'More than one write index' errors still occurring in version 7.13.4

This issue has previously been raised here, but to summarise our current issue:

We are currently seeing our APM stack (Elastic APM, Elasticsearch, Kibana) produce the following error:

alias [apm-7.13.4-metric] has more than one write index [apm-7.13.4-metric-000001,apm-7.13.4-meric-000008]

The error is triggered on our ES cluster whenever our APM pods are restarted any time after the initial index (apm-7.13.4-metric-000001) has been removed via ILM.

It should be noted that the error doesn't occur when either of the following conditions are met:

  • setup.ilm.overwrite is set to false
  • The index in question (apm-7.13.4-metric-000001) is re-created and attached to the alias with "is_write_index": false

The fix appears to have been backported to 7.13.X in this PR, and we are on 7.13.4, yet we are still running into the issue.

Is anyone still running into this problem on version >= 7.13.X or have any ideas what could still be causing this?

Happy to provide more context if relevant.