The RPC layer emitted the event as soon as the handler returned the
SSE response, before the import ran, so the event could not say which
files the import created. Emit it from inside the SSE handler once the
import ends, with the ids and count of the created files and whether
the import completed or failed.
Closes#12485
AI-assisted-by: claude-opus-5-5
The backport of docker/ from develop (2df5f417a) also brought the subpath
support from #12358 and its fix (95bc48a08). It needs the Admin Console
from 2.20, so with the current one every Admin Console URL fails. Restore
the previous nginx configuration and docker-compose.yaml.
AI-assisted-by: Claude
Signed-off-by: David Barragán Merino <david.barragan@kaleidos.net>
`SET STORAGE DEFAULT` needs PostgreSQL 16, but the official docker
images pin `postgres:15`, so migration 0156 failed on those
deployments. The migration now uses `SET STORAGE EXTENDED`, the
storage mode of the type of every column it touches (text, text[],
jsonb and bytea), which leaves the catalog in the same state on any
supported version.
The documented PostgreSQL floor goes back to >=15, the reason for
raising it was that syntax.
Verified on the test database: the 62 columns reset and no table
column keeps the legacy EXTERNAL storage.
The backend memory now records that migrations must stay within
PostgreSQL 15 syntax.
AI-assisted-by: minimax-m3.1-flash-free
The prod-infra memory now says that the `$host` multi-variable needs
its `query`, its `options` list and its `allValue` regex updated
together. A host present only in `query` is dropped when `All` is
selected, so the panels that filter with `instance=~"$host..."` stop
showing it. Panel 104 is the visible symptom.
AI-assisted-by: minimax-m3.1-flash-free
`app.redis/del` logged `:key key`, but `key` is not bound in that
`let`: it resolved to `clojure.core/key`, so the line carried
`#object[clojure.core$key ...]` instead of the keys that failed to be
deleted. Log `:keys keys`, which is always the normalized vector, so a
single key and a collection report the same shape.
The tests pin the line: the keys, the hint, the error level, the logger
and the timeout as cause, plus the `nil` the caller gets.
The prod-infra memory now notes that client-side redis timeouts mean the
JVM is paused, not that the server is slow.
AI-assisted-by: deepseek-v4.1-flash
Add the missing 2.18.4 entries for the closed storage GC,
version quota and file_change revision issues, with their
fix PRs referenced inline.
AI-assisted-by: muse-spark-1.3-contributor
Drop the legacy TOAST STORAGE EXTERNAL overrides set when PostgreSQL
ran on ZFS with transparent compression. Migration 0156 restores the
type default (SET STORAGE DEFAULT) on 62 columns across 21 tables;
existing TOAST data keeps its strategy until rewritten.
Raise the documented PostgreSQL floor to >=16, required by the new
SET STORAGE DEFAULT syntax.
AI-assisted-by: muse-spark-1.3-contributor
Comment out the media-storage-bytes-per-team check in
upload-file-media-object. The SUM over storage_object takes
3-12s per upload on prod (3.5TB DB, ~34 slow runs/hour)
and blocks uploads. Keep the SQL and quote definition
for re-enable with a cheaper implementation.
Update the upload quota test to document the disabled
state until the new implementation lands.
AI-assisted-by: muse-spark-1.3-contributor
Binfile v1/v3 import and the v3 export get-file step hold a single
transaction while the ZIP is decoded and the file data is realized.
Disabling idle_in_transaction_session_timeout left a stuck connection
holding locks (and a pool slot) forever.
Raise it to 30 minutes instead. The longest import/export observed in
production is under 8 minutes, so the limit still bounds a stuck
session. statement_timeout stays at 300 s.
Add set-transaction-idle-timeout and configure-import-transaction in
app.binfile.common, use them from both paths, and cover the helper
with a test.
AI-assisted-by: deepseek-v4.1-flash
* 🐛 Keep file_change revisions on their own deletion schedule
Restoring a file from the trash cleared the deletion mark on every
file_change row of that file, revisions included. A revision is born
already scheduled for deletion one hour later, and objects-gc only
looks at rows that carry a date, so unmarking one left it in the
table for ever: production had 358,517 of them across 3,614 files.
The file deletion cascade had the mirror problem: it rewrote the date
of rows that already had one, pushing a revision out to the trash
lifetime.
Restrict the restore to durable history and date only the rows that
have no date yet, so a revision keeps the schedule it was born with.
A pinning test covers the snapshot update path, which cannot reach a
revision because it resolves its row through a read filter. The
backend memories now carry the model of the three file_data classes
and this invariant, in place of the two superseded stubs.
Closes#12418
AI-assisted-by: deepseek-v4.1-flash
* 📚 Add autovacuum and Postgres exporter notes to Grafana memory
Record what this Grafana instance can and cannot show about autovacuum
so the next session does not have to re-derive it:
- The Postgres exporter scrapes the primary as well as the
read-only replica, so the primary's pg_stat_activity and
pg_settings_* are in Prometheus even though the SQL
datasource cannot reach them.
- It publishes no pg_stat_progress_* collector, and
pg_stat_activity_count has no backend_type label: autovacuum
workers are the usename/application_name-less bucket with
state=active. The launcher shows up as the AutoVacuumMain
wait event, and it is absent on the replica.
