metricbeat restart loop with error: data path already locked by another beat

[endorama]

• Version: 8.3.2
• Operating System: (Running in Docker with elastic-package)
• Steps to Reproduce:
  1. create policy integration (gcp package in my case, compute and loadbalancing_metrics data streams)
  2. add it to an agent
  3. change the policy
  4. observe the behaviour in Agent logs

Reporting as issue after internal discussion.

I noticed this error in Agent logs:

metricbeat stderr: Exiting: data path already locked by another beat. Please make sure that multiple beats are not sharing the same data path (path.data).

This happened consistently every time I updated a policy assigned to an agent. After this error the agent is in degraded state and tries to restart metricbeat, but it fails and gets stuck in this loop.

This is particularly painful for testing integrations, as I have to restart the agent each time to have it pick up a change to a policy.

NOTE: I've not been able to reproduce it in further tests made to gather information for opening this issue, but it happened consistently in previous tests.

[michel-laterman]

I think I made an issue in the beats repo for this: elastic/beats#31670

[fearful-symmetry]

Yah, I assume this is an extension of the lockfile issues under elastic-agent, currently addressed in elastic/beats#33169

[cmacknz]

Closing as duplicate of elastic/beats#31670 unless we find a reason to treat this separately.