Hello folks,
We need helps to understand how an method work and how we can optimize this.

Cause: We have a Go monorepo with ~20k tags, so lots of refs and for each "patchset" from gerrit, zuul take ~3min to discover jobs to execute

After debug, the method incrimined is _saveRepoState in merger.py because the method loop over ~20k "refs/tags" and the call to `refs.object` is particularly "slow" with degradation after each iterate in the loop
(0.002s on first call, 0.01 on the last call )

Is necessary to save the sha1 of all tags ? 
How can we enhance the performance of this method? 

Thx all