     splitList(trustedProxies),
 		log:                log,
 		params:             &params,
 	}
@@ -1,6 +1,8 @@
 package config_test

 import (
+	"os"
+	"slices"
 	"strings"
 	"testing"

@@ -29,6 +31,62 @@ func requireConfigError(t *testing.T, setting string) {
 	}
 }

+// requireTrustedProxies builds the config as main does and fails the
+// test unless its TrustedProxies are want.
+func requireTrustedProxies(t *testing.T, want []string) {
+	t.Helper()
+
+	var cfg *config.Config
+
+	app := fx.New(
+		fx.NopLogger,
+		fx.Provide(globals.New, logger.New, config.New),
+		fx.Populate(&cfg),
+	)
+
+	err := app.Err()
+	if err != nil {
+		t.Fatalf("config error = %v", err)
+	}
+
+	if !slices.Equal(cfg.TrustedProxies, want) {
+		t.Fatalf("TrustedProxies = %q, want %q", cfg.TrustedProxies, want)
+	}
+}
+
+// TestTrustedProxiesUnsetTrustsRFC1918: unset, the server trusts the
+// RFC 1918 ranges and nothing else, loopback included.
+func TestTrustedProxiesUnsetTrustsRFC1918(t *testing.T) {
+	// t.Setenv first, so the variable is put back as it was after the
+	// test.
+	t.Setenv("TRUSTED_PROXIES", "")
+
+	err := os.Unsetenv("TRUSTED_PROXIES")
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	requireTrustedProxies(t,
+		[]string{"10.0.0.0/8", "172.16.0.0/12", "192.168.0.0/16"})
+}
+
+// TestTrustedProxiesEmptyTrustsNothing: set and empty, the server
+// trusts no proxy. viper alone would read the empty variable as unset
+// and give the default.
+func TestTrustedProxiesEmptyTrustsNothing(t *testing.T) {
+	t.Setenv("TRUSTED_PROXIES", "")
+
+	requireTrustedProxies(t, []string{})
+}
+
+// TestTrustedProxiesListReplacesDefault: a list is used as given, with
+// none of the default ranges added.
+func TestTrustedProxiesListReplacesDefault(t *testing.T) {
+	t.Setenv("TRUSTED_PROXIES", "172.17.0.1/32, 2001:db8::/32")
+
+	requireTrustedProxies(t, []string{"172.17.0.1/32", "2001:db8::/32"})
+}
+
 // TestSettingsLoadAsGiven: valid values pass the checks and are used
 // as given. bin/entrypoint.sh starts the server with these
 // BIND_ADDRESS and PORT values.
@@ -5,7 +5,9 @@ package handlers
 import (
 	"context"
 	"encoding/json"
+	"errors"
 	"log/slog"
+	"mime"
 	"net/http"

 	"sneak.berlin/go/netwatch/internal/globals"
@@ -18,6 +20,10 @@ import (

 const jsonContentType = "application/json; charset=utf-8"

+// errNotJSON is what decodeJSON returns for a request whose
+// Content-Type is not application/json.
+var errNotJSON = errors.New("content type is not application/json")
+
 // reportAppender is the subset of the report buffer the handlers
 // depend on. Defining it here keeps the storage failure path
 // exercisable with a stub in tests.
@@ -80,14 +86,25 @@ func (s *Handlers) respondJSON(
 	}
 }

-// decodeJSON decodes the request body into v. The body is
-// expected to already be bounded by the body-size middleware, so
-// a caller can distinguish an over-limit body from malformed
-// JSON by testing the returned error for *http.MaxBytesError.
+// decodeJSON decodes the request body into v. It returns errNotJSON,
+// reading nothing, unless the request's Content-Type is
+// application/json, with or without parameters such as charset. The
+// body is expected to already be bounded by the body-size middleware,
+// so a caller can distinguish an over-limit body from malformed JSON
+// by testing the returned error for *http.MaxBytesError.
 func (s *Handlers) decodeJSON(
 	_ http.ResponseWriter,
 	r *http.Request,
 	v any,
 ) error {
+	// A page on another origin can POST text/plain or a form without
+	// a CORS preflight, so its visitors' browsers would send it
+	// whatever the CORS settings. A JSON POST waits for the
+	// preflight, so only JSON is taken.
+	mediaType, _, err := mime.ParseMediaType(r.Header.Get("Content-Type"))
+	if err != nil || mediaType != "application/json" {
+		return errNotJSON
+	}
+
 	return json.NewDecoder(r.Body).Decode(v)
 }
@@ -30,7 +30,7 @@ type report struct {
 }

 // HandleReport returns a handler that accepts telemetry
-// reports from NetWatch clients.
+// reports from NetWatch clients, sent as application/json.
 func (s *Handlers) HandleReport() http.HandlerFunc {
 	type response struct {
 		Status string `json:"status"`
@@ -66,9 +66,15 @@ func (s *Handlers) HandleReport() http.HandlerFunc {
 }

 // decodeErrorStatus logs a report decode failure and returns the
-// status to send: 413 when the body exceeded the size limit,
-// otherwise 400 for malformed JSON.
+// status to send: 415 when the report was not sent as JSON, 413 when
+// the body exceeded the size limit, otherwise 400 for malformed JSON.
 func (s *Handlers) decodeErrorStatus(err error) int {
+	if errors.Is(err, errNotJSON) {
+		s.log.Warn("report refused: Content-Type is not application/json")
+
+		return http.StatusUnsupportedMediaType
+	}
+
 	var tooLarge *http.MaxBytesError
 	if errors.As(err, &tooLarge) {
 		s.log.Warn("report body too large", "limit_bytes", tooLarge.Limit)
@@ -87,7 +93,7 @@ func (s *Handlers) decodeErrorStatus(err error) int {

 // appendErrorStatus logs a failure to store a report and returns
 // the status to send: 507 when the reports waiting to be written fill
-// the size cap, otherwise 500.
+// the size cap, otherwise 500, as when report files cannot be written.
 func (s *Handlers) appendErrorStatus(err error) int {
 	if errors.Is(err, reportbuf.ErrFull) {
 		s.log.Warn("report refused: " +
@@ -27,6 +27,17 @@ type stubAppender struct {

 func (s stubAppender) Append(any) error { return s.err }

+// recordingAppender keeps every report the handler stores.
+type recordingAppender struct {
+	reports []any
+}
+
+func (a *recordingAppender) Append(v any) error {
+	a.reports = append(a.reports, v)
+
+	return nil
+}
+
 func newTestHandlers(buf stubAppender, out io.Writer) *handlers.Handlers {
 	return handlers.NewForTest(buf, slog.New(slog.NewJSONHandler(out, nil)))
 }
@@ -83,6 +94,7 @@ func TestHandleReportAcceptsValidReports(t *testing.T) {
 				http.MethodPost, "/api/v1/reports",
 				strings.NewReader(tt.body),
 			)
+			req.Header.Set("Content-Type", "application/json")

 			h.HandleReport().ServeHTTP(rec, req)

@@ -103,6 +115,89 @@ func TestHandleReportAcceptsValidReports(t *testing.T) {
 	}
 }

+// TestHandleReportStoresOnlyJSON checks that a report is stored only
+// when sent as application/json, as the page sends it. One sent as
+// text/plain or as a form, which a page on another origin can send
+// without a CORS preflight, is refused with 415.
+func TestHandleReportStoresOnlyJSON(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name        string
+		contentType string
+		wantStatus  int
+		wantStored  int
+	}{
+		{
+			name:        "text/plain",
+			contentType: "text/plain;charset=UTF-8",
+			wantStatus:  http.StatusUnsupportedMediaType,
+			wantStored:  0,
+		},
+		{
+			name:        "a urlencoded form",
+			contentType: "application/x-www-form-urlencoded",
+			wantStatus:  http.StatusUnsupportedMediaType,
+			wantStored:  0,
+		},
+		{
+			name:        "a multipart form",
+			contentType: "multipart/form-data; boundary=x",
+			wantStatus:  http.StatusUnsupportedMediaType,
+			wantStored:  0,
+		},
+		{
+			name:        "no Content-Type",
+			contentType: "",
+			wantStatus:  http.StatusUnsupportedMediaType,
+			wantStored:  0,
+		},
+		{
+			name:        "application/json",
+			contentType: "application/json",
+			wantStatus:  http.StatusOK,
+			wantStored:  1,
+		},
+		{
+			name:        "application/json with a charset",
+			contentType: "application/json;charset=UTF-8",
+			wantStatus:  http.StatusOK,
+			wantStored:  1,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			var stored recordingAppender
+
+			h := handlers.NewForTest(&stored, slog.New(slog.DiscardHandler))
+
+			rec := httptest.NewRecorder()
+			req := httptest.NewRequestWithContext(t.Context(),
+				http.MethodPost, "/api/v1/reports",
+				strings.NewReader(`{"clientId":"c1","hosts":[]}`),
+			)
+
+			if tt.contentType != "" {
+				req.Header.Set("Content-Type", tt.contentType)
+			}
+
+			h.HandleReport().ServeHTTP(rec, req)
+
+			if rec.Code != tt.wantStatus {
+				t.Errorf("status = %d, want %d", rec.Code, tt.wantStatus)
+			}
+
+			if len(stored.reports) != tt.wantStored {
+				t.Errorf("reports stored = %d, want %d",
+					len(stored.reports), tt.wantStored)
+			}
+		})
+	}
+}
+
 func TestHandleReportStorageFailureIsNon2xx(t *testing.T) {
 	t.Parallel()

@@ -113,6 +208,7 @@ func TestHandleReportStorageFailureIsNon2xx(t *testing.T) {
 		http.MethodPost, "/api/v1/reports",
 		strings.NewReader(`{"clientId":"c1","hosts":[]}`),
 	)
+	req.Header.Set("Content-Type", "application/json")

 	h.HandleReport().ServeHTTP(rec, req)

@@ -138,6 +234,7 @@ func TestHandleReportFullIs507(t *testing.T) {
 		http.MethodPost, "/api/v1/reports",
 		strings.NewReader(`{"clientId":"c1","hosts":[]}`),
 	)
+	req.Header.Set("Content-Type", "application/json")

 	h.HandleReport().ServeHTTP(rec, req)

@@ -161,6 +258,7 @@ func TestHandleReportMalformedJSONIs400(t *testing.T) {
 		http.MethodPost, "/api/v1/reports",
 		strings.NewReader(`{not json`),
 	)
+	req.Header.Set("Content-Type", "application/json")

 	h.HandleReport().ServeHTTP(rec, req)

@@ -185,6 +283,8 @@ func TestHandleReportOversizeIs413(t *testing.T) {
 		http.MethodPost, "/api/v1/reports",
 		strings.NewReader(`{"clientId":"`+strings.Repeat("x", 200)+`"}`),
 	)
+	req.Header.Set("Content-Type", "application/json")
+
 	// No declared length, so only the middleware's read cap can
 	// stop this body.
 	req.ContentLength = -1
@@ -213,6 +313,7 @@ func TestHandleReportDoesNotLogRawGeo(t *testing.T) {
 			`{"clientId":"c1","geo":{"raw":"`+sentinel+`"},"hosts":[]}`,
 		),
 	)
+	req.Header.Set("Content-Type", "application/json")

 	h.HandleReport().ServeHTTP(rec, req)

@@ -245,6 +346,7 @@ func TestHandleReportLogsClientIDCutToBound(t *testing.T) {
 			`{"clientId":"`+long+`","timestamp":"`+long+`","hosts":[]}`,
 		),
 	)
+	req.Header.Set("Content-Type", "application/json")

 	h.HandleReport().ServeHTTP(rec, req)

@@ -284,6 +386,7 @@ func TestHandleReportDecodeErrorLogIsBounded(t *testing.T) {
 		http.MethodPost, "/api/v1/reports",
 		strings.NewReader(`{"hosts":[{"history":[{"t":`+huge+`}]}]}`),
 	)
+	req.Header.Set("Content-Type", "application/json")

 	h.HandleReport().ServeHTTP(rec, req)

@@ -11,6 +11,7 @@ import (
 	"net/http"
 	"net/netip"
 	"runtime/debug"
+	"slices"
 	"strings"
 	"time"

@@ -133,6 +134,17 @@ func ipFromHostPort(hostPort string) string {
 // X-Real-IP are honoured only when the direct peer is a
 // trusted proxy; otherwise the direct peer is returned so a
 // spoofed header cannot forge the logged address.
+//
+// Each proxy appends to X-Forwarded-For the address it took the
+// request from, so the entries on the right are written by the
+// trusted proxies and those further left by the client. The
+// client's address is therefore the right-most entry that is not
+// a trusted proxy. An entry written as an address with a port is
+// read as that address, both for the trust check and as the
+// result. When every entry is a trusted proxy, or the walk from
+// the right meets an entry that is not an address at all (such as
+// "unknown"), the peer is used. X-Real-IP is read only when there
+// is no X-Forwarded-For entry at all.
 func clientIP(
 	remoteAddr string,
 	header http.Header,
@@ -144,28 +156,49 @@ func clientIP(
 		return peer
 	}

-	if xff := firstForwardedFor(header.Get("X-Forwarded-For")); xff != "" {
-		return xff
+	forwarded := forwardedFor(header)
+
+	if len(forwarded) == 0 {
+		if xr := strings.TrimSpace(header.Get("X-Real-IP")); validIP(xr) {
+			return xr
+		}
+
+		return peer
 	}

-	if xr := strings.TrimSpace(header.Get("X-Real-IP")); validIP(xr) {
-		return xr
+	for _, entry := range slices.Backward(forwarded) {
+		if !validIP(entry) {
+			return peer
+		}
+
+		if !addrInAny(entry, trusted) {
+			return entry
+		}
 	}

 	return peer
 }

-// firstForwardedFor returns the left-most valid address in an
-// X-Forwarded-For list (the original client), or "" if none.
-func firstForwardedFor(value string) string {
+// forwardedFor returns the entries of X-Forwarded-For in order,
+// leaving out blank ones. A proxy may add its entry as a header
+// line of its own rather than append it to the line the client
+// sent, so every line is read, not only the first. Some proxies
+// write an entry as an address with a port, such as
+// 198.51.100.7:5000 or [2001:db8::1]:443; the port is removed so
+// that the entry is read as its address.
+func forwardedFor(header http.Header) []string {
+	value := strings.Join(header.Values("X-Forwarded-For"), ",")
+
+	var entries []string
+
 	for part := range strings.SplitSeq(value, ",") {
-		candidate := strings.TrimSpace(part)
-		if validIP(candidate) {
-			return candidate
+		entry := strings.TrimSpace(part)
+		if entry != "" {
+			entries = append(entries, ipFromHostPort(entry))
 		}
 	}

-	return ""
+	return entries
 }

 func validIP(s string) bool {
@@ -297,12 +330,16 @@ func (s *Middleware) MaxBodyBytes(