- A metric the exporter does not publish needs a custom query
block in its YAML queries file and a restart of the exporter.
- The Postgres server ships its own log to Loki (the job is
postgresql, primary only), whose streams carry only filename,
instance, job and level. The completed vacuum lines live
there: a per-table history, not a view of what runs now.
- The PostgreSQL section now ends with the vacuum panels,
with the Vacuum TXID age table last; an in-flight per-table
view would need a pg_stat_progress_vacuum collector in the
exporter itself.
AI-assisted-by: deepseek-v4.1-flash
* ✨ Extend backend histogram buckets beyond 7.5s
Give every backend timing histogram a tail past the shared 7500ms cap
so slow RPC calls, CLIMIT waits and background tasks report real
quantiles instead of clipping at the top bucket.
Request and wait metrics grow to 60s, background tasks to 300s and
Redis eval to 10s; the sub-7.5s buckets stay untouched to keep the
history comparable.
Closes#12408
AI-assisted-by: muse-spark-1.3-contributor
* 🐛 Fix flaky assertion in climit republish test
Read the last published waiting value instead of the exact
publication list. The sampler thread started with the component
runs its first tick asynchronously, and when it lands inside the
recording window it publishes the same count twice, which made
the full-suite CI run fail while focused runs stayed green.
AI-assisted-by: muse-spark-1.3-contributor
* ✨ Add file_data blob to JSON decode helper
Allow decoding file_data blobs (app.util.blob) to JSON
for investigations with jq, offline from a blob file or
from nREPL with the backend running.
AI-assisted-by: muse-spark-1.3-contributor
* ✨ Add validation step timings to update-file logs and metrics
Measure the six phases inside process-changes-and-validate
(resolve-libs, process-changes, soft and hard affected and
schema validations) with tpoint and try/finally, so timings
are recorded even when a hard validation throws.
Publish each phase to a new histogram
penpot_rpc_update_file_validation_timing with a fixed step
label using the request buckets (up to 60000 ms), and extend
the update-file debug log with the same *-ms fields plus
file, project and team ids. The deferred log only runs on
success, so failures get an inline error log with partial
timings.
AI-assisted-by: muse-spark-1.3-contributor
* 💄 Translate decode-file-data comments to English
Translate the Spanish docstrings, section headers and test
messages of the file_data decode helper to English, keeping
code untouched. The repo convention is English in code and
comments; the helper was first written in Spanish.
AI-assisted-by: muse-spark-1.3-contributor
* 🐛 Count saved versions in the version quota
The snapshots-per-file and snapshots-per-team quotes counted the
saved versions by file_change.data, a column that only the legacy-db
file data backend writes: versions keep their payload in file_data,
so the count returned 0 and a configured quota never applied.
Count by label instead, which tells a version apart with every file
data backend, and add a partial index on file_change with the same
predicate so both counts are answered by an index only scan instead
of walking every row of every file of the team (10.3 seconds in
production for a team with 829 versions).
Closes#12404
AI-assisted-by: deepseek-v4.1-flash
* ♻️ Drop the ! suffix from the snapshot quota test helpers
The check helpers assert the outcome of a command instead of mutating
anything, so they do not take the ! that this codebase reserves for
functions whose contract changes state or runs a command (the
th/command! call inside them already carries it).
AI-assisted-by: deepseek-v4.1-flash
The implement-plan flow asked to commit the closed plan together with
the code, so the resulting commit looked like it was missing a piece.
State the rule in the three places that describe it: the implement-plan
skill, the commit workflow memory and the agents README. A plan is
session bookkeeping, not part of the change, and its absence from the
commit is correct rather than a missing piece.
AI-assisted-by: deepseek-v4.1-flash
Local scratch files and autogenerated build output were
showing up in git status:
- /*.html and /*.json cover local reports and token
dumps at the repo root; !/package.json keeps the
tracked root manifest visible to git.
- /media-processor/target/ and /scripts/__pycache__/
cover build and Python cache output.
- render-wasm artifacts copied into the exporter go in
their own commented block.
Record the exporter and render-wasm generated paths in
the module memories so nobody edits or commits them by
hand.
AI-assisted-by: deepseek-v4.1-flash
The storage-gc-deleted task used to stream all the pending rows with
db/plan inside one long read-only transaction, blocking the table
vacuum for the whole run and generating bloat on long deletions.
Now every chunk of 25 objects is one short read-write transaction:
the SELECT reads and locks the next objects after a (deleted_at, id)
keyset cursor with FOR UPDATE SKIP LOCKED, deletes the database rows
grouped by backend and removes their backend content. Concurrent runs
never take the same chunk, and the double SELECT of the old lock-ids
step is gone.
A chunk that fails aborts its transaction: nothing of the chunk is
deleted, its locks are released, and the run skips it; the cursor
still advances to the last row read, so a failed chunk is not visited
again in the same run.
AI-assisted-by: glm-5.3-flash
Edits wait up to 3s in a buffer before they are sent. Closing the
laptop or switching away right after an edit could leave that batch
unsent, and a crash or a discarded tab then lost it.
