Improved snapshot creation logic. #2318
Conversation
|
|
Codecov Report
@@ Coverage Diff @@
## master #2318 +/- ##
==========================================
- Coverage 86.46% 86.17% -0.30%
==========================================
Files 649 648 -1
Lines 64374 64770 +396
==========================================
+ Hits 55662 55815 +153
- Misses 8712 8955 +243
Continue to review full report at Codecov.
|
bright-starry-sky
changed the title
|
1, Solved problem of unstable test cases of snapshot_command_test. Problem is caused by a deadlock. createSnapshot(get read space_lock) --> updateHostInfo ( get write space_lock) -> cancelWriteBlocking end of createSnapshot (get read space_lock) . read_lock -> write_lock -> read_lock so deadlock occurred. Thanks @critical27. |
| if (fs::FileUtils::exist(checkpointPath)) { | ||
| LOG(ERROR) << "The snapshot file already exists: " << checkpointPath; | ||
| return ResultCode::ERR_CHECKPOINT_ERROR; | ||
| if (fs::FileUtils::exist(checkpointPath) && |
There was a problem hiding this comment.
Why not return success?
Because we don't have a way to verify the validity of the snapshot that has been created. For data security, I need to delete and recreate it at here.
| @@ -620,6 +620,13 @@ bool FileBasedWal::linkCurrentWAL(const char* newPath) { | |||
| LOG(INFO) << idStr_ << "No wal files found, skip link"; | |||
| return true; | |||
| } | |||
|
|
|||
| if (fs::FileUtils::exist(newPath) && | |||
| !fs::FileUtils::remove(newPath, true)) { | |||
| @@ -13,7 +13,6 @@ namespace nebula { | |||
| namespace meta { | |||
|
|
|||
| void CreateSnapshotProcessor::process(const cpp2::CreateSnapshotReq&) { | |||
| folly::SharedMutex::ReadHolder rHolder(LockUtils::spaceLock()); | |||
The dump_syms tool path should be match with third party version. Co-authored-by: George <[email protected]>
Unable to reproduce on development machine, added some diagnostic information for snapshot_command_test.
Please ignore this PR, I will close it when I find the failure reason.