 // Recoverer returns middleware that recovers from a panic in a
 // downstream handler, logs the panic and stack trace through
-// slog, and responds 500 with no body. http.ErrAbortHandler is
-// re-panicked so the server can abort the response as intended.
+// slog, and responds 500 with no body, unless a status was already
+// written, as Write500OnPanic writes one for a handler's panic.
+// http.ErrAbortHandler is re-panicked so the server can abort the
+// response as intended.
 func (s *Middleware) Recoverer() func(http.Handler) http.Handler {
 	return func(next http.Handler) http.Handler {
 		return http.HandlerFunc(
 			func(w http.ResponseWriter, r *http.Request) {
+				ww := middleware.NewWrapResponseWriter(w, r.ProtoMajor)
+
 				defer func() {
 					rec := recover()
 					if rec == nil {
@@ -320,10 +357,43 @@ func (s *Middleware) Recoverer() func(http.Handler) http.Handler {
 						"stack", string(debug.Stack()),
 					)

-					w.WriteHeader(http.StatusInternalServerError)
+					if ww.Status() == 0 {
+						ww.WriteHeader(http.StatusInternalServerError)
+					}
 				}()

-				next.ServeHTTP(w, r)
+				next.ServeHTTP(ww, r)
+			},
+		)
+	}
+}
+
+// Write500OnPanic returns middleware that writes 500 when a downstream
+// handler panics before writing a status, and lets the panic go on to
+// Sentry and Recoverer. Recoverer, which comes first, would write its
+// 500 only after the request log and the metrics had recorded the
+// status they saw; placed after each of them, this makes them record
+// the 500.
+func (s *Middleware) Write500OnPanic() func(http.Handler) http.Handler {
+	return func(next http.Handler) http.Handler {
+		return http.HandlerFunc(
+			func(w http.ResponseWriter, r *http.Request) {
+				ww := middleware.NewWrapResponseWriter(w, r.ProtoMajor)
+
+				defer func() {
+					rec := recover()
+					if rec == nil {
+						return
+					}
+
+					if ww.Status() == 0 {
+						ww.WriteHeader(http.StatusInternalServerError)
+					}
+
+					panic(rec)
+				}()
+
+				next.ServeHTTP(ww, r)
 			},
 		)
 	}
@@ -4,6 +4,7 @@ import (
 	"bytes"
 	"encoding/json"
 	"errors"
+	"log"
 	"log/slog"
 	"net/http"
 	"net/http/httptest"
@@ -16,16 +17,24 @@ import (
 	"sneak.berlin/go/netwatch/internal/logger"
 	"sneak.berlin/go/netwatch/internal/middleware"

+	"github.com/go-chi/chi/v5"
 	chimiddleware "github.com/go-chi/chi/v5/middleware"
+	"github.com/prometheus/client_golang/prometheus"
 )

 const (
 	// loopbackPeer is a remote address inside the trusted-proxy allowlist.
 	loopbackPeer = "127.0.0.1:5000"
+	// loopbackIP is the address of loopbackPeer, which ClientIP returns
+	// when it uses the peer.
+	loopbackIP = "127.0.0.1"
 	// forwardedIP is the client address presented via X-Forwarded-For.
 	forwardedIP = "203.0.113.7"
 	// realIP is the client address presented via X-Real-IP.
 	realIP = "203.0.113.9"
+	// appendedIP is the X-Forwarded-For entry a trusted proxy appended
+	// after the entries the client wrote.
+	appendedIP = "198.51.100.7"
 )

 func mustPrefixes(t *testing.T, cidrs ...string) []netip.Prefix {
@@ -82,10 +91,35 @@ func clientIPCases() []clientIPCase {
 			want:       forwardedIP,
 		},
 		{
-			name:       "trusted proxy uses left-most of chain",
-			remoteAddr: "10.1.2.3:5000",
-			xff:        forwardedIP + ", 10.1.2.3",
-			want:       forwardedIP,
+			name:       "trusted proxy uses right-most of chain",
+			remoteAddr: loopbackPeer,
+			xff:        "203.0.113.9, " + appendedIP,
+			want:       appendedIP,
+		},
+		{
+			name:       "trusted proxy entries are skipped from the right",
+			remoteAddr: loopbackPeer,
+			xff:        "203.0.113.9, 10.0.0.5",
+			want:       "203.0.113.9",
+		},
+		{
+			name:       "chain of only trusted proxies uses peer",
+			remoteAddr: loopbackPeer,
+			xff:        "10.0.0.5, 10.1.2.3",
+			want:       loopbackIP,
+		},
+		{
+			name:       "entry that is not an address uses peer",
+			remoteAddr: loopbackPeer,
+			xff:        "203.0.113.9, unknown",
+			want:       loopbackIP,
+		},
+		{
+			name:       "x-real-ip ignored when forwarded-for has entries",
+			remoteAddr: loopbackPeer,
+			xff:        "10.0.0.5",
+			xRealIP:    realIP,
+			want:       loopbackIP,
 		},
 		{
 			name:       "trusted proxy falls back to x-real-ip",
@@ -114,7 +148,7 @@ func clientIPCases() []clientIPCase {
 			name:       "trusted proxy with garbage header uses peer",
 			remoteAddr: loopbackPeer,
 			xff:        "not-an-ip",
-			want:       "127.0.0.1",
+			want:       loopbackIP,
 		},
 	}
 }
@@ -145,6 +179,47 @@ func TestClientIP(t *testing.T) {
 	}
 }

+// TestClientIPReadsEveryForwardedForLine checks that an entry a proxy
+// adds as an X-Forwarded-For line of its own is read, and the line the
+// client sent before it is not taken.
+func TestClientIPReadsEveryForwardedForLine(t *testing.T) {
+	t.Parallel()
+
+	header := http.Header{}
+	header.Add("X-Forwarded-For", "203.0.113.9")
+	header.Add("X-Forwarded-For", appendedIP)
+
+	got := middleware.ClientIP(loopbackPeer, header,
+		mustPrefixes(t, "127.0.0.1/32"))
+	if got != appendedIP {
+		t.Errorf("ClientIP() = %q, want %q", got, appendedIP)
+	}
+}
+
+// TestClientIPReadsAddressWithPort checks that an X-Forwarded-For entry
+// written as an address with a port is read as that address, both for
+// the trust check and as the result.
+func TestClientIPReadsAddressWithPort(t *testing.T) {
+	t.Parallel()
+
+	trusted := mustPrefixes(t, "127.0.0.1/32", "10.0.0.0/8")
+
+	for xff, want := range map[string]string{
+		"203.0.113.9, " + appendedIP + ":5000": appendedIP,
+		"203.0.113.9, [2001:db8::1]:443":       "2001:db8::1",
+		"203.0.113.9, 10.0.0.5:5000":           "203.0.113.9",
+	} {
+		header := http.Header{}
+		header.Set("X-Forwarded-For", xff)
+
+		got := middleware.ClientIP(loopbackPeer, header, trusted)
+		if got != want {
+			t.Errorf("X-Forwarded-For %q: ClientIP() = %q, want %q",
+				xff, got, want)
+		}
+	}
+}
+
 func TestSecurityHeaders(t *testing.T) {
 	t.Parallel()

@@ -323,6 +398,172 @@ func TestRecovererRepanicsOnAbortHandler(t *testing.T) {
 	}
 }

+// TestHandlerPanicIsLoggedAndCountedAs500 sends a request to a handler
+// that panics through the middleware SetupRoutes puts around a route
+// with metrics, in its order, leaving out those that write no status
+// for a panic and Sentry, which TestSentry covers. The request log line
+// and the metrics must give the 500 the client gets, and the panic must
+// still reach Recoverer, which logs it.
+func TestHandlerPanicIsLoggedAndCountedAs500(t *testing.T) {
+	t.Parallel()
+
+	var logbuf bytes.Buffer
+
+	mw := middleware.NewWithLogger(
+		slog.New(slog.NewJSONHandler(&logbuf, nil)),
+	)
+	registry := prometheus.NewRegistry()
+
+	handler := chi.Chain(
+		mw.Recoverer(),
+		chimiddleware.RequestID,
+		mw.Logging(),
+		mw.Write500OnPanic(),
+		mw.Metrics(registry),
+		mw.Write500OnPanic(),
+	).HandlerFunc(func(http.ResponseWriter, *http.Request) {
+		panic("boom")
+	})
+
+	rec := httptest.NewRecorder()
+	req := httptest.NewRequestWithContext(t.Context(), http.MethodGet, "/", http.NoBody)
+	handler.ServeHTTP(rec, req)
+
+	if rec.Code != http.StatusInternalServerError {
+		t.Fatalf("status = %d, want %d",
+			rec.Code, http.StatusInternalServerError)
+	}
+
+	request := findLogLine(t, logbuf.Bytes(), "request")
+	if request.Status != http.StatusInternalServerError {
+		t.Errorf("request log status = %d, want %d",
+			request.Status, http.StatusInternalServerError)
+	}
+
+	recovered := findLogLine(t, logbuf.Bytes(), "panic recovered")
+	if recovered.Panic != "boom" {
+		t.Errorf("panic recovered line has panic %q, want %q",
+			recovered.Panic, "boom")
+	}
+
+	for _, name := range []string{
+		"http_request_duration_seconds", "http_response_size_bytes",
+	} {
+		codes := metricCodes(t, registry, name)
+		if len(codes) != 1 || codes[0] != "500" {
+			t.Errorf("%s code labels = %v, want [500]", name, codes)
+		}
+	}
+}
+
+// loggedLine holds the fields the tests read from a JSON log line.
+type loggedLine struct {
+	Msg    string `json:"msg"`
+	Status int    `json:"status"`
+	Panic  string `json:"panic"`
+}
+
+// findLogLine returns the first of the JSON log lines in logged whose
+// message is msg.
+func findLogLine(t *testing.T, logged []byte, msg string) loggedLine {
+	t.Helper()
+
+	dec := json.NewDecoder(bytes.NewReader(logged))
+
+	for {
+		var line loggedLine
+
+		err := dec.Decode(&line)
+		if err != nil {
+			t.Fatalf("no %q line in the log: %v (%q)", msg, err, logged)
+		}
+
+		if line.Msg == msg {
+			return line
+		}
+	}
+}
+
+// metricCodes returns the code label of each series of the metric
+// family name in registry.
+func metricCodes(
+	t *testing.T,
+	registry *prometheus.Registry,
+	name string,
+) []string {
+	t.Helper()
+
+	families, err := registry.Gather()
+	if err != nil {
+		t.Fatalf("gather metrics: %v", err)
+	}
+
+	var codes []string
+
+	for _, family := range families {
+		if family.GetName() != name {
+			continue
+		}
+
+		for _, metric := range family.GetMetric() {
+			for _, label := range metric.GetLabel() {
+				if label.GetName() == "code" {
+					codes = append(codes, label.GetValue())
+				}
+			}
+		}
+	}
+
+	return codes
+}
+
+// TestRecovererWritesNoSecondStatus checks, on a real server, that
+// Recoverer does not write its 500 again after Write500OnPanic has
+// written one, which the server would log as a superfluous
+// WriteHeader call.
+func TestRecovererWritesNoSecondStatus(t *testing.T) {
+	t.Parallel()
+
+	var serverLog bytes.Buffer
+
+	mw := middleware.NewWithLogger(slog.New(slog.DiscardHandler))
+
+	server := httptest.NewUnstartedServer(chi.Chain(
+		mw.Recoverer(),
+		mw.Write500OnPanic(),
+	).HandlerFunc(func(http.ResponseWriter, *http.Request) {
+		panic("boom")
+	}))
+	server.Config.ErrorLog = log.New(&serverLog, "", 0)
+	server.Start()
+
+	req, err := http.NewRequestWithContext(t.Context(),
+		http.MethodGet, server.URL, http.NoBody)
+	if err != nil {
+		t.Fatalf("new request: %v", err)
+	}
+
+	resp, err := server.Client().Do(req)
+	if err != nil {
+		t.Fatalf("request: %v", err)
+	}
+
+	_ = resp.Body.Close()
+
+	// Close waits for the handler to return, so whatever the server
+	// logs for this request is in serverLog after it.
+	server.Close()
+
+	if resp.StatusCode != http.StatusInternalServerError {
+		t.Errorf("status = %d, want %d",
+			resp.StatusCode, http.StatusInternalServerError)
+	}
+
+	if serverLog.Len() != 0 {
+		t.Errorf("server logged %q", serverLog.String())
+	}
+}
+
 // TestLoggingCutsRequestStringsToBound sends an over-long URL and
 // over-long header values, and checks the request log writes each
 // one cut to logger.MaxLoggedFieldBytes.
@@ -18,9 +18,15 @@ var ErrDataDirOutsideVolume = errors.New(
 // PrepareDataDir gets dir, the server's DATA_DIR, ready for owner, the
 // user the server runs as, so that a host directory mounted at volume,
 // /data in the image, needs no preparing: it creates dir, gives volume
-// and everything in it to owner, and gives volume and dir the mode the
-// server gives a directory it creates. dir must be volume or a path
-// below it, with no '.', '..', empty part or '/' at the end.
+// and the directories, regular files and symbolic links in it to
+// owner, and gives volume and dir the mode the server gives a directory
+// it creates. It leaves a file with a second name (hard link) as it
+// is, as that name could be outside volume, and logs its path; it
+// leaves a device, named pipe or socket as it is too (see chownFile).
+// dir must be volume or a path below it, with no '.', '..', empty part
+// or '/' at the end. On any system but Linux, the system of the image, it
+// fails at the first file in volume that is not a directory (see
+// chownFile).
 //
 // bin/entrypoint.sh runs this as root, which would follow a symbolic
 // link anywhere, so every step goes through an os.Root opened on
@@ -32,10 +38,9 @@ var ErrDataDirOutsideVolume = errors.New(
 // open, never following a link at the name it creates, and follows a
 // link on the path only as the os.Root allows, so a relative link
 // inside volume can lead it to create directories elsewhere inside
-// volume. Lchown never changes what a link points to, and the modes
-// are set on directories already opened (see chmodDir), so the most
-// that process can do is make a step fail or wait, or act on
-// something else inside volume.
+// volume. The owners and modes are set on files already opened (see
+// chownAll and chmodDir), so the most that process can do is make a
+// step fail or wait, or act on something else inside volume.
 func PrepareDataDir(volume, dir string, owner *user.User) error {
 	// rel is dir as a path from volume; IsLocal is false for one that
 	// leads out of it.
@@ -66,7 +71,7 @@ func PrepareDataDir(volume, dir string, owner *user.User) error {
 		return err
 	}

-	err = lchownAll(root, ".", uid, gid)
+	err = chownAll(root, ".", uid, gid)
 	if err != nil {
 		return err
 	}
@@ -79,28 +84,23 @@ func PrepareDataDir(volume, dir string, owner *user.User) error {
 	return chmodDir(root, rel)
 }

