Skip to content

trunk-merge/pr-138568/fe36c99d-168f-494e-bcd3-1bcebb8cbd52 - #139242

Closed
trunk-io[bot] wants to merge 47 commits into
masterfrom
trunk-merge/pr-138568/fe36c99d-168f-494e-bcd3-1bcebb8cbd52
Closed

trunk-io[bot] wants to merge 47 commits into
masterfrom
trunk-merge/pr-138568/fe36c99d-168f-494e-bcd3-1bcebb8cbd52

Conversation

@trunk-io

@trunk-io trunk-io Bot commented Nov 20, 2026

Copy link
Copy Markdown
Trunk Merge Pull Request Banner

This pull request was created and is being managed by Trunk Merge for the master merge queue.

This pull request is based on the master branch at SHA 290eb90ca5ffc6863752588dd77080a8e76b997f.

See more details about each PR in the batch here:

When CI completes, this pull request will be closed automatically.

Pull Requests Being Tested

This pull request is testing a batch with the changes from pull requests 138568, 139168, 138353, and 138441 - batching documentation.

Pull request 138353 is stacked on pull request 138343, whose changes are included here and will be merged with it.

Dependencies

This pull request depends on the changes from pull requests 136063, 139202, 139178, and 139169.

cohenroy and others added 30 commits November 15, 2026 02:57
- With no service picked and more than 40 calls, the map draws the 40 busiest, with a note and a Show all toggle.
- A caller with more than 5 external hosts gets one grouped "N external hosts" node (summed requests and errors, p95 shown as an upper bound). Clicking it opens the caller with every host.
- The map fits again whenever its drawn nodes or edges change, not only on first load.
- The tables still list every call.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Generated-By: PostHog Desktop
Task-Id: e1eccb5d-e35f-4749-baa8-6de8c1b6b5d1
With dashboard-v2, a refresh on the dashboard page opens one stream_query_results stream with cached_first. Its first cached event is the batched cached read (one MGET through _read_cached_tiles): every hit paints at once, stale hits included and marked, then the stream computes the missing and stale tiles. A failed stream, or a tile the organization had no capacity for, falls back to per-tile requests.

The client code lives in dashboardV2Logic. dashboardLogic keeps one branch at the refresh site. Streams count admitted and rejected in posthog_dashboard_query_stream_admissions_total, client events carry stream_fallback, and a board stream waits at most 10 seconds for query capacity. InsightSerializer.effective_query is the one extraction of the effective query.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…shboard-v2 module

Moves stale marking and the choice of tiles to compute into products/dashboards/backend/v2/board_stream.cached_event. stream_query_results keeps one call when cached_first is set.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
refreshThroughBoardStream in dashboardV2Logic now decides whether the stream applies, ties its abort to the refresh, reports delivered tiles and returns the tiles left for the per-tile path with the event properties for them. dashboardLogic keeps one call in refreshDashboardItems and passes an optional eventProperties through to its tile events, so it holds no stream code.

Also defers the dashboard view's import of v2/board_stream for product:lint.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…back

A tile the board stream leaves for the per-tile path already waited for query capacity on the server. The fallback now asks getInsightWithRetry for one attempt, so it does not add its capacity retries and backoff on top. Other per-tile loads keep the default retries.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…rts a tile

reportDashboardTileRefreshed now takes the tile's load timing before the
filter change. Stream tiles skip the per-tile load queue and have no request
of their own, so they report no queue wait and no request time.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2 updated
Run: 59bb3f7b-3010-401c-bffa-e7724d7629d5

Co-authored-by: DanielVisca <16750763+DanielVisca@users.noreply.github.com>
# Conflicts:
#	frontend/snapshots.yml
#	products/apm/frontend/DependenciesTab.stories.tsx
#	products/apm/frontend/DependenciesTab.tsx
#	products/apm/frontend/dependenciesLogic.ts
The button that opened an existing implementation run read "View task", which looked like it starts a new run. It now reads "View running implementation" (or "View implementation" once the run ended), and its tooltip names who started the run and says it does not start a new one. "Take over" becomes "Take over claim" in the detail actions and the report context menu.

Generated-By: PostHog Desktop
Task-Id: 7a69aba7-37a8-4a7a-93c4-ffe2d2eb2f1f
A fallback tile that hits a capacity error now ends with that error instead
of submitting force_async and polling while it holds a tile queue slot. The
board stream already waited for capacity before falling back.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Move the root project to posthog-python 8, which sends only to capture
v1. The LLM gateway is its own uv project and deploy, so the SDK major
guard holds it on 7 until its own migration.

posthog-python 8 retries a failed sync_mode send in the calling thread,
where 7 made one attempt. The production sync clients keep one attempt.

Co-authored-by: Anna Garcia <anna.g@posthog.com>
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
posthoganalytics 8 holds 1,000 AI events per client by default, down from
10,000. Each assistant run has its own client, so a long run could drop
events while capture is slow.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Bounds what one run can buffer while capture is slow. posthoganalytics 7 held 10,000 in this queue, and the longest run seen in 30 days sent under 2,000 events.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
posthoganalytics 7 retried a failed connection inside one send; 8 only
retries through `max_retries`. One retry on a short timeout covers that
and holds the calling thread for less time than one 15 second try.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Reconnecting a GitLab project that the team already has raised IntegrationConflict and returned a 500. The creator now upserts like the other integration kinds, so a reconnect replaces the token and config and clears errors. The serializer still requires admin access to overwrite an existing integration.

Generated-By: PostHog Desktop
Task-Id: bebfb380-ab1b-4195-b2b8-d71c5415256e
Move the gateway to posthog-python 8, which sends only to capture v1,
and drop the SDK major guard's hold on it.

posthog-python 8 retries a failed sync_mode send in the calling thread
and gives AI events a 30 second timeout. The gateway keeps one attempt
per send and a 15 second timeout, as posthog-python 7 behaved.

litellm.proxy imports backoff, which only posthog-python 7 installed, so
the gateway now declares it.

Co-authored-by: Anna Garcia <anna.g@posthog.com>
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The setting does not change what the gateway sends on posthoganalytics 8.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
eli-r-ph and others added 17 commits November 19, 2026 14:30
posthoganalytics 7 retried a failed connection inside one send; 8 only
retries through `max_retries`.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Generated-By: PostHog Desktop
Task-Id: c94efae8-0817-4184-a55c-279a120e6a4d
The test built its rows from the real clock, so its +02:00 case failed
on every run between 22:00 and 24:00 UTC.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Members get a create-only write, so a project connected while their request waits on GitLab raises instead of being overwritten. Admin reconnects must name the same host and project id, and keep the original creator.

Generated-By: PostHog Desktop
Task-Id: 32fa078b-562e-4a71-a430-229072150885
Only the scheme and host fold case, because GitLab can run under a case-sensitive path prefix. The identity check is now a boolean helper instead of a tuple.

Generated-By: PostHog Desktop
Task-Id: 32fa078b-562e-4a71-a430-229072150885
@trunk-io trunk-io Bot closed this Nov 20, 2026
@trunk-io
trunk-io Bot deleted the trunk-merge/pr-138568/fe36c99d-168f-494e-bcd3-1bcebb8cbd52 branch November 20, 2026 00:28
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants