Bug Report: Online DDL scheduler gives up too early on VReplication error (for vitess migrations)

[shlomi-noach]

Overview of the Issue

Historically, VReplication's error reporting was not well formed. The Online DDL scheduler relies on _vreplication.state value and/or parsing the _vreplication.message column.

However, ongoing changes in VReplication introduce values which Online DDL interprets as terminal errors, thus failing the migration, where in fact these errors are recoverable, if only we let VReplication enough time to retry.

This is one of two main causes for flakiness in onlineddl_vrepl CI test.

Reproduction Steps

Binary Version

-

Operating System and Environment details

-

Log Fragments

No response