-// lchownAll gives name, a directory inside root, and everything in it
-// to uid and gid. It reads each directory opened through root, not
-// through root.FS(), which refuses a name that is not valid UTF-8, and
-// calls Lchown on every entry, which gives a symbolic link itself to
-// them, not what it points to. It goes into an entry only when the
-// read found a directory there, so it follows no link it finds; one
-// swapped in for that directory afterwards is followed only as the
-// os.Root allows.
-func lchownAll(root *os.Root, name string, uid, gid int) error {
-	err := root.Lchown(name, uid, gid)
-	if err != nil {
-		return err
-	}
-
-	dir, err := root.Open(name)
-	if err != nil {
-		return err
-	}
-
-	entries, err := dir.ReadDir(-1)
-	_ = dir.Close()
-
+// chownAll gives name, a directory inside root, and the directories,
+// regular files and symbolic links in it to uid and gid, but not the
+// files chownFile leaves. It opens each through root and gives the
+// open file to them, as chmodDir sets a mode on the open directory.
+// Root.Lchown would not hold: it acts on whatever is at the name by
+// then, and a hard link swapped in there shares its owner with a file
+// that could be outside root. A symbolic link is given to them itself,
+// never what it points to, because in a directory that anyone can
+// write to and that has the sticky bit set, as /tmp has, the kernel
+// follows a link only for the link's owner or when the link and the
+// directory have the same owner. chownAll goes into an entry only when
+// the read found a directory there, so it follows no link it finds;
+// one swapped in afterwards for a directory is followed only as the
+// os.Root allows, and one swapped in for any other file is not
+// followed at all (see chownFile).
+func chownAll(root *os.Root, name string, uid, gid int) error {
+	entries, err := chownDir(root, name, uid, gid)
 	if err != nil {
 		return err
 	}
@@ -108,9 +108,9 @@ func lchownAll(root *os.Root, name string, uid, gid int) error {
 	for _, entry := range entries {
 		entryName := filepath.Join(name, entry.Name())
 		if entry.IsDir() {
-			err = lchownAll(root, entryName, uid, gid)
+			err = chownAll(root, entryName, uid, gid)
 		} else {
-			err = root.Lchown(entryName, uid, gid)
+			err = chownFile(root, entryName, uid, gid)
 		}

 		if err != nil {
@@ -121,6 +121,36 @@ func lchownAll(root *os.Root, name string, uid, gid int) error {
 	return nil
 }

+// chownDir gives name, a directory inside root, to uid and gid, and
+// returns what it holds. Like chmodDir, it opens name through root and
+// refuses anything but a directory, which has no second name (hard
+// link). It reads the open directory, not root.FS(), which refuses a
+// name that is not valid UTF-8.
+func chownDir(root *os.Root, name string, uid, gid int) ([]fs.DirEntry, error) {
+	dir, err := root.Open(name)
+	if err != nil {
+		return nil, err
+	}
+
+	defer func() { _ = dir.Close() }()
+
+	info, err := dir.Stat()
+	if err != nil {
+		return nil, err
+	}
+
+	if !info.IsDir() {
+		return nil, &fs.PathError{Op: "chown", Path: name, Err: syscall.ENOTDIR}
+	}
+
+	err = dir.Chown(uid, gid)
+	if err != nil {
+		return nil, err
+	}
+
+	return dir.ReadDir(-1)
+}
+
 // chmodDir gives name, a directory inside root, the mode the server
 // gives a directory it creates. Root.Chmod would not hold: on Linux it
 // checks that name is not a symbolic link, then sets the mode by name,
@@ -0,0 +1,65 @@
+//go:build linux
+
+package reportbuf
+
+import (
+	"io/fs"
+	"log/slog"
+	"os"
+	"syscall"
+
+	"golang.org/x/sys/unix"
+)
+
+// chownFile gives name, a file inside root that is not a directory, to
+// uid and gid if it is a regular file or symbolic link with one name.
+// It leaves, and logs, a file of any kind with a second name (hard
+// link): all the names of a file share its owner, and the other could
+// be outside root. It leaves any other file with one name, such as a
+// device, named pipe or socket: the server needs to own none of them.
+// It opens name through root with O_PATH, which reads nothing from the
+// file, so opening a device or named pipe neither waits nor acts on
+// it; os.Root adds O_NOFOLLOW, and the two together open a symbolic
+// link itself, never what it points to. It reads the kind of file and
+// its number of names from the open file and gives that file to them,
+// so a hard link swapped in at name is caught too.
+//
+// This file is built only on Linux: prepare-data-dir runs only in the
+// Linux image, and macOS, where the backend is also built for
+// development, has neither O_PATH nor AT_EMPTY_PATH. On any other
+// system chownFile returns an error (see datadir_other.go).
+func chownFile(root *os.Root, name string, uid, gid int) error {
+	file, err := root.OpenFile(name, unix.O_PATH, 0)
+	if err != nil {
+		return err
+	}
+
+	defer func() { _ = file.Close() }()
+
+	info, err := file.Stat()
+	if err != nil {
+		return err
+	}
+
+	stat, ok := info.Sys().(*syscall.Stat_t)
+	if !ok || stat.Nlink > 1 {
+		slog.Warn("owner not changed: file has a second name (hard link)",
+			"path", file.Name())
+
+		return nil
+	}
+
+	if !info.Mode().IsRegular() && info.Mode().Type() != fs.ModeSymlink {
+		return nil
+	}
+
+	// File.Chown would fail: fchown refuses a file opened with O_PATH.
+	// Fchownat with an empty name and AT_EMPTY_PATH acts on the open
+	// file instead.
+	err = unix.Fchownat(int(file.Fd()), "", uid, gid, unix.AT_EMPTY_PATH)
+	if err != nil {
+		return &fs.PathError{Op: "chown", Path: file.Name(), Err: err}
+	}
+
+	return nil
+}
@@ -0,0 +1,21 @@
+//go:build !linux
+
+package reportbuf
+
+import (
+	"errors"
+	"io/fs"
+	"os"
+)
+
+// chownFile returns an error on every system but Linux. Opening a file
+// that is not a directory without reading it or following it, and
+// giving it to a user on the open file, never by name, is built only
+// for Linux (see datadir_linux.go): prepare-data-dir runs only in the
+// Linux image, and macOS, where the backend is also built for
+// development, has neither O_PATH nor AT_EMPTY_PATH. This file lets
+// the package build elsewhere, such as on macOS, where a developer
+// runs make build and make run.
+func chownFile(_ *os.Root, name string, _, _ int) error {
+	return &fs.PathError{Op: "chown", Path: name, Err: errors.ErrUnsupported}
+}
@@ -254,10 +254,13 @@ func TestPrepareDataDirTakesDirectoryNamedInLatin1(t *testing.T) {
 	requireOwner(t, filepath.Join(latin1, "f"), uint32(uid), uint32(gid))
 }

-// TestPrepareDataDirGivesVolumeToOwner gives everything in the volume
-// to a uid and gid that own nothing, which only root can do. A
-// symbolic link in the volume to a directory outside it is given to
-// them itself; what it points to is left as it was.
+// TestPrepareDataDirGivesVolumeToOwner gives the directories, regular
+// files and symbolic links in the volume to a uid and gid that own
+// nothing, which only root can do. A symbolic link in the volume to a
+// directory outside it is given to them itself; what it points to is
+// left as it was. A hard link to a file outside the volume, one to a
+// symbolic link outside it and one to a named pipe outside it are left
+// as they were, and so is a named pipe in the volume.
 func TestPrepareDataDirGivesVolumeToOwner(t *testing.T) {
 	t.Parallel()

@@ -274,6 +277,16 @@ func TestPrepareDataDirGivesVolumeToOwner(t *testing.T) {
 		t.Fatal(err)
 	}

+	err = os.Symlink("f", filepath.Join(outside, "link"))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	err = syscall.Mkfifo(filepath.Join(outside, "fifo"), 0o600)
+	if err != nil {
+		t.Fatal(err)
+	}
+
 	err = os.Mkdir(old, 0o700)
 	if err != nil {
 		t.Fatal(err)
@@ -289,6 +302,21 @@ func TestPrepareDataDirGivesVolumeToOwner(t *testing.T) {
 		t.Fatal(err)
 	}

+	err = syscall.Mkfifo(filepath.Join(old, "fifo"), 0o600)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	// os.Link does not follow a symbolic link: it gives the link itself
+	// a second name.
+	for _, name := range []string{"f", "link", "fifo"} {
+		err = os.Link(filepath.Join(outside, name),
+			filepath.Join(old, "hardlink-to-"+name))
+		if err != nil {
+			t.Fatal(err)
+		}
+	}
+
 	err = reportbuf.PrepareDataDir(volume, filepath.Join(volume, reports),
 		&user.User{Uid: "4242", Gid: "4343"})
 	if err != nil {
@@ -296,12 +324,16 @@ func TestPrepareDataDirGivesVolumeToOwner(t *testing.T) {
 	}

 	for _, path := range []string{
-		volume, filepath.Join(volume, reports), old,
-		filepath.Join(old, "f"), filepath.Join(old, "link"),
+		volume, filepath.Join(volume, reports), old, filepath.Join(old, "f"),
+		filepath.Join(old, "link"),
 	} {
 		requireOwner(t, path, 4242, 4343)
 	}

-	requireOwner(t, outside, 0, 0)
-	requireOwner(t, filepath.Join(outside, "f"), 0, 0)
+	for _, path := range []string{
+		outside, filepath.Join(outside, "f"), filepath.Join(outside, "link"),
+		filepath.Join(outside, "fifo"), filepath.Join(old, "fifo"),
+	} {
+		requireOwner(t, path, 0, 0)
+	}
 }
@@ -1,6 +1,9 @@
 // Package reportbuf accumulates telemetry reports in memory
 // and periodically flushes them to zstd-compressed JSONL files,
-// deleting the oldest files to keep them under a size cap.
+// deleting the oldest files to keep them under a size cap. Before the
+// server starts, PrepareDataDir gets the directory those files are
+// kept in, DATA_DIR, ready for the user the server runs as: it creates
+// it and gives it and the volume it is on to that user.
 package reportbuf

 import (
@@ -13,7 +16,6 @@ import (
 	"log/slog"
 	"os"
 	"path/filepath"
-	"slices"
 	"strings"
 	"sync"
 	"sync/atomic"
@@ -70,11 +72,9 @@ type Buffer struct {
 	// except in tests that hold the write open or make it fail.
 	fileCreated func(f *os.File)
 	// files are the report files that may be deleted to make room,
-	// in name order, which is oldest first: those in dataDir at
-	// start, and each one this buffer writes, put in at its place by
-	// name once it is complete, even when an older file's write
-	// completes after a newer one's. A file still being written is
-	// not among them. filesBytes is their total size.
+	// oldest first: those in dataDir at start, in name order, then
+	// each one this buffer writes, once it is complete. A file still
+	// being written is not among them. filesBytes is their total size.
 	files      []reportFile
 	filesBytes int64
 	log        *slog.Logger
@@ -85,12 +85,30 @@ type Buffer struct {
 	now func() time.Time
 	// seq numbers the report files, so that two named in the same
 	// millisecond still get different names.
-	seq      atomic.Uint64
+	seq      atomic.Int64
 	stopOnce sync.Once
+	// thresholdWriteWaiting is set while a write Append started at
+	// flushSizeThreshold waits for writeMu; Append starts no other
+	// while it is set.
+	thresholdWriteWaiting atomic.Bool
 	// usedBytes is what Append checks against maxBytes: the size
 	// of the report files in dataDir, plus the reports waiting to
 	// be written to one at their uncompressed size.
 	usedBytes int64
+	// writeErr is the error of the last write of a report file to
+	// finish, nil if it succeeded or none has. While it is set, Append
+	// refuses reports, and the reports of that write wait in buf for
+	// the next flush.
+	writeErr error
+	// writeMu is held across each write of a report file, so writes
+	// never overlap: a failed write puts its reports back in buf before
+	// the next write takes them, which keeps them in the order they
+	// were appended.
+	writeMu sync.Mutex
+	// writers are the goroutines that write report files: flushLoop,
+	// and each one Append starts at flushSizeThreshold. OnStop waits
+	// for them before the final flush.
+	writers sync.WaitGroup
 }

 // New creates a Buffer and registers lifecycle hooks to
@@ -115,16 +133,33 @@ func New(

 	lc.Append(fx.Hook{
 		OnStart: func(_ context.Context) error {
+			// The errors name DATA_DIR, as config.New's do for the
+			// settings it checks.
 			err := os.MkdirAll(b.dataDir, dirPerms)
 			if err != nil {
-				return fmt.Errorf("create data dir: %w", err)
+				return fmt.Errorf("DATA_DIR %q: %w", b.dataDir, err)
+			}
+
+			// Creating and removing a file stops the start when no
+			// file can be created in DATA_DIR, instead of the first
+			// write of a report file failing after reports were
+			// answered 200. A full disk is not caught: the empty
+			// file is still created.
+			f, err := os.CreateTemp(b.dataDir, "")
+			if err != nil {
+				return fmt.Errorf("DATA_DIR %q: %w", b.dataDir, err)
+			}
+
+			err = errors.Join(f.Close(), os.Remove(f.Name()))
+			if err != nil {
+				return fmt.Errorf("DATA_DIR %q: %w", b.dataDir, err)
 			}

 			// Report files left by earlier runs count too, and are
 			// the first to be deleted to make room.
 			files, err := reportFiles(b.dataDir)
 			if err != nil {
-				return err
+				return fmt.Errorf("DATA_DIR %q: %w", b.dataDir, err)
 			}

 			b.mu.Lock()
@@ -142,7 +177,7 @@ func New(

 			b.mu.Unlock()

-			go b.flushLoop()
+			b.writers.Go(b.flushLoop)

 			return nil
 		},
@@ -153,7 +188,14 @@ func New(
 			var err error

 			b.stopOnce.Do(func() {
+				// Closing done ends flushLoop after any flush it is
+				// running. Wait for it and for the writes Append
+				// started, which the process exiting would cut off;
+				// one that fails puts its reports back in buf, for the
+				// final flush to write.
 				close(b.done)
+				b.writers.Wait()
+
 				err = b.flushLocked()
 			})

@@ -167,11 +209,13 @@ func New(
 }

 // Append marshals v as a single JSON line and appends it to
-// the buffer. If the line would take usedBytes past maxBytes, the
-// oldest report files are deleted to make room; it stores nothing
-// and returns ErrFull if that cannot make room. If the buffer
-// reaches the size threshold, it is drained and written to disk
-// asynchronously.
+// the buffer. It stores nothing and returns an error if the last
+// write of a report file failed, until a write succeeds again. If the
+// line would take usedBytes past maxBytes, the oldest report files are
+// deleted to make room; it stores nothing and returns ErrFull if that
+// cannot make room. If the buffer reaches the size threshold, it starts
+// a write of the reports in the background (see flushAtThreshold),
+// unless one it started is still waiting.
 func (b *Buffer) Append(v any) error {
 	line, err := json.Marshal(v)
 	if err != nil {
@@ -182,6 +226,13 @@ func (b *Buffer) Append(v any) error {

 	b.mu.Lock()