The batch is now sent at once when the tab is hidden (another tab, a
minimized window, a closed tab or laptop) or the window loses focus
(another app or a plugin iframe). The 3s window itself is unchanged.
pagehide and beforeunload are left out: they come too late for an
async save, and the tab is already hidden by then.
The .penpot download is built from the server copy. Users who
downloaded a file to keep their work while saves were failing got an
old or empty file, with no warning.
Opening the download dialog now sends the pending changes of the files
and waits up to 10s for them. If some are still not saved, the dialog
says the download will not include them.
A failed export raised the exporter error to the global handler, which
replaced the workspace with the error page. Users then reloaded and
lost the changes that were not saved yet.
- A failed single or multiple export now shows an error toast and
keeps the workspace. Exporter defects still report as unhandled.
- A failed multiple export request ends its progress listeners, so the
leave-page warning it turned on is released.
- Before exporting, the pending changes of the exported files are sent
and the export waits up to 10s for them, instead of 400ms. Exports
are built from the server copy, so if some changes are still unsaved
a toast warns that the export may not include them.
force-persist-files-and-wait waits only for the given files, so a stuck
save of another file does not delay the export.
We only learn how often users lose work from support tickets. Each time
the user is told a save failed, persistence now emits a save-failed
audit event with the cause type, the code, whether the queue is still
retrying and how many changes are waiting.
It follows the same rule as the failure toast: a repeated failure of
the slow retry cycle is not counted again. Failures handed to another
handler, such as an expired session, are counted too, since they also
stop the save.
A save waits for the lock that another save holds on the same file. A
slow save, such as one on a large file, blocks the other saves to that
file. The waiting request also holds the user's only save slot on the
backend, so the user's saves to other files wait behind it too.
Those saves ended in answers that stopped autosave for good:
- a 429 after 60s, when no save slot freed up in time;
- a 504 written by the backend after 300s, with :statement-timeout,
when Postgres cancelled the wait for the lock.
In both cases the backend never applied the save, so sending it again
is safe. transient-error? now treats them as temporary, and the save
takes the retry path. A statement timeout takes 300s per attempt, so
the slow failure limit still bounds it. GET requests and error reports
are unchanged: the check lives in the save pipeline, not in
retryable-types.
Both answers were reproduced in the devenv by holding the file lock
from a separate database session.
* ✨ Count the application logs as a metric, by logger and level
The error panels had no metric to read: they counted log lines in Loki,
because the backend published nothing about the errors of the application
(the only thing alive in Prometheus was the http listener error count,
which measures another thing). Now there is
`penpot_log_events_total{logger, level}`.
The counter comes from a new component, `app.loggers.metrics/reporter`,
which hooks a watch on the log record like the database and mattermost
reporters do, but without a thread or a channel: the increment is in
memory, so it runs in the watcher itself. Only `warn`, `error` and `fatal`
are counted, and only the records that go through `app.common.logging`:
the lines a library writes straight to log4j2 are not counted (one in a
thousand today).
The body of the watcher is wrapped in a `try/catch` that swallows the
error on purpose: the watcher fires from the `swap!` of the log record,
inside the logging path, so an exception there would reach whoever was
logging and would take the logging executor down with it. A test pins that
a failing publication does not propagate.
The suite removes the reporter (it would count the logs of the tests and
the recording of their `mtx/run!`), so a test of its own pins the wiring.
The memories carry the counter, its two limits and why the `try/catch` is
not noise; the grafana one also brings the notes of this afternoon about
the service account, the dashboard search and Cloudflare, which were
waiting for a commit.
AI-assisted-by: deepseek-v4.1-flash
* 📚 Address the review findings on the log-events metric
The review of the counter found that the reason written for the
`try/catch` was not the real one, so the memory and the comments say now
what happens if it is removed: with `::l/sync?` the exception reaches
whoever was logging, and on the async path it dies in the uncaught
handler of the logging executor -- outside the application's logging --
where the pool replaces the worker, so logging continues and that line is
lost. They also say that a publication failing every time is silent: the
counter stays flat and nothing is written about it.
The rest are small things the review asked for: the test of the `fatal`
level (it was in the counted set and nothing verified it), the HELP of the
metric naming the three levels, the require of the test without the unused
alias that clj-kondo flagged, and the note that the `level` is lowercase
here and uppercase in Loki.
The grafana memory loses what was said twice: the two pairs of bullets
about reading the API (the service account 403s and the search that lies)
are merged, the CLIMIT notes are one bullet instead of two, and the four
bullets about editing dashboards are grouped in their own section. Its
host name is gone as well, which the first line of that file forbids.
AI-assisted-by: deepseek-v4.1-flash
Port staging c74d415 (#12252) to main, adapted to this branch.
The :export-shapes modal event carried the full :exports vector,
each export embedding its shape, which could reach megabytes and
wedge the audit archive queue past the proxy body limit. Build a
compact body instead: totals, distinct shapes and per-type counts,
plus the page name. Other modals keep their props.
Move the modal :origin prop to ::ev/origin so make-proto-event
records it as :event-origin; the UI still reads the plain prop.
Bound exception messages at the source: get-hint clamps to
max-message-length (1000 chars) and the CLJS format-throwable
clamps the trace message line and honors :trace-length. The
Context Hint header in generate-report prints get-hint instead
of the raw hint. Also fix first-line on the JVM, where
cuerdas/index-of returns nil and (pos? nil) throws.
Adaptations to main: skip audit-log.md (absent here), skip the
generate-report huge-message test (staging async helpers are
absent), and fit the errors.cljs hunk to main's generate-report.
AI-assisted-by: muse-spark-1.3-contributor
* ✨ Add async test helpers and the with-mocks* macro
Backport of the test scaffolding that the persistence retry work needs,
taken from staging. Purely additive: no existing helper changes behavior,
so the suite result is unchanged (740 tests, 0 failures).
- frontend-tests.helpers.async: ->promise, await-response, settle,
observe and wait-for. wait-for checks immediately and then polls one
macrotask at a time, bounded, so it fails instead of hanging.
- mock/with-mocks*: installs mocks with set! and evaluates its body in a
generated ^:async fn, so the test reads sequentially and there is no
done to thread. Evaluates to a promise.
- mock/asap and mock/run-mocked: run-mocked awaits the body and restores
the originals always, in that order, reporting a rejection or a
non-promise return as an :error.
The matching change that made with-mocks defer its body via asap is NOT
backported: it would force all 58 existing uses across 14 files into
done-chained contexts, which is a separate refactor.
AI-assisted-by: space-bunny-free
* ✨ Add an error report governor with compact payloads
One browser session could emit an unbounded number of identical error
reports, and a dropped connection was reported as an application defect
with the full payload. Both are fixed here, in one place, because the
persistence retry work that follows depends on the new submit-report
contract.
- A report governor in app.main.errors. Each report carries a
fingerprint (report-name|type|code|hint|first stack frame); the first
occurrence is always emitted, repeats inside a 2 minute window are
counted, and the next emitted report carries them as :occurrences. The
cache is bounded by evicting the oldest entry, so memory stays fixed.
- flash reserves the report before generating it, so a suppressed
occurrence never pays the report-building cost. Its whole body now
runs inside one ts/schedule callback, a reporting or notification
failure is logged instead of propagating, and it returns a total
promise so tests can await it.
- generate-report accepts :format :compact|:full. The compact payload
keeps the context header plus type, code and uri, and skips the
stack, the ex-data dump (it can carry request headers) and the
last-events list.
- environment-error-types is derived from rp/retryable-types, so the
transport classification has a single source of truth. Environment
failures are audit-only handled-exceptions and never reach the
exception page or the alerts. :offline gets its own handler instead
of falling through to :default.
:occurrences travels inside the audit event :props, which the backend
schema accepts as a free-form map, so no backend change is needed.
The dedup governor test file is new: 7 tests over the fingerprint, the
pure reservation decision, and the flash pipeline.
Relates to #11726, #11743
AI-assisted-by: space-bunny-free
* ♻️ Answer commit refusals with an empty stream
commit-changes used to return nil when it refused to build a commit,
which forced every caller that wanted to observe the outcome to
special-case nil. It now returns an empty stream, so a refusal and an
empty result look the same and callers can subscribe uniformly.
Only the refusal branch changes: the accepted branch already answered
with (rx/of commit). Potok treats a nil watch result as "nothing", so
both spellings were already correct at the store level.
The 57 call sites that wrap this event in rx/of are unaffected. They
hold the ::commit-changes event itself, not the value its watch
returns, so they keep processing exactly as before.
AI-assisted-by: space-bunny-free
* ✨ Retry transient saves with backoff and a reconnect notice
Any failed update-file landed on the same path: the queue halted in
:error, the retained changes were flushed away, and the user was told
to reload. A dropped connection and a revoked permission are not the
same thing, so they now take different paths.
transient-error? splits them, checking :type and :cause-type because
persistence-failed wraps the original cause. Transient failures enter
a :retrying episode: the head commit stays queued and resends with
backoff at 2s, 8s and 20s, then falls through to today's terminal path.
Everything else is terminal and keeps the exact :error path.
- Resends rotate the ::request-id stamp only once the old request has
left active-requests, carry the same :commit-id, and never double
send while one request is in flight.
- Retry timers carry an episode token; a timer whose episode was
superseded stays silent instead of sending or failing.
- One tagged reconnect notice per episode, hidden by tag on save and on
terminal failure. Recovery is silent.
- The browser online event re-enters the runner, but only for a live
:retrying episode, guarded by exists? js/window.
- wait-persisted-or-error waits through :retrying instead of cutting the
wait short.
The episode keeps its :run-id, so a new local edit only joins the queue.
Re-entering the runner on every edit would bypass the backoff and spend
one attempt per batched commit.
The retry test namespace is new: 13 tests over the classification, the
backoff schedule, stamp rotation, stale tokens and the notice lifecycle.
Relates to #11724
AI-assisted-by: space-bunny-free
* 🐛 Fix stalled saves that never answer and edits lost on leave
Two holes the retry work did not close.
