Resolve locks could not based on GC safepoint
Open
sig/transaction
type/enhancement
- Dominant language
- Go
- Stars
- 40.5k
- Forks
- 6.2k
- PR merge metrics
- PR metrics pending
Description
## Enhancement
In some scenarios, the user wants to set the GC lifetime longer to keep they can flashback to any point when some mistakes happen. But they may not want to resolve lock too late.
When TiDB stopped, there may lots of running transactions, many locks need to be resolved by GC or some other transactions after TiDB restarted. This will cause some latency spike for the new transaction. It will be OK after the first round GC, but if the GC life time is too large, the spike may happened in a very long time.
It may be better to set the resolve lock time earlier in GC, which will not make the latency spike take too long time.
Contributor guide
Assessment
This issue has not been assessed yet.