Disable the slot compaction while the cluster slot is migrating

[git-hulk]

Search before asking

• I had searched in the issues and found no similar issues.

Motivation

Currently, we allow to run compaction even though the slot is migrating,
so it may cause data inconsistent in the corner case like below:

1. The cluster migration was started at T1 with the snapshot
2. The key 'foo' expired at T2, but it's extended between T1 and T2 to T3
3. Then the slot starts migrating the key 'foo' after T2, then it will be skipped since it's expired
4. Finally, the extended time write batch will be applied by iterating WAL items, but it will miss the subkeys since all of them are skipped.

This issue was found by @caipengbo, just record to prevent forgetting.

Solution

Don't compact the slot if it's migrating, and also migrate them if it's expired.

Are you willing to submit a PR?

• I'm willing to submit a PR!