Thread ETags, stagger sweep bursts, and attribute budget consumption - #32
Merged
Conversation
Three companion reductions from the 2026-08-04 rate-limit incident, where check-runs, PR metadata, and PR files were the largest 403 categories after CODEOWNERS and a single C-R1 sweep enqueued ~1,300 refreshes at once: Conditional requests: the REST refresh paths for PR metadata, files, and check runs now send persisted ETags as If-None-Match. A 304 is a confirmed observation only when its validator still belongs to the current endpoint representation -- it resets that endpoint's freshness clock without rewriting entity state, and a validator compare-and-set rejects a late stale response arriving after a newer write, keeping C-C2 intact under interleaving (tested). 304s flow through C-B4 accounting with correct zero-cost budget header handling. Burst smoothing: sweep refresh deadlines get bounded jitter applied at enqueue (River unique-key coalescing preserved), clamped so the latest enqueue plus completion headroom stays inside the C-R1 staleness bound (0.95B < B); hard deadlines remain last_checked_at + B and the existing bound tests pass unchanged. Attribution: the budget gate exports a per-request counter labeled by auth context, endpoint family, and outcome (200/304/403/error) -- bounded cardinality, no raw paths -- built on the gate's public RequestHook seam so it composes with #25's per-context budget restructure. Collector shape is test-pinned and the PromQL is documented in ops/DASHBOARD.md, turning the next incident's attribution from hours of 403 sampling into one query. Fixes #28 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
ewhauser
force-pushed
the
refresh-etags-attribution
branch
from
August 4, 2026 21:33
7647511 to
dbe1864
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
1 participant
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.
Fixes #28 — the three companion reductions to #25/#26/#27 from the rate-limit incident.
If-None-Match; 304s reset endpoint freshness without rewriting state, with a validator CAS rejecting late-stale responses (C-C2 interleaving test). C-B4-observable, zero-cost budget accounting verified.0.95B < B); hard deadlines and existing bound tests unchanged; River unique-key coalescing preserved.RequestHookseam — deliberately composed against Separate rate-budget accounting for App-JWT and installation-token requests #25's restructure; collector-test-pinned, documented inops/DASHBOARD.md.Built and adversarially reviewed by paired Codex sol-xhigh agents in a dedicated worktree (parallel with #25/#26/#27). Full gate green locally.
🤖 Generated with Claude Code