Flush log store meta when compacting log #633
Conversation
|
Codecov ReportAttention: Patch coverage is
❗ Your organization needs to install the Codecov GitHub app to enable full functionality. Additional details and impacted files@@ Coverage Diff @@
## master #633 +/- ##
==========================================
+ Coverage 56.51% 66.10% +9.58%
==========================================
Files 108 109 +1
Lines 10300 10986 +686
Branches 1402 1505 +103
==========================================
+ Hits 5821 7262 +1441
+ Misses 3894 3005 -889
- Partials 585 719 +134 ☔ View full report in Codecov by Sentry. |
| REPL_STORE_LOG(DEBUG, "Simulating log compaction with delay, delay:{}", delay.get()); | ||
| std::this_thread::sleep_for(std::chrono::milliseconds(delay.get())); | ||
| } | ||
| #endif | ||
| m_log_store->truncate(to_store_lsn(compact_lsn)); |
There was a problem hiding this comment.
IMO, we can change this line to
m_log_store->truncate(to_store_lsn(compact_lsn), false);
this will be enough.
home_raft_log_store will have one seperate logdev, and there will be only one logstore on this logdev.
this line will change in_memory_truncate_only to false and as result it will trigger logdev truncation, where the metadata of logstore will be updated
There was a problem hiding this comment.
Thanks, that make it even simpler!
- apply_snp_resync_data is similar to apply_snapshot in raft
yuwmao
changed the title
In baseline resync scenario, log truncation won't flush the truncate result immediately(periodically flushed by resource manager), thus if durable_commit_lsn has been flushed but start_lsn of the logstore hasn't been flushed, it will trigger the issue6.
Add a flush meta function in logstore, and manually trigger flush in log truncation.