Repository navigation
feat(observability): summarize runs and sessions - #397
Merged
Merged
Conversation
ShubyM
force-pushed
the
observability/agent-summaries
branch
from
October 30, 2026 18:24
a01ea1d to
699cf24
Compare
ShubyM
marked this pull request as ready for review
October 30, 2026 20:13
droot
approved these changes
Oct 30, 2026
Add three inspection routes. /api/v1/inspect/runs gives each run in one line from small records. /runs/<id>/summary gives the facts recorded about one run, and /sessions/<id>/summary the facts about one client session. A summary shows a block only when it has content. The API server now keeps each session's record and the SDK's error events through the telemetry buffer. Session models come from the model records, which bind_session already writes. Ages count from the request and never go negative. An ended run shows no ages. Goodput splits the trainer's idle time into GPU wait, queued, rollout and client. The ledger reports a lost request only after the gap has held across reads more than a flush apart. Log pointers use the /logs arguments run_id, pod, since and until.
ShubyM
force-pushed
the
observability/agent-summaries
branch
from
October 30, 2026 20:32
699cf24 to
fa56944
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why
Phase 8 of #390. Give agents one call that answers "why is this run not behaving".
What changed
/api/v1/inspect/runslists one line per run, with failure and SDK error counts./api/v1/inspect/runs/{run_id}/summaryreturns facts in blocks that appear only when they have content: pace, goodput, lost and unanswered requests with their holder, failures with a pointer to their logs, saves, non-finite metrics, samplers, config mismatches and placement./api/v1/inspect/sessions/{session_id}/summaryreturns a session's runs and its SDK errors.What you configure
Nothing.
Verification
make testandmake lintpass. Tests cover eight scenarios through the routes: a lost request, a request held by a dead worker, step pace, a config mismatch, a failure with its log pointer, a non-finite metric, sampler sends, and a session with SDK errors.save_stateheld by a dead trainer.