A full artifact storage quota must not fail a green suite (#74) - #76
Merged
Merged
Conversation
An organisation hit the Actions artifact storage quota today and every consumer's
CI went red at `Shard labels`:
Failed to CreateArtifact: Artifact storage quota has been hit.
before a single test executed. That job's real product is the `matrix` OUTPUT, and
the log of the failing run reads `121 timing rows.` and `Set output 'matrix'` —
the work succeeded and was discarded because an upload after it failed. Every
shard `needs:` that matrix, so the whole suite was skipped, on every PR, in every
repository using this workflow.
The uploads that report on a run are now `continue-on-error`. The split is this
file's own: coverage and timings are a REPORT, "every unit ran exactly once" is a
CORRECTNESS CLAIM, and only the second should be able to fail a suite. They were
sharing one exit status.
- `-timings` (label job) — the load-bearing one; it gated the matrix.
- `-coverage-<sid>`, `-out-<sid>` (shards) — both run AFTER the suite, so failing
there converts a green shard into a red one over a report.
- `-lcov`, `-records`, and the counter merge (collect) — same category.
`-prebuild` stays fatal: the shards DOWNLOAD it, so a failure there is a real
missing dependency rather than a lost report.
`Every unit ran, exactly once` stays fatal and ungated. So does the refusal inside
`restore_counters`: what changes is that a missing coverage report annotates the
run with `::error::` and skips the Codecov upload instead of failing the gate. The
silent-coverage-loss failure this file exists to prevent — 54.5 % reported for a
94.8 % suite — is still impossible, because a missing report is visible as an
error and an absent upload, never as a quietly wrong number.
Contributor
|
📚 Docs preview: https://codes.sota-shimozono.com/TestShards.jl/previews/PR76/ (updates on each push to this PR) |
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
sotashimozono
added a commit
that referenced
this pull request
Aug 19, 2026
…#77) #76 stopped a full artifact storage quota from failing the label job, and the shards then ran — and all eight failed one step later: Unable to download artifact(s): Artifact not found for name: testshards-timings An upload that is allowed to fail has a consumer that is not. Both halves have to degrade together, and for `timings` the degradation is NOT free to make per shard: > The ONE history `labels` read, not one this shard fetched for itself. A per-shard > fetch let a single transient failure put one shard on a different assignment > function from the rest — which double-runs some units and skips others, with > every shard still green. So `continue-on-error` on the DOWNLOAD would reintroduce exactly that. Instead the label job publishes its upload's outcome as an output, and every shard gates on that one value: either the whole matrix has the history or none of it does. "None" is a state the package already defines — `load_timings` returns an empty table for a missing path, which is what the first run of any suite gets — so the fallback is the ordinary unweighted assignment, not a new code path. A `::warning::` says the balance may be worse and that the result is not. `prebuild` gets the same shape for the same reason, and it is an optimisation outright: a shard without the shared compiled cache precompiles for itself and computes the same answer. Its upload becomes non-fatal, its outcome an output, and both the download and the adopt step gate on it. What stays fatal is unchanged: `Every unit ran, exactly once`, and the refusal inside `restore_counters`.
This was referenced Aug 19, 2026
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 of #74, and independent of the package change in #75 — this is workflow-only, so it can merge and take effect immediately without waiting for a registration.
What happened
lab-sotashimozonohit the Actions artifact storage quota today. Every consumer's CI went red atShard labels:before a single test executed.
That job's real product is the
matrixoutput, and the log of the failing run reads:The work succeeded, and was thrown away because an upload after it failed. Every shard
needs:that matrix, so the whole suite was skipped — on every PR, in every repository using this workflow. A storage problem became a test failure.The split this restores
It is this file's own vocabulary: coverage and timings are a report;
Every unit ran, exactly onceis a correctness claim. Only the second should be able to fail a suite. They were sharing one exit status.Now
continue-on-error:-timings-coverage-<sid>-out-<sid>-lcov,-recordsUnchanged and still fatal:
-prebuild— the shards download it, so a failure there is a genuine missing dependency, not a lost report.Every unit ran, exactly once— ungated, as before.restore_counters— just as loud. What changes is that a missing coverage report annotates the run with::error::and skips the Codecov upload, instead of failing theAll shards passedgate.The silent-coverage-loss failure this file exists to prevent — 54.5 % reported for a suite that covered 94.8 % — stays impossible: a missing report is visible as an error and an absent upload, never as a quietly wrong number.
Relationship to #75
#75 stops the cause (per-shard coverage artifacts carry 8 copies of the source tree; 61.8× reducible) but has to wait on registration, and it cannot help a quota that is already full. This PR makes the symptom survivable and applies the moment it merges.