Repository navigation
Conversation
- ci.yml: fmt/clippy/test, then the telemetry-functions acceptance smoke on real collector containers (stages share the runner's CPUs). - asap-sketch-benchmark.yml: run on vars.BENCH_RUNNER when set, compare summary.csv with the previous successful main nightly and fail on a drop beyond regression_margin (default 10%); fix stale verify counts for the five scenarios. - benchmark-on-demand.yml: `/benchmark` PR comment runs a reduced sweep on main and the PR head and posts the comparison. - benchmarks/compare-summaries.py and README tier docs.
On a shared 4-CPU CI runner a source's sketch sometimes reached window_merge after the scaled 2 s grace and was dropped as late, so one window's sketch held one source fewer than exact. Scaled runs now keep the deployed 5 s grace, and the check lists any window whose counts differ.
Data crossing a boundary already closes a window on time. On a contended host (the 4-CPU CI runner) the timer could close a window while its last batches were still queued, moving them into the next window: sketch windows then differed from exact by a batch (e.g. 388,736 vs 401,568 points). The timer now only closes idle streams' windows, 5 s after their end.
Loki keeps one entry per stream, timestamp and line, so a sketch and an exact event with the same text (identical top-k counts) collapsed into whichever arrived first; on the CI runner that dropped the sketch's novelty event. Each event line now starts with [sketch] or [exact]. The acceptance novelty check reports event counts and the asap_topk_new gauge when it fails.
zzylol
force-pushed
the
feat/e2e-acceptance
branch
from
October 13, 2026 00:20
9a62ef4 to
ff1cc27
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.
Part 9 of the RFC 0001 stack: the three CI benchmark tiers from the RFC.
ci.ymlacceptance.pyon real collector containers, with stages sharing the runner's 4 CPUs (a correctness check, not a capacity measurement);acceptance.jsonuploadedasap-sketch-benchmark.ymlvars.BENCH_RUNNER(falls back toubuntu-24.04with shared infrastructure); a new step compares against the previous night'ssummary.csvwithbenchmarks/compare-summaries.pyand fails on a throughput drop beyondregression_margin(default 10%)benchmark-on-demand.yml/benchmarkPR comment by an owner, member or collaboratorAlso fixes the nightly's verify step, which still expected the 3-scenario row counts (189/63/441) after the top-k scenarios made it 5 (315/105/735).
To enable dedicated-CPU runs: register a self-hosted runner with Docker and
gh, and set the repository variableBENCH_RUNNERto its label.Validated with actionlint 1.7.7 (no findings) and
yaml.safe_load.compare-summaries.pywas tested on synthetic CSVs (−15% fails at a 10% margin; −5%, +4% and −1.2% pass). This PR's own CI run is the first real run ofci.yml.Fixes found by this PR's first real CI runs
The shared 4-CPU runner exposed three problems that dedicated cores hid. All three were reproduced on a 56-core x86 server with every stage on CPUs 0–3 and fixed; the full acceptance then passed 25/25 under that contention.
window_mergeafter the scaled 2 s grace and was dropped as late. Acceptance now keeps the deployed 5 s grace, and the check lists any window whose counts differ.[sketch]or[exact].