+	if b.writeErr != nil {
+		err = fmt.Errorf("report files cannot be written: %w", b.writeErr)
+		b.mu.Unlock()
+
+		return err
+	}
+
 	b.deleteOldestFiles(lineBytes)

 	if b.usedBytes+lineBytes > b.maxBytes {
@@ -194,18 +245,9 @@ func (b *Buffer) Append(v any) error {
 	b.buf.Write(line)
 	b.buf.WriteByte('\n')

-	if b.buf.Len() >= flushSizeThreshold {
-		data := b.drainBuf()
-		b.mu.Unlock()
-
-		go func() {
-			writeErr := b.writeFile(data)
-			if writeErr != nil {
-				b.log.Error("flush reports failed", "error", writeErr)
-			}
-		}()
-
-		return nil
+	if b.buf.Len() >= flushSizeThreshold &&
+		b.thresholdWriteWaiting.CompareAndSwap(false, true) {
+		b.writers.Go(b.flushAtThreshold)
 	}

 	b.mu.Unlock()
@@ -263,21 +305,28 @@ func (b *Buffer) flushLoop() {
 	}
 }

-// flushLocked acquires the lock, drains the buffer, and
-// writes the data to a compressed file.
+// flushLocked waits for any write of a report file under way, then
+// writes the reports waiting in the buffer to a new one.
 func (b *Buffer) flushLocked() error {
-	b.mu.Lock()
+	b.writeMu.Lock()
+	defer b.writeMu.Unlock()

-	if b.buf.Len() == 0 {
-		b.mu.Unlock()
+	return b.writeFile()
+}

-		return nil
+// flushAtThreshold is the write Append starts when the buffer reaches
+// flushSizeThreshold. Once any write under way is done, it lets Append
+// start another and writes the reports waiting in the buffer.
+func (b *Buffer) flushAtThreshold() {
+	b.writeMu.Lock()
+	defer b.writeMu.Unlock()
+
+	b.thresholdWriteWaiting.Store(false)
+
+	err := b.writeFile()
+	if err != nil {
+		b.log.Error("flush reports failed", "error", err)
 	}
-
-	data := b.drainBuf()
-	b.mu.Unlock()
-
-	return b.writeFile(data)
 }

 // drainBuf copies the buffer contents and resets it.
@@ -290,9 +339,22 @@ func (b *Buffer) drainBuf() []byte {
 	return data
 }

-// writeFile writes data, reports drained from the buffer, to a new
-// timestamped zstd-compressed JSONL file in the data directory.
-func (b *Buffer) writeFile(data []byte) error {
+// writeFile takes the reports waiting in the buffer, if any, and writes
+// them to a new timestamped zstd-compressed JSONL file in the data
+// directory. If the write fails, it puts the reports back in the
+// buffer. The caller must hold b.writeMu.
+func (b *Buffer) writeFile() error {
+	b.mu.Lock()
+
+	if b.buf.Len() == 0 {
+		b.mu.Unlock()
+
+		return nil
+	}
+
+	data := b.drainBuf()
+	b.mu.Unlock()
+
 	// The timestamp comes first, so the names sort by time; the number
 	// after it tells apart files named in the same millisecond.
 	ts := b.now().UTC().Format("2006-01-02T15-04-05.000Z")
@@ -300,29 +362,30 @@ func (b *Buffer) writeFile(data []byte) error {

 	size, err := b.createFile(filepath.Join(b.dataDir, name), data)

-	// The reports no longer wait to be written, so they stop counting
-	// at their uncompressed size. If the write failed they are lost;
-	// otherwise they count as the file, which from here on may be
-	// deleted to make room.
 	b.mu.Lock()
 	defer b.mu.Unlock()

-	b.usedBytes -= int64(len(data))
+	b.writeErr = err

 	if err != nil {
+		// The reports go back in front of any appended since, for the
+		// next flush to write, and still count at their uncompressed
+		// size.
+		appended := bytes.Clone(b.buf.Bytes())
+		b.buf.Reset()
+		b.buf.Write(data)
+		b.buf.Write(appended)
+
 		return err
 	}

+	// The reports no longer wait to be written, so they stop counting
+	// at their uncompressed size and count as the file, which from here
+	// on may be deleted to make room.
+	b.usedBytes -= int64(len(data))
 	b.usedBytes += size
 	b.filesBytes += size
-
-	// At its place by name, not at the end: another write, of a newer
-	// file, may have completed while this one was being written.
-	i, _ := slices.BinarySearchFunc(b.files, name,
-		func(f reportFile, target string) int {
-			return strings.Compare(f.name, target)
-		})
-	b.files = slices.Insert(b.files, i, reportFile{name: name, size: size})
+	b.files = append(b.files, reportFile{name: name, size: size})

 	return nil
 }
@@ -5,6 +5,8 @@ import (
 	"errors"
 	"fmt"
 	"io/fs"
+	"net/http"
+	"net/http/httptest"
 	"os"
 	"path/filepath"
 	"slices"
@@ -13,10 +15,13 @@ import (
 	"sync"
 	"sync/atomic"
 	"testing"
+	"testing/synctest"
 	"time"

 	"sneak.berlin/go/netwatch/internal/config"
 	"sneak.berlin/go/netwatch/internal/globals"
+	"sneak.berlin/go/netwatch/internal/handlers"
+	"sneak.berlin/go/netwatch/internal/healthcheck"
 	"sneak.berlin/go/netwatch/internal/logger"
 	"sneak.berlin/go/netwatch/internal/reportbuf"

@@ -103,9 +108,236 @@ func TestFailedFinalFlushFailsStop(t *testing.T) {
 	}
 }

-// startBuffer starts a Buffer through fx, as main does, with the
-// DATA_DIR and DATA_DIR_MAX_BYTES the calling test has set.
-func startBuffer(t *testing.T) *reportbuf.Buffer {
+// TestUncreatableDataDirStopsStart sets DATA_DIR below a regular file,
+// where no directory can be created, even by root, as the tests run in
+// the backend image. The start must fail with an error naming DATA_DIR
+// and its value, as the errors for the other settings do.
+func TestUncreatableDataDirStopsStart(t *testing.T) {
+	file := filepath.Join(t.TempDir(), "file")
+	writeBytes(t, file, 1)
+
+	dir := filepath.Join(file, "reports")
+	t.Setenv("DATA_DIR", dir)
+
+	app := fxtest.New(t,
+		fx.Provide(
+			globals.New,
+			logger.New,
+			config.New,
+			reportbuf.New,
+		),
+		fx.Invoke(func(*reportbuf.Buffer) {}),
+	)
+
+	err := app.Start(t.Context())
+	if err == nil {
+		t.Fatal("start succeeded with DATA_DIR below a regular file")
+	}
+
+	want := fmt.Sprintf("DATA_DIR %q: ", dir)
+	if !strings.Contains(err.Error(), want) {
+		t.Fatalf("start error = %q, want it to contain %q", err, want)
+	}
+}
+
+// TestUnwritableDataDirStopsStart sets DATA_DIR to /proc, which exists
+// and can be read, but where no file can be created, even by root, as
+// the tests run in the backend image: root ignores a directory the test
+// makes read-only. The start must fail with an error naming DATA_DIR and
+// its value, rather than the first write of a report file.
+func TestUnwritableDataDirStopsStart(t *testing.T) {
+	dir := "/proc"
+	t.Setenv("DATA_DIR", dir)
+
+	app := fxtest.New(t,
+		fx.Provide(
+			globals.New,
+			logger.New,
+			config.New,
+			reportbuf.New,
+		),
+		fx.Invoke(func(*reportbuf.Buffer) {}),
+	)
+
+	err := app.Start(t.Context())
+	if err == nil {
+		t.Fatal("start succeeded with DATA_DIR /proc")
+	}
+
+	want := fmt.Sprintf("DATA_DIR %q: ", dir)
+	if !strings.Contains(err.Error(), want) {
+		t.Fatalf("start error = %q, want it to contain %q", err, want)
+	}
+}
+
+// TestStopWaitsForPeriodicWrite holds open the write the periodic flush
+// starts, then stops the app. The stop must wait until the write is
+// released, and the report must then be in its file. The test runs in
+// a synctest bubble, whose clock moves on at once when every goroutine
+// in it is waiting, so the periodic flush does not take a minute.
+func TestStopWaitsForPeriodicWrite(t *testing.T) {
+	dir := t.TempDir()
+	t.Setenv("DATA_DIR", dir)
+
+	synctest.Test(t, func(t *testing.T) {
+		buf, app := startApp(t)
+
+		created := make(chan struct{})
+		release := make(chan struct{})
+
+		buf.OnFileCreated(func(*os.File) {
+			created <- struct{}{}
+
+			<-release
+		})
+
+		err := buf.Append(map[string]string{"id": "periodic"})
+		if err != nil {
+			t.Fatalf("append report: %v", err)
+		}
+
+		<-created
+
+		err = stopWhileHeld(t, app, release)
+		if err != nil {
+			t.Fatalf("stop: %v", err)
+		}
+
+		files, err := readReportFiles(dir)
+		if err != nil {
+			t.Fatalf("read report files: %v", err)
+		}
+
+		want := `{"id":"periodic"}` + "\n"
+		if len(files) != 1 || files[0] != want {
+			t.Fatalf("report files = %q, want one holding %q", files, want)
+		}
+	})
+}
+
+// TestStopWaitsForWriteAtSizeThreshold holds open the write Append
+// starts when the buffer reaches FlushSizeThreshold, then stops the
+// app. The stop must wait until the write is released, and the reports
+// must then be in its file.
+func TestStopWaitsForWriteAtSizeThreshold(t *testing.T) {
+	dir := t.TempDir()
+	t.Setenv("DATA_DIR", dir)
+
+	synctest.Test(t, func(t *testing.T) {
+		buf, app := startApp(t)
+
+		created := make(chan struct{})
+		release := make(chan struct{})
+
+		buf.OnFileCreated(func(*os.File) {
+			created <- struct{}{}
+
+			<-release
+		})
+
+		appended := appendToThreshold(t, buf)
+
+		<-created
+
+		err := stopWhileHeld(t, app, release)
+		if err != nil {
+			t.Fatalf("stop: %v", err)
+		}
+
+		files, err := readReportFiles(dir)
+		if err != nil {
+			t.Fatalf("read report files: %v", err)
+		}
+
+		if len(files) != 1 || files[0] != appended {
+			t.Fatalf("%d report files, want one holding the %d bytes "+
+				"appended", len(files), len(appended))
+		}
+	})
+}
+
+// TestFinalFlushWritesReportsOfFailedWrite holds open the write the
+// periodic flush starts, makes it fail, then stops the app. The stop
+// waits for the write, which puts its report back in the buffer, and
+// the final flush writes it, so the stop succeeds with nothing lost.
+func TestFinalFlushWritesReportsOfFailedWrite(t *testing.T) {
+	dir := t.TempDir()
+	t.Setenv("DATA_DIR", dir)
+
+	synctest.Test(t, func(t *testing.T) {
+		buf, app := startApp(t)
+
+		created := make(chan *os.File)
+		release := make(chan struct{})
+
+		buf.OnFileCreated(func(f *os.File) {
+			created <- f
+
+			<-release
+		})
+
+		err := buf.Append(map[string]string{"id": "failed"})
+		if err != nil {
+			t.Fatalf("append report: %v", err)
+		}
+
+		held := <-created
+
+		// Closing the file under the write makes the write fail.
+		_ = held.Close()
+
+		// The final flush's write goes through.
+		buf.OnFileCreated(func(*os.File) {})
+
+		err = stopWhileHeld(t, app, release)
+		if err != nil {
+			t.Fatalf("stop: %v", err)
+		}
+
+		files, err := readReportFiles(dir)
+		if err != nil {
+			t.Fatalf("read report files: %v", err)
+		}
+
+		want := `{"id":"failed"}` + "\n"
+		if len(files) != 1 || files[0] != want {
+			t.Fatalf("report files = %q, want one holding %q", files, want)
+		}
+	})
+}
+
+// stopWhileHeld stops app while a report file write is held open until
+// release is closed, and fails the test if the stop returns before
+// that. It then closes release and returns the stop's error. It must
+// be called in a synctest bubble.
+func stopWhileHeld(
+	t *testing.T,
+	app *fxtest.App,
+	release chan<- struct{},
+) error {
+	t.Helper()
+
+	stopped := make(chan error, 1)
+
+	go func() { stopped <- app.Stop(t.Context()) }()
+
+	// Wait returns once every other goroutine in the bubble is waiting:
+	// the stop has got as far as it can while the write is held.
+	synctest.Wait()
+
+	if len(stopped) != 0 {
+		t.Error("stop returned while a report file write was held open")
+	}
+
+	close(release)
+
+	return <-stopped
+}
+
+// startApp starts a Buffer through fx, as main does, with the DATA_DIR
+// and DATA_DIR_MAX_BYTES the calling test has set, and returns it with
+// the app, for tests that stop the app themselves.
+func startApp(t *testing.T) (*reportbuf.Buffer, *fxtest.App) {
 	t.Helper()

 	var buf *reportbuf.Buffer
@@ -123,6 +355,16 @@ func startBuffer(t *testing.T) *reportbuf.Buffer {
 	app.RequireStart()
 	t.Cleanup(app.RequireStop)

+	return buf, app
+}
+
+// startBuffer starts a Buffer through fx, as main does, with the
+// DATA_DIR and DATA_DIR_MAX_BYTES the calling test has set.
+func startBuffer(t *testing.T) *reportbuf.Buffer {
+	t.Helper()
+
+	buf, _ := startApp(t)
+
 	return buf
 }

@@ -339,6 +581,62 @@ func TestWrittenFilesDeletedToMakeRoom(t *testing.T) {
 	}
 }

+// TestWrittenFilesDeletedOldestFirst writes three report files, then
+// appends a report that fits only once one of them is deleted. The
+// first one written must be the one deleted.
+func TestWrittenFilesDeletedOldestFirst(t *testing.T) {
+	const maxBytes = 1000
+
+	report := map[string]string{"id": "written"}
+
+	t.Setenv("DATA_DIR", t.TempDir())
+	t.Setenv("DATA_DIR_MAX_BYTES", strconv.Itoa(maxBytes))
+
+	buf := startBuffer(t)
+
+	var written []string
+
+	buf.OnFileCreated(func(f *os.File) { written = append(written, f.Name()) })
+
+	for range 3 {
+		err := buf.Append(report)
+		if err != nil {
+			t.Fatalf("append report: %v", err)
+		}
+
+		err = buf.Flush()
+		if err != nil {
+			t.Fatalf("flush: %v", err)
+		}
+	}
+
+	// The three files hold the same report, so they are the same size.
+	info, err := os.Stat(written[0])
+	if err != nil {
+		t.Fatalf("stat %s: %v", written[0], err)
+	}
+
+	// A report that fits beside two of the files, so deleting one makes
+	// exactly the room it needs.
+	pad := maxBytes - 2*int(info.Size()) -
+		lineBytes(t, map[string]string{"id": ""})
+
+	err = buf.Append(map[string]string{"id": strings.Repeat("a", pad)})
+	if err != nil {
+		t.Fatalf("report that needs room: %v", err)
+	}
+
+	if exists(t, written[0]) {
+		t.Fatal("first report file written kept when room was needed")
+	}
+
+	for _, path := range written[1:] {
+		if !exists(t, path) {
+			t.Fatalf("%s deleted before the first report file written", path)
+		}
+	}
+}
+
 // TestFailedDeletionStillCounts makes deleting the oldest report file
 // fail. It is still there, so it still takes room, and the next oldest
 // is deleted in its place.
@@ -424,18 +722,23 @@ func TestFileDeletedByHandFreesRoom(t *testing.T) {
 	}
 }

-// TestFileBeingWrittenIsNeverDeleted holds the write of one report file
-// open while a second write completes, then sends a report that needs
-// room. Deleting either file would make it, and the one being written is
-// the older, but only the complete one may be deleted. Once the first
-// write is complete, its file is deleted when room is needed.
+// TestFileBeingWrittenIsNeverDeleted holds the write of a report file
+// open, then sends a report that needs room. The report file from an
+// earlier run is deleted to make it, and the one being written is not.
+// Once the write is complete, its file is deleted when room is needed.
 func TestFileBeingWrittenIsNeverDeleted(t *testing.T) {
-	report := map[string]string{"id": "writing"}
+	const fileBytes = 100

-	t.Setenv("DATA_DIR", t.TempDir())
-	// Room for two reports waiting to be written, but not for two
-	// beside a report file.
-	t.Setenv("DATA_DIR_MAX_BYTES", strconv.Itoa(2*lineBytes(t, report)))
+	report := map[string]string{"id": "writing"}
+	dir := t.TempDir()
+	earlier := reportFilePath(dir, 1)
+
+	writeBytes(t, earlier, fileBytes)
+
+	t.Setenv("DATA_DIR", dir)
+	// Room for the earlier file and one report.
+	t.Setenv("DATA_DIR_MAX_BYTES",
+		strconv.Itoa(fileBytes+lineBytes(t, report)))

 	buf := startBuffer(t)

@@ -459,24 +762,8 @@ func TestFileBeingWrittenIsNeverDeleted(t *testing.T) {

 	writing := <-created

-	// Only the first write is held; the second goes through, and
-	// so do the writes after it, the final one at stop included.
-	var complete string
-
-	buf.OnFileCreated(func(f *os.File) { complete = f.Name() })
-
-	// Errorf, not Fatalf, until the first write is released, so that a
+	// Errorf, not Fatalf, until the write is released, so that a
 	// failure here does not leave it held.
-	err = buf.Append(report)
-	if err != nil {
-		t.Errorf("second report: %v", err)
-	}
-
-	err = buf.Flush()
-	if err != nil {
-		t.Errorf("flush of the second report: %v", err)
-	}
-
 	err = buf.Append(report)
 	if err != nil {
 		t.Errorf("report that needs room: %v", err)
@@ -486,10 +773,14 @@ func TestFileBeingWrittenIsNeverDeleted(t *testing.T) {
 		t.Error("report file deleted while it was being written")
 	}

-	if exists(t, complete) {
-		t.Error("complete report file kept, though room was needed")
+	if exists(t, earlier) {
+		t.Error("earlier report file kept, though room was needed")
 	}

+	// Only this write is held; the writes after it, the final one at
+	// stop included, go through.
+	buf.OnFileCreated(func(*os.File) {})
+
 	close(release)

 	err = <-flushed
@@ -497,9 +788,13 @@ func TestFileBeingWrittenIsNeverDeleted(t *testing.T) {
 		t.Fatalf("flush of the first report: %v", err)
 	}

-	err = buf.Append(report)
+	// As large as the earlier file, so it fits only once the file just
+	// written is deleted too.
+	pad := fileBytes - lineBytes(t, map[string]string{"id": ""})
+
+	err = buf.Append(map[string]string{"id": strings.Repeat("a", pad)})
 	if err != nil {
-		t.Fatalf("report after the first write was complete: %v", err)
+		t.Fatalf("report after the write was complete: %v", err)
 	}

 	if exists(t, writing) {
@@ -507,13 +802,17 @@ func TestFileBeingWrittenIsNeverDeleted(t *testing.T) {
 	}
 }

-// TestFailedWriteStopsCounting makes a write fail once its file is
-// created. Its reports are lost, so they stop counting, and the part of
-// the file written is removed, so it takes no room.
-func TestFailedWriteStopsCounting(t *testing.T) {
-	report := map[string]string{"id": "failed"}
+// TestFailedWriteKeepsItsReports makes a write fail once its file is
+// created, under a cap with room for one report. The part of the file
+// written is removed, so it takes no room, and the report waits in
+// memory for the next flush, which writes it. Once written, it counts
+// as its file and no longer as a report waiting, so the cap still holds
+// exactly: one more report is taken and the one after it is refused.
+func TestFailedWriteKeepsItsReports(t *testing.T) {
+	report := map[string]string{"id": "kept"}
+	dir := t.TempDir()

-	t.Setenv("DATA_DIR", t.TempDir())
+	t.Setenv("DATA_DIR", dir)
 	t.Setenv("DATA_DIR_MAX_BYTES", strconv.Itoa(lineBytes(t, report)))

 	buf := startBuffer(t)
@@ -544,10 +843,289 @@ func TestFailedWriteStopsCounting(t *testing.T) {
 	// Writes from here on, the final one at stop included, succeed.
 	buf.OnFileCreated(func(*os.File) {})

+	err = buf.Flush()
+	if err != nil {
+		t.Fatalf("flush after the failed write: %v", err)
+	}
+
+	files, err := readReportFiles(dir)
+	if err != nil {
+		t.Fatalf("read report files: %v", err)
+	}
+
+	want := `{"id":"kept"}` + "\n"
+	if len(files) != 1 || files[0] != want {
+		t.Fatalf("report files = %q, want one holding %q", files, want)
+	}
+
+	// The file is deleted to make room if need be.
 	err = buf.Append(report)
 	if err != nil {
-		t.Fatalf("report after the failed write: %v", err)
+		t.Fatalf("report after the written one: %v", err)
 	}
+
+	err = buf.Append(report)
+	if !errors.Is(err, reportbuf.ErrFull) {
+		t.Fatalf("report past the cap: error = %v, want ErrFull", err)
+	}
+}
+
+// TestFailedWriteKeepsReportsAppendedDuringIt holds a write open,
+// appends a report while it is held, then makes the held write fail.
+// Both reports are kept, and the next flush writes them in the order
+// they were appended, the held write's first.
+func TestFailedWriteKeepsReportsAppendedDuringIt(t *testing.T) {
+	dir := t.TempDir()
+	t.Setenv("DATA_DIR", dir)
+
+	buf := startBuffer(t)
+
+	created := make(chan *os.File)
+	release := make(chan struct{})
+
+	buf.OnFileCreated(func(f *os.File) {
+		created <- f
+
+		<-release
+	})
+
+	err := buf.Append(map[string]string{"id": "held"})
+	if err != nil {
+		t.Fatalf("report of the held write: %v", err)
+	}
+
+	flushed := make(chan error)
+
+	go func() { flushed <- buf.Flush() }()
+
+	held := <-created
+
+	// Errorf, not Fatalf, until the held write is released, so that a
+	// failure here does not leave it held.
+	err = buf.Append(map[string]string{"id": "appended"})
+	if err != nil {
+		t.Errorf("report appended while the write was held: %v", err)
+	}
+
+	// Closing the file under the write makes the write fail.
+	_ = held.Close()
+
+	close(release)
+
+	err = <-flushed
+	if err == nil {
+		t.Fatal("held write succeeded, though its file was closed under it")
+	}
+
+	// Writes from here on, the final one at stop included, succeed.
+	buf.OnFileCreated(func(*os.File) {})
+
+	err = buf.Flush()
+	if err != nil {
+		t.Fatalf("flush after the failed write: %v", err)
+	}
+
+	files, err := readReportFiles(dir)
+	if err != nil {
+		t.Fatalf("read report files: %v", err)
+	}
+
+	want := `{"id":"held"}` + "\n" + `{"id":"appended"}` + "\n"
+	if len(files) != 1 || files[0] != want {
+		t.Fatalf("report files = %q, want one holding %q", files, want)
+	}
+}
+
+// TestFailedWritesKeepReportsInOrder holds a write open, then appends a
+// report that takes the buffer to FlushSizeThreshold, so that Append
+// starts a second write. The held write fails first, then the second,
+// which waited for it and took both reports. The next flush writes them
+// in the order they were appended, the held write's first.
+func TestFailedWritesKeepReportsInOrder(t *testing.T) {
+	dir := t.TempDir()
+	t.Setenv("DATA_DIR", dir)
+
+	buf := startBuffer(t)
+
+	created := make(chan struct{})
+	release := make(chan struct{})
+
+	// Each write is held until release is closed, then fails, as its
+	// file is closed under it.
+	buf.OnFileCreated(func(f *os.File) {
+		created <- struct{}{}
+
+		<-release
+
+		_ = f.Close()
+	})
+
+	err := buf.Append(map[string]string{"id": "held"})
+	if err != nil {
+		t.Fatalf("report of the held write: %v", err)
+	}
+
+	flushed := make(chan error)
+
+	go func() { flushed <- buf.Flush() }()
+
+	<-created
+
+	big := strings.Repeat("a", reportbuf.FlushSizeThreshold)
+
+	// Errorf, not Fatalf, until the held write is released, so that a
+	// failure here does not leave it held.
+	err = buf.Append(map[string]string{"id": big})
+	if err != nil {
+		t.Errorf("report that starts the second write: %v", err)
+	}
+
+	close(release)
+
+	err = <-flushed
+	if err == nil {
+		t.Error("held write succeeded, though its file was closed under it")
+	}
+
+	// The second write has created its file, so it has taken the
+	// reports.
+	<-created
+
+	// Writes from here on, the final one at stop included, succeed.
+	buf.OnFileCreated(func(*os.File) {})
+
+	// Flush waits for the second write to fail before it writes.
+	err = buf.Flush()
+	if err != nil {
+		t.Fatalf("flush after the failed writes: %v", err)
+	}
+
+	files, err := readReportFiles(dir)
+	if err != nil {
+		t.Fatalf("read report files: %v", err)
+	}
+
+	want := `{"id":"held"}` + "\n" + `{"id":"` + big + `"}` + "\n"
+	if len(files) != 1 || files[0] != want {
+		t.Fatalf("%d report files, want one holding the held write's "+
+			"report, then the second one", len(files))
+	}
+}
+
+// TestReportsRefusedWhileWritesFail posts reports through the report
+// handler, as the page does. Once a write has failed, a report is
+// refused with 500 until a write succeeds again; the reports taken
+// before and after are stored, and the one refused is not.
+func TestReportsRefusedWhileWritesFail(t *testing.T) {
+	dir := t.TempDir()
+	t.Setenv("DATA_DIR", dir)
+
+	buf, h := startBufferWithHandlers(t)
+
+	if code := postReport(t, h, "before"); code != http.StatusOK {
+		t.Fatalf("report before the failed write: status = %d, want %d",
+			code, http.StatusOK)
+	}
+
+	// With the data directory gone, the write has nowhere to go.
+	err := os.Remove(dir)
+	if err != nil {
+		t.Fatalf("remove data dir: %v", err)
+	}
+
+	err = buf.Flush()
+	if err == nil {
+		t.Fatal("flush succeeded with the data directory gone")
+	}
+
+	if code := postReport(t, h, "refused"); code != http.StatusInternalServerError {
+		t.Fatalf("report after the failed write: status = %d, want %d",
+			code, http.StatusInternalServerError)
+	}
+
+	err = os.Mkdir(dir, 0o750)
+	if err != nil {
+		t.Fatalf("make data dir: %v", err)
+	}
+
+	err = buf.Flush()
+	if err != nil {
+		t.Fatalf("flush with the data directory back: %v", err)
+	}
+
+	if code := postReport(t, h, "after"); code != http.StatusOK {
+		t.Fatalf("report after a write succeeded: status = %d, want %d",
+			code, http.StatusOK)
+	}
+
+	err = buf.Flush()
+	if err != nil {
+		t.Fatalf("flush of the report after: %v", err)
+	}
+
+	files, err := readReportFiles(dir)
+	if err != nil {
+		t.Fatalf("read report files: %v", err)
+	}
+
+	stored := strings.Join(files, "")
+
+	for _, clientID := range []string{"before", "after"} {
+		if !strings.Contains(stored, `"clientId":"`+clientID+`"`) {
+			t.Errorf("report %q not stored", clientID)
+		}
+	}
+
+	if strings.Contains(stored, `"clientId":"refused"`) {
+		t.Error("refused report stored")
+	}
+}
+
+// startBufferWithHandlers starts a Buffer and the Handlers that store
+// reports in it through fx, as main does, with the DATA_DIR the calling
+// test has set.
+func startBufferWithHandlers(
+	t *testing.T,
+) (*reportbuf.Buffer, *handlers.Handlers) {
+	t.Helper()
+
+	var (
+		buf *reportbuf.Buffer
+		h   *handlers.Handlers
+	)
+
+	app := fxtest.New(t,
+		fx.Provide(
+			config.New,
+			globals.New,
+			handlers.New,
+			healthcheck.New,
+			logger.New,
+			reportbuf.New,
+		),
+		fx.Populate(&buf, &h),
+	)
+
+	app.RequireStart()
+	t.Cleanup(app.RequireStop)
+
+	return buf, h
+}
+
+// postReport posts a report from clientID to the report handler, as the
+// page does, and returns the status of the answer.
+func postReport(t *testing.T, h *handlers.Handlers, clientID string) int {
+	t.Helper()
+
+	rec := httptest.NewRecorder()
+	req := httptest.NewRequestWithContext(t.Context(),
+		http.MethodPost, "/api/v1/reports",
+		strings.NewReader(`{"clientId":"`+clientID+`","hosts":[]}`),
+	)
+	req.Header.Set("Content-Type", "application/json")
+	h.HandleReport().ServeHTTP(rec, req)
+
+	return rec.Code
 }

 // TestConcurrentAppendsStopAtCap appends from many goroutines at once
@@ -681,6 +1259,132 @@ func TestFlushAtSizeThreshold(t *testing.T) {

 	buf := startBuffer(t)

+	appended := appendToThreshold(t, buf)
+
+	// Append writes the file in the background, so wait for it.
+	deadline := time.Now().Add(10 * time.Second)
+
+	for {
+		files, err := readReportFiles(dir)
+		if err == nil && len(files) == 1 && files[0] == appended {
+			return
+		}
+
+		if time.Now().After(deadline) {
+			t.Fatalf("%d report files (error: %v), want one holding the "+
+				"%d bytes appended", len(files), err, len(appended))
+		}
+
+		time.Sleep(10 * time.Millisecond)
+	}
+}
+
+// TestEachSizeThresholdStartsAWrite reaches FlushSizeThreshold twice,
+// with no call to Flush and the periodic flush a minute away. Append
+// must start a write each time, so that the reports end up in two
+// report files. The test runs in a synctest bubble, so that
+// synctest.Wait returns once each write is done.
+func TestEachSizeThresholdStartsAWrite(t *testing.T) {
+	dir := t.TempDir()
+	t.Setenv("DATA_DIR", dir)
+
+	synctest.Test(t, func(t *testing.T) {
+		buf := startBuffer(t)
+
+		var appended string
+
+		for range 2 {
+			appended = appendToThreshold(t, buf)
+
+			synctest.Wait()
+		}
+
+		files, err := readReportFiles(dir)
+		if err != nil {
+			t.Fatalf("read report files: %v", err)
+		}
+
+		if !slices.Equal(files, []string{appended, appended}) {
+			t.Fatalf("%d report files, want two, each holding the %d "+
+				"bytes appended to reach the threshold",
+				len(files), len(appended))
+		}
+	})
+}
+
+// TestAppendStartsOneWaitingWrite holds a write open, then appends two
+// reports of FlushSizeThreshold bytes each. Append starts a write at
+// the first, which waits for the held one, and none at the second,
+// since that write is still waiting. Every write fails, so a write
+// started at the second report would find the reports back in the
+// buffer and try them again: once the held write is released, two
+// writes must have been started, the held one and one more.
+func TestAppendStartsOneWaitingWrite(t *testing.T) {
+	t.Setenv("DATA_DIR", t.TempDir())
+
+	synctest.Test(t, func(t *testing.T) {
+		buf := startBuffer(t)
+
+		var writes atomic.Int64
+
+		release := make(chan struct{})
+
+		// Each write is held until release is closed, then fails, as its
+		// file is closed under it.
+		buf.OnFileCreated(func(f *os.File) {
+			writes.Add(1)
+
+			<-release
+
+			_ = f.Close()
+		})
+
+		err := buf.Append(map[string]string{"id": "one-waiting"})
+		if err != nil {
+			t.Fatalf("report of the held write: %v", err)
+		}
+
+		go func() { _ = buf.Flush() }()
+
+		// Wait returns once the flush's write is held.
+		synctest.Wait()
+
+		big := map[string]string{
+			"id": strings.Repeat("a", reportbuf.FlushSizeThreshold),
+		}
+
+		// Errorf, not Fatalf, until the held write is released, so that a
+		// failure here does not leave it held.
+		for range 2 {
+			err = buf.Append(big)
+			if err != nil {
+				t.Errorf("report of FlushSizeThreshold bytes: %v", err)
+			}
+		}
+
+		close(release)
+
+		// Wait returns once every write started is done.
+		synctest.Wait()
+
+		started := writes.Load()
+
+		// Writes from here on, the final one at stop included, succeed.
+		buf.OnFileCreated(func(*os.File) {})
+
+		if started != 2 {
+			t.Fatalf("%d writes started, want 2: the held one and one "+
+				"at the threshold", started)
+		}
+	})
+}
+
+// appendToThreshold appends reports to buf until it holds
+// FlushSizeThreshold bytes, so that the last append starts writing them
+// to a report file, and returns the lines appended.
+func appendToThreshold(t *testing.T, buf *reportbuf.Buffer) string {
+	t.Helper()
+
 	// Large reports, so the threshold takes a few hundred appends.
 	pad := strings.Repeat("a", 64<<10)

@@ -703,22 +1407,7 @@ func TestFlushAtSizeThreshold(t *testing.T) {
 		appended.WriteByte('\n')
 	}

-	// Append writes the file in the background, so wait for it.
-	deadline := time.Now().Add(10 * time.Second)
-
-	for {
-		files, err := readReportFiles(dir)
-		if err == nil && len(files) == 1 && files[0] == appended.String() {
-			return
-		}
-
-		if time.Now().After(deadline) {
-			t.Fatalf("%d report files (error: %v), want one holding the "+
-				"%d bytes appended", len(files), err, appended.Len())
-		}
-
-		time.Sleep(10 * time.Millisecond)
-	}
+	return appended.String()
 }

 // reportFilesBytes returns the total size of the report files in dir.
@@ -835,85 +1524,3 @@ func hasReportFile(t *testing.T, dir string) bool {

 	return false
 }
-
-// TestFilesDeletedOldestFirstWhenWritesOverlap holds the write of an
-// older report file open until a newer one's write completes, then
-// releases it. When room is needed, the older file is deleted first,
-// though its write was the last to complete.
-func TestFilesDeletedOldestFirstWhenWritesOverlap(t *testing.T) {
-	const maxBytes = 1000
-
-	report := map[string]string{"id": "overlap"}
-
-	t.Setenv("DATA_DIR", t.TempDir())
-	t.Setenv("DATA_DIR_MAX_BYTES", strconv.Itoa(maxBytes))
-
-	buf := startBuffer(t)
-
-	created := make(chan string)
-	release := make(chan struct{})
-
-	buf.OnFileCreated(func(f *os.File) {
-		created <- f.Name()
-
-		<-release
-	})
-
-	err := buf.Append(report)
-	if err != nil {
-		t.Fatalf("older report: %v", err)
-	}
-
-	flushed := make(chan error)
-
-	go func() { flushed <- buf.Flush() }()
-
-	older := <-created
-
-	// Only the older write is held; the newer one goes through, and
-	// so do the writes after it, the final one at stop included.
-	var newer string
-
-	buf.OnFileCreated(func(f *os.File) { newer = f.Name() })
-
-	// Errorf, not Fatalf, until the older write is released, so that a
-	// failure here does not leave it held.
-	err = buf.Append(report)
-	if err != nil {
-		t.Errorf("newer report: %v", err)
-	}
-
-	err = buf.Flush()
-	if err != nil {
-		t.Errorf("flush of the newer report: %v", err)
-	}
-
-	close(release)
-
-	err = <-flushed
-	if err != nil {
-		t.Fatalf("flush of the older report: %v", err)
-	}
-
-	info, err := os.Stat(newer)
-	if err != nil {
-		t.Fatalf("stat %s: %v", newer, err)
-	}
-
-	// A report that fits beside the newer file alone, so deleting the
-	// older one makes exactly the room it needs.
-	pad := maxBytes - int(info.Size()) - lineBytes(t, map[string]string{"id": ""})
-
-	err = buf.Append(map[string]string{"id": strings.Repeat("a", pad)})
-	if err != nil {
-		t.Fatalf("report that needs room: %v", err)
-	}
-
-	if exists(t, older) {
-		t.Fatal("older report file kept when room was needed")
-	}
-
-	if !exists(t, newer) {
-		t.Fatal("newer report file deleted before the older one")
-	}
-}
@@ -34,13 +34,16 @@ func (s *Server) SetupRoutes() {
 	s.router.Use(s.mw.Recoverer())
 	s.router.Use(middleware.RequestID)
 	s.router.Use(s.mw.Logging())
+	// s.mw.Recoverer would write a panic's 500 only after the panic has
+	// passed the request log, so it is written here, where the log sees it.
+	s.router.Use(s.mw.Write500OnPanic())
 	s.router.Use(s.mw.SecurityHeaders())
 	s.router.Use(s.mw.CORS(s.params.Config.CORSAllowedOrigins))
 	s.router.Use(s.mw.MaxBodyBytes(maxRequestBodyBytes))
 	s.router.Use(middleware.Timeout(requestTimeout))

-	// Sentry reports a panic, then panics again, so that s.mw.Recoverer
-	// still answers 500.
+	// Sentry reports a panic, then panics again, so that it still
+	// reaches s.mw.Recoverer.
 	if s.params.Config.SentryDSN != "" {
 		s.router.Use(sentryhttp.New(sentryhttp.Options{Repanic: true}).Handle)
 	}
@@ -64,6 +67,8 @@ func (s *Server) SetupRoutes() {
 		// other.
 		if s.params.Config.MetricsUsername != "" {
 			r.Use(s.mw.Metrics(registry))
+			// Likewise inside the metrics.
+			r.Use(s.mw.Write500OnPanic())
 		}

 		r.Get("/.well-known/healthcheck", s.h.HandleHealthCheck())
@@ -66,6 +66,7 @@ func TestReportsAreRateLimited(t *testing.T) {
 			http.MethodPost, "/api/v1/reports",
 			strings.NewReader(`{"clientId":"c1","hosts":[]}`),
 		)
+		req.Header.Set("Content-Type", "application/json")
 		srv.ServeHTTP(rec, req)

 		return rec.Code
@@ -2,9 +2,10 @@
 # script/frontend-test: run the frontend test suite on the host: the
 # unit tests in test/unit/, through the test script in package.json,
 # then the production build, which fails on broken code. make test runs
-# the same in the frontend stage of Dockerfile. The tests print a dot
-# each; if any fails, they run again with every test listed, and the
-# script fails even if that run passes. NODE_OPTIONS chooses the
+# the same tests in the frontend-test stage of Dockerfile, and the image
+# build runs the production build in its frontend-build stage. The tests
+# print a dot each; if any fails, they run again with every test listed,
+# and the script fails even if that run passes. NODE_OPTIONS chooses the
 # reporter because yarn adds its arguments after the test files, where
 # node would take a reporter option for one more file.
 set -eu
@@ -7,6 +7,8 @@
 # headless Chrome at it over CDP, and asserts on computed layout at every
 # viewport width derived from the app's own CSS. See
 # test/viewport/README.md for what this covers and what it cannot.
+# Before the browser starts, it checks that the page is served with
+# Cache-Control: no-cache and Content-Security-Policy.
 #
 # Deliberately not part of script/check: it needs Docker and takes far
 # longer than the 20s budget make test has to stay inside.
@@ -64,8 +66,8 @@ main() {
    # configs/nginx.conf is a template: the image renders it over its own
    # default.conf, with the same port and limit bin/entrypoint.sh uses.
    # The empty file it includes trusts no proxy, as bin/entrypoint.sh
-    # writes it when TRUSTED_PROXIES is unset. It also includes the
-    # security headers, so the page runs under the shipped policy.
+    # writes it when TRUSTED_PROXIES is set and empty. It also includes
+    # the security headers, so the page runs under the shipped policy.
    docker run -d --rm --name "$SERVER" \
        --network "$NETWORK" --network-alias netwatch \
        -e PORT=8080 -e NGINX_ENVSUBST_FILTER='^PORT$' \
@@ -75,6 +77,32 @@ main() {
        -v "$ROOT/configs/security-headers.conf:/etc/nginx/security-headers.conf:ro" \
        "$SERVER_IMAGE" > /dev/null

+    # The page, and /no-such-page, which location / answers with the
+    # page through its try_files fallback, must go out with
+    # Cache-Control: no-cache. That add_header in its location drops the
+    # server's own unless the location includes the security headers
+    # again; only Content-Security-Policy is checked, as it comes from
+    # the same file as the rest. curl tries up to ten more times, a
+    # second apart, while nginx starts. It runs without -S, which would
+    # print an error for every failed attempt, even when a later one
+    # succeeds.
+    for path in / /no-such-page; do
+        if ! headers="$(docker exec "$SERVER" curl -s -o /dev/null -D - \
+            --retry 10 --retry-connrefused --retry-delay 1 \
+            "http://127.0.0.1:8080$path")"; then
+            echo "frontend-viewport-test: nginx did not answer $path" >&2
+            exit 1
+        fi
+        headers="$(printf '%s\n' "$headers" | tr -d '\r')"
+        if ! printf '%s\n' "$headers" | grep -qix 'Cache-Control: no-cache' ||
+            ! printf '%s\n' "$headers" | grep -qi '^Content-Security-Policy:'; then
+            echo "frontend-viewport-test: $path needs" \
+                "Cache-Control: no-cache and Content-Security-Policy; got:" >&2
+            printf '%s\n' "$headers" >&2
+            exit 1
+        fi
+    done
+
    # The image's own entrypoint already exposes CDP on 9222 and passes
    # --no-sandbox, so only extra flags belong here; re-specifying the
    # debugging port collides with it and leaves the endpoint bound to
@@ -6,6 +6,7 @@ import { beforeEach, test } from "node:test";
 import assert from "node:assert/strict";
 import {
    AppState,
+    buildUI,
    CONFIG,
    greyOutUI,
    hostRowHTML,
@@ -13,20 +14,26 @@ import {
    humanDuration,
    latencyClass,
    latencyHex,
+    log,
    measureLatency,
    Reporter,
    tick,
+    updateClocks,
+    updateSummary,
 } from "../../web/src/main.js";

 // There is no page here, so the tests stand in for it. The debug log looks
-// for its panel by id and finds none. Each element of a host's row that
-// tick or greyOutUI draws into is a plain object, made the first time a
-// test looks it up and kept in elements under its selector until the next
-// test starts. As on a page, writing its text replaces its markup; the
-// status dot greyOutUI looks for in it is not there. Drawing a sparkline
-// does nothing; it looks for the pixel ratio on window and finds none. In
-// each test, console.error and console.info print nothing and keep what
-// they are given.
+// for its panel by id and finds none, an element the page creates is a
+// plain object, and no frame the page asks for ever comes. Each element of
+// a host's row that tick or greyOutUI draws into is a plain object, made
+// the first time a test looks it up and kept in elements under its
+// selector until the next test starts. As on a page, writing its text
+// replaces its markup; the status dot greyOutUI looks for in it is not
+// there. Drawing a sparkline does nothing; it looks for the pixel ratio on
+// window and finds none. The commit the page's footer names, which
+// vite.config.js fills in when it builds the page, is blank. In each test,
+// console.error and console.info print nothing and keep what they are
+// given.
 const doNothing = () => {};
 let elements;
 beforeEach((t) => {
@@ -46,8 +53,13 @@ const canvasContext = {
    arc: doNothing,
 };
 globalThis.window = {};
+globalThis.requestAnimationFrame = doNothing;
+globalThis.cancelAnimationFrame = doNothing;
+globalThis.__COMMIT_HASH__ = "";
+globalThis.__COMMIT_FULL__ = "";
 globalThis.document = {
    getElementById: () => null,
+    createElement: () => ({}),
    querySelector: (selector) =>
        (elements[selector] ??= {
            getContext: () => canvasContext,
@@ -71,6 +83,11 @@ function statusText(state, host) {
    return elements[`.status-text[data-host="${index}"]`]?.innerHTML;
 }

+// What this test wrote to the browser console with console[method].
+function consoleLines(method) {
+    return console[method].mock.calls.map((call) => call.arguments[0]);
+}
+
 // Mocks the clock for test t, so that a check lasting seconds takes no real
 // time, and replaces fetch with targets that each answer after
 // answerAfter(url) milliseconds of that clock, or never when that is
@@ -190,9 +207,10 @@ test("at a 30000ms interval, a target answering after 1000ms shows in its row wh
 });

 // In the next three tests, the answering target's check is still waiting
-// when something happens after which its result must not show.
+// when something happens after which its result must not be recorded or
+// show in its row, and the check must write nothing to the console.

-test("at a 30000ms interval, a check still waiting when its round is given up does not show in its row", async (t) => {
+test("at a 30000ms interval, a check still waiting when its round is given up is not recorded and does not show in its row", async (t) => {
    CONFIG.updateInterval = 30000;
    const state = new AppState([
        { name: "Answering", url: "https://answering.test" },
@@ -208,9 +226,13 @@ test("at a 30000ms interval, a check still waiting when its round is given up do
    roundChecks.abort();
    assert.notEqual(await settled(round), "still waiting");
    assert.equal(latencyFigure(state, answering), undefined);
+    assert.deepEqual(answering.history, []);
+    assert.equal(answering.status, "pending");
+    assert.deepEqual(consoleLines("error"), []);
+    assert.deepEqual(consoleLines("info"), []);
 });

-test("at a 30000ms interval, a check still waiting when the user pauses does not show in its row", async (t) => {
+test("at a 30000ms interval, a check still waiting when the user pauses is not recorded and does not show in its row", async (t) => {
    CONFIG.updateInterval = 30000;
    const state = new AppState([
        { name: "Answering", url: "https://answering.test" },
@@ -225,9 +247,13 @@ test("at a 30000ms interval, a check still waiting when the user pauses does not
    t.mock.timers.tick(500);
    assert.equal(await settled(round), "still waiting");
    assert.equal(latencyFigure(state, answering), undefined);
+    assert.deepEqual(answering.history, []);
+    assert.equal(answering.status, "pending");
+    assert.deepEqual(consoleLines("error"), []);
+    assert.deepEqual(consoleLines("info"), []);
 });

-test("at a 30000ms interval, a check in the first round does not show in its row", async (t) => {
+test("at a 30000ms interval, a check in the first round is not recorded and does not show in its row", async (t) => {
    CONFIG.updateInterval = 30000;
    const state = new AppState([
        { name: "Answering", url: "https://answering.test" },
@@ -239,6 +265,10 @@ test("at a 30000ms interval, a check in the first round does not show in its row
    t.mock.timers.tick(1000);
    assert.equal(await settled(round), "still waiting");
    assert.equal(latencyFigure(state, answering), undefined);
+    assert.deepEqual(answering.history, []);
+    assert.equal(answering.status, "pending");
+    assert.deepEqual(consoleLines("error"), []);
+    assert.deepEqual(consoleLines("info"), []);
 });

 test("at a 30000ms interval, after the user pauses and resumes during a round, no row reads paused once its last check ends", async (t) => {
@@ -288,11 +318,6 @@ function stateWithOneTarget() {
    return { state, target: state.local[0] };
 }

-// What this test wrote to the browser console with console[method].
-function consoleLines(method) {
-    return console[method].mock.calls.map((call) => call.arguments[0]);
-}
-
 test("a check that fails with a network error writes one console line with the target, the time, the error and how long the request took", async (t) => {
    const { state, target } = stateWithOneTarget();
    mockTargets(t, () => 23, target.url);
@@ -471,6 +496,39 @@ test("of the checks interval changes cleared before they were reported, the next
    assert.equal(sent.at(-1).t, 120000);
 });

+// panel stands in for the open debug log panel, and for the element in it
+// that holds its lines, whichever the page looks up by id; the lines are
+// kept in order in lines. Drawing the whole panel, as opening it does,
+// replaces every line with a new one.
+test("with the debug log open, writing more lines than it keeps adds each to the panel and leaves the lines already shown as they are", (t) => {
+    const lines = [];
+    const panel = {
+        classList: { contains: () => false },
+        children: lines,
+        get firstElementChild() {
+            return lines[0];
+        },
+        append: (line) => lines.push(line),
+        removeChild: (line) => lines.splice(lines.indexOf(line), 1),
+        replaceChildren: (...newLines) =>
+            lines.splice(0, lines.length, ...newLines),
+    };
+    t.mock.method(globalThis.document, "getElementById", () => panel);
+    const kept = CONFIG.maxLogEntries;
+
+    for (let i = 1; i <= kept; i++) log.info(`line ${i}`);
+    const shown = [...lines];
+    for (let i = kept + 1; i <= kept + 5; i++) log.info(`line ${i}`);
+
+    assert.deepEqual(
+        lines.map((line) => line.textContent.match(/line \d+$/)?.[0]),
+        Array.from({ length: kept }, (_, i) => `line ${i + 6}`),
+    );
+    // The first five lines shown were dropped, and the rest are the same
+    // lines, not drawn again.
+    shown.slice(5).forEach((line, i) => assert.equal(lines[i], line));
+});
+
 // The page shows &lt; &gt; &quot; &amp; and &#39; in a row's markup as
 // < > " & and '.
 test(`a target whose name and URL hold < > " & and ' shows those characters in its row`, () => {
@@ -540,6 +598,32 @@ test("a check that timed out or found its target unreachable has a grey figure a
    assert.equal(latencyHex(null), "#6b7280");
 });

+// buildUI draws the page into #app, here a plain object. Each entry in the
+// footer's color legend is a colored dot and the latencies it stands for,
+// with < and > written as &lt; and &gt;. A latency takes the color of the
+// first entry in CONFIG.latencyColors whose limit it is below, so the
+// legend has, for each of those entries in order, a dot of its color (a
+// bg- class where CONFIG.latencyColors has a text- one) and the latencies
+// below its limit, or, for the last, which has no limit, the latencies
+// from the limit before it up. A gray dot for offline comes after them.
+test("the footer's color legend has each entry in CONFIG.latencyColors, with its color and limit, then offline", (t) => {
+    const app = {};
+    t.mock.method(globalThis.document, "getElementById", () => app);
+    buildUI(new AppState([]));
+
+    const legend = [
+        ...app.innerHTML.matchAll(/(bg-\w+-500) [^"]*"><\/span>(\S+)/g),
+    ].map(([, dot, latencies]) => [dot, latencies]);
+    const colors = CONFIG.latencyColors;
+    const expected = colors.map((color, i) => [
+        color.className.replace("text-", "bg-"),
+        color.below === Infinity
+            ? `&gt;=${colors[i - 1].below}ms`
+            : `&lt;${color.below}ms`,
+    ]);
+    assert.deepEqual(legend, [...expected, ["bg-gray-500", "offline"]]);
+});
+
 // A target whose checks, in turn, answered after each of latencies ms, or,
 // for null, found it unreachable.
 function hostAfter(latencies) {
@@ -616,6 +700,102 @@ test("the summary's min, max, median and average latency over the WAN targets' l
    });
 });

+// In the next two tests each element of the summary that updateSummary
+// writes into is a plain object, made the first time it is looked up and
+// kept in elements under "#" and its id.
+
+// The interval menu resets each target and counts the rounds from 0 again,
+// then updates the summary, before any check since the change is recorded.
+test("after an interval change, the summary's Reachable slot reads --/-- in white, as at page load", (t) => {
+    t.mock.method(
+        globalThis.document,
+        "getElementById",
+        (id) => (elements[`#${id}`] ??= {}),
+    );
+    const state = new AppState([]);
+    for (const host of state.wan) {
+        host.pushSample(Date.now(), { latency: 30, error: null });
+    }
+    state.tickCount = 5;
+
+    for (const host of state.allHosts) host.reset();
+    state.tickCount = 0;
+    updateSummary(state);
+
+    assert.equal(elements["#summary-reachable"].textContent, "--/--");
+    assert.equal(elements["#summary-reachable"].className, "text-white");
+});
+
+test("once every WAN target's check has failed, the summary's Reachable slot reads 0 of them, in red", (t) => {
+    t.mock.method(
+        globalThis.document,
+        "getElementById",
+        (id) => (elements[`#${id}`] ??= {}),
+    );
+    const state = new AppState([]);
+    for (const host of state.wan) {
+        host.pushSample(Date.now(), { latency: null, error: "timeout" });
+    }
+
+    updateSummary(state);
+
+    assert.equal(
+        elements["#summary-reachable"].textContent,
+        `0/${state.wan.length}`,
+    );
+    assert.equal(elements["#summary-reachable"].className, "text-red-400");
+});
+
+// In the next two tests each of the header's clocks is a plain object, made
+// the first time updateClocks looks it up and kept in elements under "#" and
+// its id.
+
+// The mocked clock reads 1970-01-01T00:00:00.000Z.
+test("the header's clocks show the local time with its time zone's name, and the UTC time", (t) => {
+    t.mock.method(
+        globalThis.document,
+        "getElementById",
+        (id) => (elements[`#${id}`] ??= {}),
+    );
+    t.mock.timers.enable({ apis: ["Date"] });
+
+    updateClocks();
+
+    assert.match(
+        elements["#clock-local"].textContent,
+        /^Local: \d{4}-\d\d-\d\d \d\d:\d\d:\d\d \S/,
+    );
+    assert.equal(
+        elements["#clock-utc"].textContent,
+        "UTC: 1970-01-01 00:00:00 UTC",
+    );
+});
+
+// Here Intl.DateTimeFormat is as in Chrome before 95, Firefox before 91 or
+// Safari before 15.4: it throws a RangeError when asked for the time zone's
+// "shortGeneric" name.
+test("in a browser that cannot name the time zone, updating the clocks throws nothing and leaves them blank", (t) => {
+    t.mock.method(
+        globalThis.document,
+        "getElementById",
+        (id) => (elements[`#${id}`] ??= {}),
+    );
+    const DateTimeFormat = Intl.DateTimeFormat;
+    t.mock.method(Intl, "DateTimeFormat", function (locales, options) {
+        if (options?.timeZoneName === "shortGeneric") {
+            throw new RangeError(
+                "Value shortGeneric out of range for Intl.DateTimeFormat options property timeZoneName",
+            );
+        }
+        return new DateTimeFormat(locales, options);
+    });
+
+    updateClocks();
+
+    assert.equal(elements["#clock-local"].textContent, "");
+    assert.equal(elements["#clock-utc"].textContent, "");
+});
+
 // An app state in which, of the WAN targets, the first timedOut timed out,
 // the next unreachable were found unreachable, the next answered answered
 // after latency ms, and the rest have not been checked yet.
@@ -144,13 +144,15 @@ const WAN_HOSTS = [

 const debugLog = [];

-const log = (() => {
+export const log = (() => {
    function append(level, message) {
-        debugLog.push({ timestamp: new Date(), level, message });
+        const entry = { timestamp: new Date(), level, message };
+        debugLog.push(entry);
        if (debugLog.length > CONFIG.maxLogEntries)
            debugLog.splice(0, debugLog.length - CONFIG.maxLogEntries);
        const panel = document.getElementById("debug-panel");
-        if (panel && !panel.classList.contains("hidden")) renderDebugLog();
+        if (panel && !panel.classList.contains("hidden"))
+            appendDebugLogLine(entry);
    }
    return {
        error: (msg) => append("error", msg),
@@ -827,7 +829,7 @@ function wanRowsHTML(wanHosts) {
    return rows.join("");
 }

-function buildUI(state) {
+export function buildUI(state) {
    const app = document.getElementById("app");
    app.innerHTML = `
    <div class="mx-auto px-[5%] py-8">
@@ -909,7 +911,7 @@ function buildUI(state) {
          <span class="inline-block w-3 h-3 rounded-full bg-lime-500 mr-1 ml-3 align-middle"></span>&lt;100ms
          <span class="inline-block w-3 h-3 rounded-full bg-yellow-500 mr-1 ml-3 align-middle"></span>&lt;200ms
          <span class="inline-block w-3 h-3 rounded-full bg-orange-500 mr-1 ml-3 align-middle"></span>&lt;500ms
-          <span class="inline-block w-3 h-3 rounded-full bg-red-500 mr-1 ml-3 align-middle"></span>&gt;500ms
+          <span class="inline-block w-3 h-3 rounded-full bg-red-500 mr-1 ml-3 align-middle"></span>&gt;=500ms
          <span class="inline-block w-3 h-3 rounded-full bg-gray-500 mr-1 ml-3 align-middle"></span>offline
        </p>
        <p class="mt-2"><a href="https://git.eeqj.de/sneak/netwatch/commit/${__COMMIT_FULL__}" target="_blank" rel="noopener" class="text-gray-600 hover:text-gray-400">${__COMMIT_HASH__}</a></p>
@@ -993,7 +995,7 @@ function updateHostRow(host, index) {
    SparklineRenderer.draw(canvas, host.history);
 }

-function updateSummary(state) {
+export function updateSummary(state) {
    const stats = state.wanStats();
    const hstats = state.wanHistoryStats();

@@ -1012,13 +1014,20 @@ function updateSummary(state) {
    const hmaxUnitEl = document.getElementById("summary-hmax-unit");
    if (!reachableEl) return;

-    reachableEl.textContent = `${stats.reachable}/${stats.total}`;
-    reachableEl.className =
-        stats.reachable === stats.total
-            ? "text-green-400"
-            : stats.reachable === 0
-              ? "text-red-400"
-              : "text-yellow-400";
+    // Before any WAN host's check is recorded, as after an interval change,
+    // no host has answered or failed, so the slot reads as at page load.
+    if (state.wan.every((h) => h.status === "pending")) {
+        reachableEl.textContent = "--/--";
+        reachableEl.className = "text-white";
+    } else {
+        reachableEl.textContent = `${stats.reachable}/${stats.total}`;
+        reachableEl.className =
+            stats.reachable === stats.total
+                ? "text-green-400"
+                : stats.reachable === 0
+                  ? "text-red-400"
+                  : "text-yellow-400";
+    }

    if (stats.min !== null) {
        minEl.textContent = `${stats.min}`;
@@ -1105,11 +1114,37 @@ function updateHealthBox(state) {
    else if (health === "healthy") log.notice(`Health: ${labels[health]}`);
 }

+// Writes the local time, with its time zone's name, and the UTC time into
+// the header's clocks. The clocks only decorate the page: in a browser
+// whose Intl.DateTimeFormat rejects the "shortGeneric" time zone name
+// (Chrome before 95, Firefox before 91, Safari before 15.4) they are left
+// blank.
+export function updateClocks() {
+    let local = "";
+    let utc = "";
+    try {
+        const now = new Date();
+        const zone = new Intl.DateTimeFormat("en-US", {
+            timeZoneName: "shortGeneric",
+        })
+            .formatToParts(now)
+            .find((p) => p.type === "timeZoneName").value;
+        const pad = (n) => String(n).padStart(2, "0");
+        local = `Local: ${now.toLocaleString("sv-SE", { hour12: false })} ${zone}`;
+        utc =
+            `UTC: ${now.getUTCFullYear()}-${pad(now.getUTCMonth() + 1)}-${pad(now.getUTCDate())} ` +
+            `${pad(now.getUTCHours())}:${pad(now.getUTCMinutes())}:${pad(now.getUTCSeconds())} UTC`;
+    } catch {
+        // Both clocks are left blank.
+    }
+    document.getElementById("clock-local").textContent = local;
+    document.getElementById("clock-utc").textContent = utc;
+}
+
 // --- Debug Log Renderer ------------------------------------------------------