An update-file request that never gets an answer kept its attempt
stamp forever, so the queue could not move and the user saw a saving
state that never resolved. The request now has a 7 minute timeout that
fails it as a transient network error, which hands it to the retry path
instead of blocking. The timeout is longer than the 5 minute stall
threshold on purpose: the stalled report must fire before the retry,
not instead of it.
The leave-page warning also had gaps. It now tracks unsaved edits
directly, through sync-unload-guard, which holds :persistence in
st/ongoing-tasks while edits sit in the buffer or the queue. Edits kept
in the queue after a failed save keep the guard on too, so closing the
tab warns instead of dropping them.
store.cljs guards its ongoing-tasks watcher with exists? js/window,
because the new tests exercise it without a window.
Four new tests: three for the guard across buffering, an in-flight save
and a retained edit, and one for a hung request that must resend and
save.
Closes#12223
AI-assisted-by: space-bunny-free
* 🐛 Skip the save when a missing font cannot be remapped
fix-deleted-fonts emitted text and typography changes even when no
replacement font existed, so opening a page that uses a deleted custom
font rewrote identical content and cleared position-data on every
visit. Every visit became a save with nothing to save.
Both loops now compare the remapped value with the original and emit
nothing when they match. The remapped path is unchanged.
Two new tests cover the no-replacement case and the remapped case.
Closes#12151
AI-assisted-by: space-bunny-free
* ✨ Add a debug helper that simulates an intermediary response
A save that never reaches the backend is hard to reproduce: you need a
proxy, a captive portal or a CDN in the way. This wraps fetch and
answers the named API method with a status and a body that carries no
Penpot error code, leaving every other request alone.
Without a duration the simulation is spent after answering one call.
With one it answers every matching call for that long, which is how to
watch the save retry, warn, keep retrying and finally recover.
debug.simulateHttpError('update-file')
debug.simulateHttpError('get-comment-threads', 403)
debug.simulateHttpError('update-file', 524, 120)
debug.clearHttpErrorSimulation()
Dev-only and self-contained: no production path reads it.
AI-assisted-by: space-bunny-free
* 🌐 Add the save retry and connection error strings
Three strings the new persistence and error paths need:
- errors.save-retrying, shown once per retry episode while the queue
resends a queued save after a connection loss.
- workspace.header.retrying, the save indicator state for an episode
in progress.
- errors.connection-error, shown by the network and offline handlers,
which no longer fall through to the generic message.
Added to en.po only. The translations script also normalizes every
other locale and drops 439 strings that were already stale before this
work; that belongs in its own maintenance pass, not in a backport.
AI-assisted-by: space-bunny-free
* 🐛 Report save failures as unhandled when they are our defect
flash-persistence reported every failure it did not delegate as
handled-exception, including the ones that are plainly our fault. A
validation failure means the client sent something the server refused,
yet it landed in the audit log with a name that skips the internal
reports and the alerts. Nobody was paged.
The report type now follows the cause. :validation, :server-error and
:internal report as unhandled-exception, so they alert. The rest are
unchanged.
Two exclusions are deliberate:
- Environment failures stay handled. They reach this point after the
retry budget is spent, and the user's connection is not our bug.
- :assertion stays handled, because its own handler already reports it
that way and the two paths should not disagree.
The cause type is read from :cause-type first, since persistence-failed
wraps the original under :type :persistence. Reading :type alone would
match the wrapper every time and change nothing.
The user still gets the same sticky toast and report link; only the
audit event name moves.
The test that asserted a validation failure is "still notified" now
asserts which event name it gets, across the bare and wrapped shapes,
the environment types that must stay handled, and an unclassified one.
AI-assisted-by: space-bunny-free
* 🐛 Unblock persistence when opening a file after a terminal failure
A terminal save failure set the state to :error and left the queue
alone. recover-persistence only resumed for :save-permission-denied, so
for every other terminal code both of its branches were false and it
emitted nothing. Leaving and reopening the same file therefore changed
nothing: the editor stayed halted until a reload, which rebuilds the
whole store and loses the queue anyway.
The path had two jobs under one name. initialize-persistence fires when
the workspace view mounts and tears down the previous subscription;
recover-persistence fires when a file is opened or the role changes.
They are now separate: initialization picks up a queue left behind by a
previous session, and deliberately does not lift a halt, so undeliverable
edits are released when the user acts rather than when the view mounts.
Recovery lifts the halt by releasing what can never be saved.
release-failed-file drops every queued commit of the file that failed,
then resume-persistence clears the error and settles an empty queue to
:saved. Retaining them until that point is deliberate: it is what keeps
wait-persisted-or-error rejecting, which is what stops
workspace/versions.cljs from building a snapshot out of unsaved edits.
Releasing at failure time would make waiters resolve instead.
Permission loss is treated as permanent too, which the previous
exception got wrong. The backend raises :not-found for a missing edit
permission, so a resend does not succeed — it routes the user to the
exception page. The old branch made that reachable: it resumed the save
whenever the cached can-edit flag said yes, and the server disagreed.
The only case where the flag can return to true is a team switch, where
the queue is gone anyway, so nothing real is lost by releasing.
Because the queue is shared, dropping the failed file is also what lets
the other files' pending edits drain. One file that cannot be saved no
longer blocks everything else.
