test(dissemination): investigate legacy partition convergence timeout
Failure
The existing dissemination cost-comparison workflow timed out while checking exact deployment-load state after a partition in its pinned original-runtime lane:
Orleans.Dissemination.IntegrationHarness.ScalingTests.Scaling_OriginalLegacy_DefaultOff_EnabledSupported
Timed out during exact production load state at every active process.
at ProcessCluster.Eventually(...) in ProcessCluster.cs:128
at ProcessCluster.PublishAndConverge() in ProcessCluster.cs:107The captured scenario is scale-4-partition-OriginalLegacy-0-3bbda7faea5a43d1ba4ca456fdaf31fd. The single scaling test failed after approximately 2m 48s.
Evidence
- PR: #10236, head
8d322362c83391f868b5b1e1fe879fe485d6efff. - Existing workflow: https://github.com/dotnet/orleans/actions/runs/34570001652.
- Failed job: https://github.com/dotnet/orleans/actions/runs/34570001652/job/103170572149 (
Bounded original, default-off and enabled costs). - Artifact:
dissemination-bounded-cost-evidence(artifact ID10187553682), containing the Cost TRX and scenario snapshots/logs. - The companion pinned-binary upgrade, rollback and recovery job passed: https://github.com/dotnet/orleans/actions/runs/34570001652/job/103169906991.
Scope
The failure occurred in OriginalLegacy, before the comparison reached the modified dissemination-enabled runtime path. The cleanup commit changes local dissemination attempt admission and its coupled tests/docs; the comparison harness is unchanged. The underlying reason for the legacy lane's convergence timeout remains undetermined.
Investigation
Use the saved per-process snapshots and logs to identify which production load value did not converge after the partition. Reproduce the pinned original-runtime scenario and distinguish a convergence defect from a harness synchronization or timing issue. Searches for the exact failure message, OriginalLegacy, and dissemination scaling found no existing matching issue or merged fix.
Source: dotnet/orleans