-function renderDebugLog() {
-    const el = document.getElementById("debug-log");
-    if (!el) return;
+// One line of the debug log panel, its message set as text.
+function debugLogLine(entry) {
    const levelColors = {
        error: "text-red-400",
        warning: "text-orange-400",
@@ -1117,19 +1152,43 @@ function renderDebugLog() {
        info: "text-gray-300",
        debug: "text-gray-500",
    };
-    el.replaceChildren(
-        ...debugLog.map((entry) => {
-            const ts = formatUTCTimestamp(entry.timestamp);
-            const lvl = entry.level.toUpperCase().padEnd(7);
-            const line = document.createElement("div");
-            line.className = levelColors[entry.level] || "text-gray-400";
-            line.textContent = `${ts} ${lvl} ${entry.message}`;
-            return line;
-        }),
-    );
+    const ts = formatUTCTimestamp(entry.timestamp);
+    const lvl = entry.level.toUpperCase().padEnd(7);
+    const line = document.createElement("div");
+    line.className = levelColors[entry.level] || "text-gray-400";
+    line.textContent = `${ts} ${lvl} ${entry.message}`;
+    return line;
+}
+
+// Draws every line of the debug log into the panel, when it is opened.
+function renderDebugLog() {
+    const el = document.getElementById("debug-log");
+    if (!el) return;
+    el.replaceChildren(...debugLog.map((entry) => debugLogLine(entry)));
    el.scrollTop = el.scrollHeight;
 }

+// The request that scrolls the open panel to its newest line in the next
+// frame. One waits at most, also in a hidden tab, where frames do not run.
+let debugLogScrollFrame = 0;
+
+// Adds entry to the open panel as its last line, and drops the first line
+// once the panel holds more than maxLogEntries. A check that ends while the
+// page is busy is timed only when the page is done, so the lines already
+// shown are left as they are, and the panel is scrolled once a frame
+// rather than once a line: scrolling it lays out all of its lines again.
+function appendDebugLogLine(entry) {
+    const el = document.getElementById("debug-log");
+    if (!el) return;
+    el.append(debugLogLine(entry));
+    if (el.children.length > CONFIG.maxLogEntries)
+        el.removeChild(el.firstElementChild);
+    cancelAnimationFrame(debugLogScrollFrame);
+    debugLogScrollFrame = requestAnimationFrame(() => {
+        el.scrollTop = el.scrollHeight;
+    });
+}
+
 // --- Sorting -----------------------------------------------------------------

 // Sort WAN hosts: pinned hosts first (alphabetically), then unpinned hosts
@@ -1477,24 +1536,6 @@ async function init() {
        sortAndRebuildWAN(state);
    });

-    function updateClocks() {
-        const now = new Date();
-        const pad = (n) => String(n).padStart(2, "0");
-        const utc =
-            `${now.getUTCFullYear()}-${pad(now.getUTCMonth() + 1)}-${pad(now.getUTCDate())} ` +
-            `${pad(now.getUTCHours())}:${pad(now.getUTCMinutes())}:${pad(now.getUTCSeconds())} UTC`;
-        const local =
-            now.toLocaleString("sv-SE", { hour12: false }) +
-            " " +
-            new Intl.DateTimeFormat("en-US", { timeZoneName: "shortGeneric" })
-                .formatToParts(now)
-                .find((p) => p.type === "timeZoneName").value;
-        document.getElementById("clock-local").textContent = "Local: " + local;
-        document.getElementById("clock-utc").textContent = "UTC: " + utc;
-    }
-    updateClocks();
-    setInterval(updateClocks, 1000);
-
    // Rounds never overlap: a round first gives up the last round's checks
    // if they are still waiting, and the last round then records nothing
    // more. At a steady interval they never are, as they time out at 80% of
@@ -1574,6 +1615,11 @@ async function init() {

    window.addEventListener("resize", () => handleResize(state));
    setTimeout(() => handleResize(state), CONFIG.resizeDelay);
+
+    // The clocks start last, after the rounds: they only decorate the page,
+    // so nothing in them may stop the monitor from probing.
+    updateClocks();
+    setInterval(updateClocks, 1000);
 }

 // Bootstrap only when loaded as the page: a real DOM containing the #app
Author	SHA1	Message	Date
sneak	6092c62223	Report file writes run one at a time (closes #150 ) check / check (push) Waiting to run Details Two writes that failed at the same time put their reports back in the wrong order when the older one failed first. A mutex is now held across each write, from taking the waiting reports out of the buffer to putting them back if the write fails, so the next write takes them in order. The write Append starts at the 10 MiB threshold waits for one under way, and Append starts no other while it waits. Files written since the start are now deleted to make room in the order they were written. A test that needed two writes at once is removed; new tests cover the one waiting write and the order files are deleted in. Model: opus-5-5	2026-10-19 01:59:25 +00:00
clawbot	648c225919	Discarded-check unit tests check nothing was recorded (closes #151 ) check / check (push) Waiting to run Details The three unit tests of a check whose result is discarded (its round given up, the user paused, or the first round) checked only that the row showed nothing. Each now also checks that the host's history is still empty, its status still pending, and nothing was written to the console, so a change that recorded the result and skipped only the row fails them. The console helper moves above its first use. Model: opus-5-5	2026-10-19 03:44:13 +02:00
clawbot	be6dc93eaa	Unusable DATA_DIR error names DATA_DIR; report file number is atomic.Int64 (closes #148 , closes #161 ) check / check (push) Waiting to run Details When the backend runs on its own, a DATA_DIR that could not be created or read stopped the start with an error that named the path but not DATA_DIR, and one where no file could be created let it start, to fail at the first write of a report file. The start now creates a file in DATA_DIR and removes it, and all three cases fail with DATA_DIR "<value>": <reason>, the form config.New uses for the other settings. Tests start the buffer with DATA_DIR below a regular file and with DATA_DIR set to /proc. The report file sequence counter is atomic.Int64 instead of atomic.Uint64, as the Go style guide asks for int types. File names are unchanged. Model: opus-5-5	2026-10-19 03:15:35 +02:00
clawbot	6df743f7fa	Page is sent with Cache-Control: no-cache (closes #141 ) check / check (push) Waiting to run Details nginx sent index.html with no Cache-Control, so a browser could keep it after a deploy and ask for hashed assets the new build no longer has, and show an empty page. location / now adds Cache-Control: no-cache, which covers its try_files fallback to /index.html, and includes security-headers.conf again, since its own add_header drops the server's. /assets/ and the paths passed to the backend are unchanged. script/frontend-viewport-test fails unless / and a path that falls back to the page carry Cache-Control: no-cache and Content-Security-Policy, which comes from the same file as the other security headers. Model: opus-5-5	2026-10-19 03:10:39 +02:00
clawbot	984047e6ca	README states what refused reports, one sender and nginx's 413 do (closes #145 , closes #146 , closes #147 ) check / check (push) Waiting to run Details The README said the page keeps refused samples and sends them later. Those in the 100-round history are sent only while it holds them: at the 1-second interval one refused report loses the oldest 20 seconds of each host's samples, and a pause longer than the history loses every sample between the last delivered report and the pause. Samples an interval change cleared are kept apart, the last 100 per host, until delivered. Report limits now says one address at the defaults can fill the 1 GiB cap in about 17 minutes. In the image, nginx answers a body over 1 MiB itself, with 413 and an HTML page; the backend's own 413 sentences now apply to it run on its own. Model: opus-5-5	2026-10-19 02:55:29 +02:00
clawbot	007190b541	Clock errors no longer stop probing; legend reads >=500ms (closes #149 , closes #152 ) check / check (push) Waiting to run Details The header's clocks start last in init, after the rounds, and an error formatting them, as in a browser that rejects the "shortGeneric" time zone name, leaves both clocks blank instead of stopping the page before its first round. updateClocks moved out of init so a unit test can call it. The footer legend's red entry reads >=500ms, as README.md and CONFIG.latencyColors have it. buildUI is exported so a unit test can draw the page and check the legend against every entry of CONFIG.latencyColors. Model: opus-5-5	2026-10-19 01:34:45 +02:00
clawbot	fd32f348b7	prepare-data-dir leaves files with a second name to their owner (closes #143 ) check / check (push) Waiting to run Details A hard link in /data shares its owner with a file that can be anywhere on the same filesystem, so chowning it gave that file to the netwatch user. A file with more than one name now keeps its owner and its path is logged. Owners are now set on the opened directory, file or link after checking it, as chmodDir sets modes, so a hard link swapped in afterwards is caught. Every other file is opened without following it; a symbolic link is given to the netwatch user itself, and devices, named pipes and sockets keep their owner. This is built only for Linux: prepare-data-dir runs only in the Linux image, and macOS has neither O_PATH nor AT_EMPTY_PATH; elsewhere it returns an error. A failed owner change names the file. Model: opus-5-5	2026-10-19 01:11:41 +02:00
clawbot	2532da55e9	Client address is the right-most untrusted X-Forwarded-For entry (closes #144 ) check / check (push) Waiting to run Details From a peer in TRUSTED_PROXIES, clientIP took the left-most X-Forwarded-For entry, which the client writes itself behind a proxy that appends to the header, so a client could choose the address the rate limit counts. It now walks the entries from the right past trusted proxies and takes the first that is not one. An entry with a port is read as its address; with only trusted entries, or one that is not an address at all, it uses the peer. Every X-Forwarded-For header line is read, and X-Real-IP only when there is no X-Forwarded-For entry. Model: opus-5-5	2026-10-19 00:28:18 +02:00
clawbot	e43a710114	Summary reads --/-- after an interval change, not 0/26 in red (closes #140 ) check / check (push) Waiting to run Details The interval menu resets every host to pending and updates the summary, which counted the WAN hosts with a latency recorded, none, and showed 0/26 in red until the second round after the change ended. While every WAN host still waits for its first check, the Reachable slot now reads --/-- in white, as at page load. updateSummary is exported for a unit test that resets the hosts as the interval menu does; a second test checks that the slot still reads 0/26 in red once every check failed. Model: opus-5-5	2026-10-18 23:00:35 +02:00
clawbot	2861be1c4d	Stopping the server waits for report file writes under way (closes #122 ) check / check (push) Canceled after 0s Details The report buffer's stop now waits, through a sync.WaitGroup, for flushLoop and for the write Append starts at 10 MiB before its final flush. It used to return at once, so the process exiting cut such a write off and its reports were lost. A write under way that fails puts its reports back in the buffer, so the final flush writes them; the stop fails only if that flush fails too. Tests hold the periodic write and the one at 10 MiB open in a synctest bubble and find the stop waiting until each is released; a third makes the held write fail and finds its report written by the final flush. Model: opus-5-5	2026-10-14 14:47:01 +02:00
clawbot	580d12a934	make test no longer runs the frontend's production build (refs #118 ) check / check (push) Canceled after 0s Details yarn build moves out of the stage that runs the frontend's unit tests, now named frontend-test, into a frontend-build stage of its own. It starts from frontend-test for its yarn packages and source, installs the git that vite.config.js reads the commit with, and copies from the lint and test phases as builder does, so it cannot run unless they passed. The runtime stage copies dist/ from it. The Go compile, most of make test's time, is unchanged and waits on the ruling asked for in sneak/prompts#113; no build cache is added. Model: opus-5-5	2026-10-14 14:30:03 +02:00
clawbot	12f3dd5b07	Debug log adds each line instead of drawing the whole panel (closes #131 ) check / check (push) Canceled after 0s Details With the debug log open, every line written drew the panel again with up to 1000 lines, and a check that ends while the page draws is timed only when the drawing is done, so latencies measured with the panel open read too high. Each line is now appended to the open panel and its first line dropped once it holds more than maxLogEntries; the whole panel is drawn only when it is opened. Scrolling to the newest line reads the panel's height, which lays out all its lines, so it runs once a frame, with at most one frame request waiting. Model: opus-5-5	2026-10-14 14:13:03 +02:00
clawbot	2a47fb9a8e	Unset TRUSTED_PROXIES trusts RFC 1918, empty trusts none (closes #129 ) check / check (push) Canceled after 0s Details The backend's own TRUSTED_PROXIES now follows the ruling for every app: unset, it trusts 10.0.0.0/8, 172.16.0.0/12 and 192.168.0.0/16, and no longer loopback; a list replaces that; set and empty, it trusts no proxy. viper reads an empty environment variable as unset, so an empty value got the default. config.New now reads this one setting with os.LookupEnv first. viper's AllowEmptyEnv would apply to every setting, so an empty PORT or DEBUG would stop the start instead of counting as unset, as README.md says it does. bin/entrypoint.sh still starts the backend with TRUSTED_PROXIES=127.0.0.1/32. Model: opus-5-5	2026-10-14 14:01:37 +02:00
clawbot	cff7b07bbc	Reports refused while report files cannot be written (closes #123 ) check / check (push) Canceled after 0s Details A failed write of a report file used to drop its reports, and POST /api/v1/reports kept answering 200 for as long as writes failed. Now a failed write puts its reports back in the buffer, still counted against DATA_DIR_MAX_BYTES, and Append refuses every report, which the handler answers 500, until a write succeeds. The minute flush tries the held reports again, so storing resumes with no restart once writing works, and the buffer takes nothing new while writes fail. Tests also cover a report appended while a write that fails runs, and the cap once a failed write's reports are written. Model: opus-5-5	2026-10-14 13:47:14 +02:00
clawbot	e628dfe6cf	A handler's panic is logged and counted as 500 (closes #125 ) check / check (push) Canceled after 0s Details Recoverer comes first, so it wrote the 500 only after the panic had passed the request log and the metrics, which both recorded 200. Write500OnPanic, placed after the request log and after the metrics, writes the 500 as the panic passes and panics again, so Sentry and Recoverer still get it. Recoverer now writes its 500 only when no status was written, so the server logs no superfluous WriteHeader call. The test also requires Recoverer's log line for the handler's panic, so it fails if Write500OnPanic stops passing the panic on. Model: opus-5-5	2026-10-14 13:30:11 +02:00
clawbot	cbcde098dc	Unset TRUSTED_PROXIES trusts the RFC 1918 ranges (closes #119 ) check / check (push) Canceled after 0s Details With TRUSTED_PROXIES unset, bin/entrypoint.sh gives nginx one set_real_ip_from line for each of 10.0.0.0/8, 172.16.0.0/12 and 192.168.0.0/16, as sneak ruled on 2026-09-22 for every app. A list replaces them, and set and empty trusts no proxy; each entry is checked as before. The backend inside the container still trusts only 127.0.0.1/32. README.md, configs/nginx.conf and the comment in script/frontend-viewport-test say the new default. Model: opus-5-5	2026-10-14 13:01:34 +02:00
clawbot	73b39e485b	Reports are taken only as application/json (closes #126 ) check / check (push) Canceled after 0s Details A page on another origin can POST text/plain or a form without a CORS preflight, so its visitors' browsers could store reports although the server sends no CORS headers. decodeJSON now refuses a request whose Content-Type is not application/json (parameters such as charset allowed), and the report handler answers it 415 with the usual error body, storing nothing. A JSON POST waits for the preflight, which only an origin in CORS_ALLOWED_ORIGINS passes; the page already sends application/json. Model: opus-5-5	2026-10-14 12:43:25 +02:00