Tests: the failure retains the queue and notifies once, recovery
releases it without replaying, a failed file does not block a second
one, permission loss is released rather than resumed, and initialization
leaves a halt alone.
AI-assisted-by: space-bunny-free
* ✨ Offer a reload on the save failure toast
A terminal save failure stops autosave for the whole session, but the
toast asked the user to reload without offering the reload. It listed
five implied actions, none of them reachable from the banner, and it
never said that the changes could not be saved.
flash now takes a :links vector of actions, appended after the download
link that :report-link? produces, so a caller can add recovery actions
without reaching into the report handling. flash-persistence passes a
reload built from labels.reload-page and location.reload, the pattern
already used elsewhere in the app for a full restart.
The message now says what happened, what is lost, and what to do about
it: autosave has stopped for this session, the changes since the last
successful save cannot be saved, and either reloading or leaving the
file and opening it again restarts it. Naming both matters, because the
toast links are easy to miss.
Saying the edits are lost is not optional: recovery releases them, so
without it the user finds out when they come back to the file.
Translated in en.po and es.po only. The translations script would also
normalize all 40 locales and drop entries that were already stale, which
is a maintenance pass of its own.
The reload callback cannot be invoked under the test runner, since
js/location does not exist there, so it is covered by its label and by
being callable. Whether the reload works is a manual check.
AI-assisted-by: space-bunny-free
* 🐛 Classify the responses the backend never wrote
A reply written by a proxy instead of the backend carries no Penpot
error code, so handle-response raised it as :internal. A 504 on a save
skipped the retry cycle and stopped autosave at once, and because
:internal reports as unhandled, it raised an alert. In a standard
deployment this is how a slow save ends: nginx answers 504 after 300s,
before the client's 7 minute timeout fires.
The status now decides the type:
- 504 and the Cloudflare 520-524 family raise :gateway-error, which
joins retryable-types. Saves retry with backoff like 502 and 503,
GETs retry, and a save that spends its retries reports as handled,
because the report rules derive from the same set.
- 429 raises :rate-limit, and any other error status
:unexpected-response. Neither retries; both show a toast instead of
the error page.
- :internal is kept for a response below 400 that does not decode.
Bodies we cannot read are cut to 500 characters, since proxy error
pages run to tens of kilobytes.
Tests: classification and retries per status, the transport handlers
keep the page, a 504 and a 524 built by the real handle-response go
through the save path and land on a retry, and the new types stay
handled once the retries are spent.
AI-assisted-by: claude-opus-5-5
* ✨ Keep retrying failed saves on a slow cycle
Once the three backoff retries were spent, a save that failed on the
connection stopped autosave like any terminal failure, and opening a
file again dropped its edits. A connection that comes back after a
minute lost the work all the same.
A transport failure that outlasts the backoff now arms a slow cycle:
one silent resend every 30s, until a save lands or 24h have passed
since the failure started. The cycle stops when persistence restarts,
the queue resumes another way, or the workspace closes.
A save that keeps failing slowly is not resent for a day. A gateway
that answers 504 after its own timeout may leave the backend applying
the save, so each resend can cost the whole save again, and the failing
file holds back every other file's queued edits. An attempt that fails
after 90s or more counts as a slow failure; at three, the slow cycle
stops sending the commit and the save fails as before. Fast failures,
such as being offline, never count.
- The first failure shows a toast that says the changes are kept and
keep being sent, with no reload link, since a reload loses them. A
save that lands hides it.
- Opening a file sends a queue still inside its retry window once
more instead of releasing it. Failures outside the window, and every
non-transport failure, are still released.
- A freshly loaded file gets back the queued edits of its version, so
reopening it shows the changes that are still waiting to be saved. An
edit that no longer applies to the loaded data leaves the file as the
server sent it, so the file still opens; the edits stay queued.
- A commit whose earlier request is no longer in flight is simply sent
again under the same commit id: changes are idempotent, so there is
no unknown-outcome guard any more. A request still in flight still
blocks a parallel send.
flash takes a :tag so the save failure toast can be hidden by tag.
Tests: the spent backoff arms the slow cycle, a 504 arms it, a terminal
failure does not, the cycle stops when the window closes or at the slow
failure limit, recovery resends instead of releasing, reopening restores
only the loaded file's edits and survives one that no longer applies,
and the retrying toast offers no reload.
* 🐛 Turn off the leave-page warning when failed edits are released
The leave-page warning only followed the queue on appends and discards.
Opening a file again releases the edits of a file that failed for good,
but that release changed the queue without either event, so the warning
stayed on with nothing left to save. Leaving the page asked about
unsaved changes until the next save went through.
The warning now also follows the release.
Tests: releasing the last failed edit turns the warning off.
* 🐛 Stop retrying failed saves outside the workspace
Leaving the workspace was meant to stop the slow retry cycle, but only
a cycle already armed stopped. A backoff timer kept firing after the
user left, and a backoff spent outside the workspace armed a slow cycle
that kept sending for up to a day from the dashboard. Those sends check
the edit permission of whatever team is open, so a switch to a team
where the user only views failed the queued edits for good, and the
next file they opened dropped them.
Leaving the workspace now suspends retries. A pending backoff timer
goes stale, and while suspended a transport failure schedules no
backoff, a spent backoff arms no slow cycle and the online event sends
nothing. The edits stay queued and a request in flight still lands.
Opening a file lifts the suspension and resends a retrying save whose
backoff was cut short; a role change does not lift it.
Tests: leaving stops a pending backoff, a failure after leaving
schedules no retry and ignores the online event, a backoff spent
outside the workspace arms no slow cycle, and opening a file saves the
held edit in each case.
---------
Co-authored-by: alonso.torres <alonso.torres@kaleidos.net>
Add a check-commit-guidelines job to commit-checker.yml
that runs scripts/check-commit on every new commit, so
the Co-authored-by ban holds in CI too. PR events check
out the base ref, keeping the executed script trusted.
AI-assisted-by: muse-spark-1.3-contributor
Signed-off-by: Andrey Antukh <niwi@niwi.nz>
The Co-authored-by trailer is now forbidden. The
commit checker fails any message holding it, whatever
the case, and the docs record the ban.
AI-assisted-by: muse-spark-1.3-contributor
Signed-off-by: Andrey Antukh <niwi@niwi.nz>
State in the contributing guide that issues under `needs triage` and
`internal` are not open to contributions. Every new issue starts under
`needs triage` until the product team reviews it, so contributors could
mistake a fresh issue for an invitation and send a PR that we then turn
down.
A PR opened on one of those issues may be closed without review,
without an explanation and without a reply. The wording follows the
existing note about AI-assisted content.
The help center page keeps its current wording for now.
AI-assisted-by: deepseek-v4.1-flash
Repair scripts run against production with a plain SQL client, not the
devenv psql: no `\set`, `\if` or `\copy` there. The SQL datasource the
agent can query is a read-only replica that rejects writes (a temporary
table included) and cuts long queries, so its queries are plain SELECTs
bounded by a list of ids: an unbounded scan over tens of millions of rows
does not come back.
AI-assisted-by: deepseek-v4.1-flash
The queue gauge was written by whichever limiter of a name ran last, so a
per-profile pile-up never showed: a healthy profile writing 0 overwrote the
15 of the profile that was queueing (4 of 1441 samples in production), and
a limiter whose requests never come back read 0 while 72 threads stayed
parked.
A queue is an entry and an exit, so the new gauge is accumulated: one
counter per limiter name and instance, up when a call starts waiting for a
permit, down when it stops, so a call that never leaves stays visible. It
is always written with `:val`, and a daemon sampler republishes every count
every 15s so a name whose traffic died does not go stale in the scrape;
halting the component stops that thread.
Rejections had no metric at all: penpot_rpc_climit_rejected_total counts
them by name and reason (queue-full or timeout), and the log lines now
carry queue-max, permits, code and reason, so an operator can tell whether
the limiter did its job or was saturated.
The gauges that could not tell the truth are gone: queue and permits (same
last-write-wins problem, and waiting plus the wait histogram replace them)
and http_worker_busy_threads (reads 0 on an instance with 72 blocked
threads; pool_size is the signal). No other dashboard or alert rule reads
them.
AI-assisted-by: deepseek-v4.1-flash
Add a per-process availability flag (app.health) that /readyz
reports as 503 OFFLINE while offline, so the load balancer
drains the instance and in-flight requests finish untouched.
Operator surface is app.srepl.main/set-offline and
app.srepl.main/set-online; a restart comes back online.
Covered by backend-tests.health-test; HA docs updated.
AI-assisted-by: muse-spark-1.3-contributor
`file_object_thumbnail` was replaced by `file_tagged_object_thumbnail`
and nothing refers to it anymore: no query in the backend touches it,
no table has a foreign key pointing at it, no view depends on it, and
it is empty in production. It also lacked `deleted_at`, which is why
restoring files had to leave it out. Only legacy naming survives in two
helpers (`get-file-object-thumbnails` and
`delete-file-object-thumbnails!`, which operate on
`file_tagged_object_thumbnail`).
The storage bucket named "file-object-thumbnail", the one that
`file_tagged_object_thumbnail` uses, is a different thing and stays.
AI-assisted-by: deepseek-v4.1-flash
Restoring a file only un-marked the database rows (`file`,
`file_media_object`, `file_change`, `file_data`, `file_thumbnail`,
`file_tagged_object_thumbnail`), not the storage objects those rows
reference. An object that was already marked for deletion stayed
marked while a live row pointed at it, and those foreign keys are `NO
ACTION`: the storage GC cannot delete it and skips it forever (and
before the GC learned to skip, it cost the whole nightly run).
Restoring a team had the same gap with its font variants.
Both restores now unmark the storage objects of the rows they bring
back: the media and the thumbnail of the file media objects, the media
of the file and object thumbnails, the `storage-ref-id` of the file
data, and the four files of a font variant. Objects referenced by other
rows are left alone.
AI-assisted-by: deepseek-v4.1-flash
objects-gc deleted every kind of object inside one transaction per proc,
looping until empty, so a single row that a foreign key still references
aborted the whole run: the task died on its first file every night, after
the earlier steps had already spent around twenty minutes. The per-file
savepoint only saved the files proc and left the same shape everywhere
else.
Now every proc streams its pending keys with db/plan inside a read-only
transaction of its own and deletes them chunk by chunk, each chunk in a
transaction of its own with FOR UPDATE SKIP LOCKED. A chunk that fails is
reported with l/err and skipped, so the run reaches the end and the next
run retries the skipped rows.
AI-assisted-by: muse-spark-1.3-contributor
`storage-gc-deleted` ran its whole batch inside one transaction: the
handler opened it and `delete-in-bulk` reused that connection, so the
"separate transaction" its comment promised did not exist. A single
row that a foreign key still references aborted that transaction, the
cursor then died on its next fetch ("current transaction is aborted")
and the task ended up throwing `unhandled exception on running task`
after rolling back the whole night of work. That is what has kept
production from deleting a single object since late April.
Now `clean-deleted` walks the pending objects with the cursor in its
own read-only transaction and every chunk runs in its own one, so a
blocked chunk is logged with its ids and skipped while the rest is
committed. `task finished` is reachable again.
The private functions also drop the `!` suffix: the namespace is all
mutation, so the mark adds nothing and reads as noise.
The test builds a blocked object (a `file_media_object` pointing at it
as its thumbnail) plus a full chunk of pending objects and one more in
the next chunk, and checks that the run finishes and that the object of
the next chunk is deleted.
AI-assisted-by: deepseek-v4.1-flash
The four `process-deleted-*-cascade` helpers of `app.srepl.main`
loaded every soft-deleted row of a table into memory (`db/exec!`)
and ran `:delete-object` inline (`wrk/invoke!`), so they could not
be used against production: `file` alone has 2.1M deleted rows.
They now share a single `submit-deleted-cascade!` that reads the
rows with a server-side cursor (`db/plan`, in a read-only
transaction) and submits one task per row, so the work goes
through the workers like any other task. The row id goes as
`label` with `dedupe` on, and the helpers accept `:limit`, `:ids`
and `:dry-run?` to run bounded batches. Public names and their
zero-arity form are unchanged.
Tests cover the task params, the row's own date (not `now`), the
`limit`/`ids`/`dry-run?` options and that each helper reads its own
table.
AI-assisted-by: deepseek-v4.1-flash
An instance left the audit events pipeline dead: every push-audit-events
answered with 429 and 69 worker threads parked for good inside a limiter.
Two code paths of one profile acquired the same two limiters in opposite
order: a command declaring its own limits (upload-file-media-object) took
process-image/by-profile and then root/by-profile, while a command with no
limits of its own took the roots first and then re-entered climit from
inside its handler with the same ids (create-file-media-object-from-url).
Each path held what the other one waited for, and no limiter had a timeout
to break the wait.
Now wrap puts the root limiters after the command's own, so they end up
outermost and every path meets them first: the nested chains carry only the
command's own limits and run inside them, which puts both paths in the same
order. impl-invoke also keeps, per thread, the set of limiters that thread
is inside of and runs the handler inline when a call asks for one it
already holds, so a handler can no longer queue behind itself.
The shipped climit.edn sets a timeout on every limiter (60s for operations
that can be slow, 30s otherwise), so no wait can park a thread for ever.
The production file lives outside the repository and has to be mirrored.
Tests cover the acquisition order, the deadlock, the reentrancy and the
timeouts of the shipped config.
AI-assisted-by: deepseek-v4.1-flash
The command declared its limit as a plain keyword next to
`::climit/key-fn`, metadata that `climit/wrap` does not read: it
takes the key only from the positional `key-fn` of the vector form.
The limit was a single shared one, so when one call got stuck it
held the only permit and every audit batch of that instance was
rejected with 429 from then on.
The test builds the limiter from the command metadata and checks
that two profiles end up with different limiters.
AI-assisted-by: deepseek-v4.1-flash
The footer had a hardcoded 2025 copyright and links that no longer match
penpot.app. The year is now set at build time. Sections, links and
social icons follow the current penpot.app footer, and the grid wraps
into two rows to fit the new sections.
Closes#12132
AI-assisted-by: claude-opus-5-5
Check a milestone against the Main project board with
scripts/project-anomalies.py (check writes
tmp/<MILESTONE>-ANOMALIES.md): open issues with merged PRs,
milestone mismatches either way, needs triage labels and
unassigned non-community issues. Outsiders resolve in bulk;
only a missing milestone auto-fixes, on confirmation.
Teach scripts/gh.py batched issue lookup (issue subcommand),
milestone on prs output, plus assignees and projects fields,
and retry transient HTTP 504s.
AI-assisted-by: muse-spark-1.3
Move the inline programs out of the update-changelog skill into
scripts/changelog.py (check-merged, cross-ref, report) and shrink
the skill to the workflow alone.
Teach scripts/gh.py to retry transient HTTP 504s, resolve lookups
in batches of 50 (new issue subcommand, milestone on prs output),
and report 💥 entries without the breaking change label.
Add the nine missing 2.17.0 changelog entries found while
trying the new flow.
AI-assisted-by: muse-spark-1.3
