feat(correctness): fan out from one millstone process#1724
Open
webern wants to merge 6 commits into
Open
Conversation
Millstone now holds multiple sinks per process and writes the same payload bytes to all configured targets. Eliminates per-payload divergence between Agent (baseline) and Agent+ADP (comparison) by construction. - millstone: `targets:` named map replaces single `target:`; in-process fan-out via TargetSender; fail-fast with named-target errors. - panoramic: single millstone process per test for both Docker and k8s paths; shared helpers for $GROUP placeholder resolution and socket enumeration. - 19 `test/correctness/*/millstone.yaml` migrated to `targets:` shape. - 4 new millstone unit tests + 6 new panoramic unit tests.
dd-octo-sts
Bot
added
the
area/test
This comment has been minimized.
This comment has been minimized.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 4c353efc6f
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Binary Size Analysis (Agent Data Plane)Baseline: f907c91 · Comparison: 5e9ddc9 · diff ✅ Binary size difference within thresholdChanges by Module
Detailed Symbol Changes |
Regression Detector (Agent Data Plane)Run ID: Optimization Goals: ✅ No significant changes detectedFine details of change detection per experiment (35)Experiments configured
Bounds Checks: ✅ Passed (5)
ExplanationA change is flagged as a regression when |Δ mean %| > 5.00% in the regressing direction for its optimization goal AND SMP marks the experiment as a regression ( |
webern
commented
May 22, 2026
webern
changed the title
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.
Summary
Reduce payload timing divergence by running a single millstone process that fans out to both
baselineandcomparisonin correctness tests.Before this change, the millstone container ran
sh -c '... & P1=$! ... & P2=$! ...'to fork twoindependent millstones - one per target which is one place where drift may be introduced.
In follow up PRs I intend to extend the solution to include
Key changes:
bin/correctness/millstone:Configacceptstargets:as a named map (baseline:/comparison:) instead of a singletarget:;TargetSenderfans out each generated payload toall configured sinks; driver and corpus collapsed to a single send loop; errors include the
target name.
bin/correctness/panoramic: Docker (runner.rs) and k8s (k8s.rs) paths each spawn onemillstone container/pod per test instead of two. New shared helpers in
correctness/config.rs(
resolve_group_placeholders,millstone_targets_all_sockets,millstone_first_network_port)substitute the
$GROUPplaceholder per-target on the host. The resolved YAML is written underthe per-test
log_dir(deliberately notmounts_dir, which is overlaid into the agentcontainers).
test/correctness/*/millstone.yamlmigrated mechanically to thetargets:shape.Change Type
How did you test this PR
Locally on macOS against rebuilt
correctness-toolsanddatadog-agentimages at this commit:-d test/integration/cases -d test/correctness) parallel (-p 4default): 51/52pass. Single failure:
dsd-origin-detection-matrix/unified-high-cardinality— window-edgedivergence, the residual window boundary failure class that subsequent PRs intend to target.
-p 1): 52/52 pass. This is slow! Parallelization is a good idea if we can fix it.Compare to the pre-fanout parallel baseline of 50/52 with all four
dsd-origin-detection-matrix/*variants plus
dsd-service-checksfailing under value-divergence.No leaked airlock resources after either run. I did see leaked airlock resources on an aborted run, and that is something I also intend to work on downstream.
References
Not directly, but documenting my ongoing sensitivity to local flakes:
dsd-mapper-blocklistfails locally. #1579