Ganesh Vernekar

Do you mean to say drop both checkpoint and also the WAL? (btw, in that case, we should also remove the data from the memory)

Hold on before working on this. https://github.com/prometheus/prometheus/issues/7530#issuecomment-655393626 makes sense and need to re-think this. If checkpoint is corrupt, replaying WAL for partial series might not be useful/meaningful.

Hold on before working on this. https://github.com/prometheus/prometheus/issues/7530#issuecomment-655393626 makes sense and need to re-think this. If checkpoint is corrupt, replaying WAL for partial series might not be useful/meaningful.

In ~Prometheus Storage Working Group~ contributors office hours we wen't over these 3 options of how we want to handle this 1. Drop everything from Head since there will be...

Then that sounds like option 1. Also, checkpoint corruptions should be rare, it will only happy if disk is faulty and not on Prometheus crashes.

To not block this any further, let's go with the simpler option 1, and we can revisit this later if that is causing any problems.

We could. But for some reason, I prefer `ref` outside. Reason: `ref` is used widely, almost everywhere. `hint` I would assume to be not used in many places, so it...

Alternate idea: release a 2.55 with the only diff from 2.54 being the code to support rollback from 2.55. 2.55 release could add rollback code for other features as well...

I have a wild unconventional idea: keep the index version as V2 but don't write these fields anymore, and keep the TOC structure intact. The 2.x version from last couple...

So, [2.15.0 / 2019-12-23](https://github.com/prometheus/prometheus/releases/tag/v2.15.0) is the release after which we stopped reading these fields (~5 years ago). So I feel keeping the index version as V2 and stop writing it...