If a node crashes after committing and index build, but before that write makes it into a checkpoint, startup recovery will require a full index rebuild before the node is available again. This can be catastrophic if a node crashes immediately after committing an index build.

We should evaluate a targeted, backportable solution for avoiding this worst-case recovery scenario. For example, could we persist resume info after finishing the index build and before voting for commit?

One caveat is that this may still be subject to the limitations of resumable index builds:

• Index builds can only be resumed once
• Index builds with non-default commitQuorum cannot be resumed