Skip to content

Failed catchup ranges are retried only after all other ranges finish, adding hours to runtime #409

Description

@sisuresh

Summary

In a parallel catchup run (HistoryPubnetParallelCatchupV2), a single ledger range appears to have been dispatched early, failed/stalled without completing, and then was not retried until every other range had finished — a free worker picked it up hours later and completed it. This added roughly 4h to the total mission runtime.

Suggestion

Improve retry handling so a failed/stalled range is re-queued and picked up promptly rather than waiting for all other ranges to finish — e.g. a visibility timeout / lease reaper on the in-progress set, or explicit failure detection that re-dispatches to any free worker. This would avoid a single straggler serializing at the tail end.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions