Repository navigation
·
133 commits
to main
since this release
Immutable
release. Only release title and notes can be modified.
This release adds incident watchers, a per-project notification policy, an AI-assisted incident workflow and MCP tools that act on incidents. It also includes a database hardening pass that must be applied before you upgrade.
Upgrade notes
Read Upgrade before upgrading a self-hosted installation.
- Database migrations. Apply migrations
0047to0058before you update the app and the worker.0053converts every timestamp column totimestamptzand rebuilds the indexes on those columns, locking each table while it does. Plan a maintenance window on large installations. The migration stops, instead of guessing, if the database holds data and itsTimeZoneis not UTC.0056dropsplan_limits.sso_enabledand the legacystatus_page_subscribers.webhook_secretandunsubscribe_tokencolumns. Before it runs, the migrator encrypts any plaintext webhook secret withSECRET_ENCRYPTION_KEY, so nothing needs rotating; keep the key the app already uses, or the upgrade stops and rolls back. Unsubscribe links in emails sent before 1.6.0 stop working; every new email carries a signed link. The previous release still reads the dropped columns, so upgrade with workers stopped or in a short window, and don't roll back the image afterwards.0057and0058add the incident watcher and notification policy tables.
- Migration command.
db-migrate.js(npm run db:migrate) refuses to run unlessDATABASE_URLorDB_HOSTis set, instead of defaulting tolocalhost:5432. - New optional settings.
APP_HOST_PORTandRUSTFS_HOST_PORTchange the host ports the app and RustFS publish on.install.shandbackup.sh --restorefollow them. - MCP scopes. Clients can request the new optional scopes
mcp:incidents:writeandmcp:sre. Existing connections stay read-only until the user approves the new scopes. - Reports origin removed. The optional
REPORTS_ORIGINsetting andPOST /api/report-urlsare gone, and report HTML is served from the app origin again. RemoveREPORTS_ORIGINfrom your environment, and remove any DNS, TLS or proxy routing you added for that host, which would otherwise serve the full app. - Diagnostic queries. Saving a PromQL, LogQL or TraceQL query now rejects a free-text parameter outside quotes and a pattern that accepts any value. Add a pattern, enum or allowlist, or quote the placeholder. Existing queries are not rewritten, and at run time they accept plain tokens only.
- Monitor targets. Creating an HTTP, website, ping or port monitor, or changing its target or type, returns 400 when the target is a private, loopback or reserved address, including a hostname that resolves to one. Set
ALLOW_INTERNAL_TARGETS=trueon both the app and the worker to allow them. Existing monitors keep running.
Added
- Incident watchers. Watch an incident to receive its notes and severity changes by email. Viewers can watch too.
- Default notification policy for each project: the channels used for incidents, email to service owners for SEV1 and SEV2, and an optional reminder when an incident is still unacknowledged.
- Post to status page from the incident timeline, for incidents linked to a status page incident.
- Incident Copilot opens in a side drawer on the incident page. AI drafts the customer-facing status update and a recovery-aware resolution note for you to review.
- Automatic investigation, off by default: SEV1 and SEV2 incidents get a full investigation, SEV3 and SEV4 a triage. A failed run is retried on a later trigger, up to three attempts per incident.
- MCP incident tools.
assign_incident,acknowledge_incidentandresolve_incident(scopemcp:incidents:write), andtriage_incident,investigate_incidentandgenerate_incident_brief(scopemcp:sre). Every call is checked against the user's current role. - Organization deletion. Owners can delete their organization, and super admins can delete a user's sole-member organizations. Both remove the projects, data and stored files.
- Telemetry retention. Operational telemetry, such as Private Agent heartbeats and service health snapshots, is purged on a schedule set by
*_RETENTION_DAYSsettings (for examplePRIVATE_AGENT_HEARTBEAT_RETENTION_DAYS). Incident history and chat are never purged.
Changed
- Permission failures return 403 instead of 401 when the caller is signed in but not permitted, including the realtime queue and test status streams.
- Requirement coverage is calculated per test, from each test's own result in a job run.
- The runs table's Triggered by column shows the schedule in plain words for scheduled runs (for example "Every 4 hours"), and the key name and its owner for API runs.
- Job failure alerts are sent once per run and channel, even when runs complete at the same time.
- Uploaded avatars are converted to a 256 px WebP image.
- The run report fits the window height.
Fixed
- Database integrity. Foreign keys have explicit delete rules, missing indexes were added and redundant ones dropped, relationships and tenant columns are enforced, and status and type columns accept known values only.
- Status page subscribers. Verification tokens are hashed, webhook secrets are stored encrypted, and unsubscribe links use signed tokens.
- Accounts. Suspended accounts are rejected on every session path, and the verify-key, status page upload and extension sign-in routes are rate limited.
- A Playwright run is no longer executed again when only saving its result fails.
- The recorder asks before replacing unsaved changes in the Playground editor.
- MCP client registration and MCP requests no longer fail with a server error.
- AI test-fix requests with a malformed body return 400 instead of 500.
- MCP endpoint URLs that contain a key are stored encrypted.
- Status page incident emails are sent when no encryption key is configured.
- Monitor alerts are no longer lost when an internal check briefly fails; the check cycle is retried instead of aborted.