changelog
Everything we shipped, including the parts that broke
Named changes with the numbers that moved. Breaking changes are marked and explained rather than folded into a release summary, and when we cause an incident it gets an entry like anything else.
Currently on 2.11.0 · tracepoint.dev/changelog.atom
2.11.0
AddedDeploy correlation on the incident timeline
Incidents now show the deploys that landed in the 30 minutes before the signal moved, with the diff range for each. The first question in every incident channel is 'did we ship something', and now it is answered before anyone asks.
- AddedDeploy markers on every latency and error chart, sourced from the /v1/deploys endpoint or the GitHub Actions integration.
- AddedSuspected-cause ranking: deploys touching files that own the affected route pattern are surfaced first.
- ImprovedIncident timelines load in one request instead of three; median open time dropped from 1.9s to 380ms.
2.10.2
FixedPercentile accuracy at low request volume
Routes seeing fewer than roughly 40 requests per minute could report a p99 well above the true value, because the sketch was being read before it had enough observations to be meaningful.
- Fixedp95 and p99 now return null rather than a misleading figure below 30 observations in the window, and the chart draws a gap instead of a spike.
- FixedAlert rules on p99 no longer fire on the first minute after a service comes back from zero traffic.
- ImprovedSparse-route charts label the gap explicitly ('insufficient sample') on hover rather than leaving a bare hole.
2.10.0
AddedLog anomaly baselining, generally available
Pattern fingerprinting and per-pattern volume baselines are out of beta and on for every Team and Enterprise workspace. Baselines build over a trailing 14 days, so a workspace that enables it today starts alerting reliably in about two weeks.
- AddedNovel-pattern detection: a stack trace never seen before that appears more than 25 times in an hour opens a low-severity incident.
- AddedPer-pattern mute, with an expiry. Mutes without an expiry are the reason nobody trusts log alerting, so they are not offered.
- ImprovedFingerprinting now strips UUIDs, ULIDs, and 12+ digit numerics before hashing, which collapsed one design-partner workspace from 41,000 patterns to 1,900.
2.9.0
BreakingTwo-region agreement before an incident opens
Synthetic check failures now require two probe regions to agree before an incident is created. This removes the largest single source of false pages we had, and it changes behaviour for anyone relying on single-region alerting.
- BreakingWorkspaces with only one probe region enabled will no longer open uptime incidents. Starter workspaces are unaffected — single-region checks there now use a three-consecutive-failure rule instead.
- ImprovedMedian false-page rate across all workspaces fell from 2.1 per week to 0.3 in the four weeks after rollout.
- AddedThe check detail view shows each region's result for the last 200 runs, so a consistently flaky region is visible rather than inferred.
2.8.1
ImprovedRoute pattern inference for Go and Elixir
The Go and Elixir middleware now read the matched route pattern from the router rather than the raw path, which was producing one series per URL on any app using path parameters.
- Improvedchi, gin, echo, and net/http ServeMux patterns are read directly from the request context in the Go middleware.
- ImprovedPhoenix routes resolve through the router's route_info/4 rather than being regex-matched.
- FixedWorkspaces that had already ingested high-cardinality paths can collapse them retroactively from Settings → Routes; the rewrite applies to historical data.
2.8.0
AddedOwnership sync from CODEOWNERS
Service and route ownership can now be derived from the CODEOWNERS file in your repository, so the routing table stops drifting from reality the moment a team reorganises.
- AddedGitHub and GitLab CODEOWNERS sync, re-read on every push to the default branch.
- AddedUnowned-route report, emailed weekly to workspace admins. A route with no owner pages the fallback rotation, and knowing which ones do is most of the fix.
- ImprovedManual ownership overrides survive a sync and are marked as overrides in the UI rather than silently reverting.
2.7.0
AddedPrometheus read-back endpoint
Every workspace now exposes its computed series in Prometheus exposition format. If you want the numbers in a Grafana you already run, you can have them, on any plan including the free one.
- AddedGET /v1/metrics returns latency percentiles, error rates, and check results as gauges, scoped by API key.
- AddedA published recording-rules file for teams migrating off a self-hosted Prometheus, mapping our series names to the common ones.
2.6.3
FixedAlert delivery during our own degradation
During a 22-minute degradation of our us-west-2 ingest on 4 February, alert delivery was delayed for 41 workspaces rather than failing over. The failover path existed and was not triggered. Full write-up is on the status page.
- FixedAlert dispatch now fails over on queue depth rather than on health-check failure, which is the signal that actually moved during the incident.
- AddedDelivery-latency monitoring on our own alert pipeline, published on the status page alongside ingest availability.
- ImprovedWorkspaces affected by a delivery delay now see a banner on the incident itself, rather than having to correlate with our status page by hand.
Subscribe to the feed, or don't — we won't email you about a patch release
The Atom feed carries every entry on this page. Breaking changes also go to workspace admins by email, 30 days ahead of the change, with the affected workspaces named.