You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Summary:
Fixes issues uncovered by YbAdminSnapshotScheduleTest.UndeleteIndex test.
1) DeleteTableInMemory could be called multiple times in the case of the index table.
There is a check that just does noop when the table was already deleted.
Adjusted this check to do the same when the table is being hidden.
2) Don't remove the table from names map during delete, when it was previously hidden.
Otherwise, it would crash with fatal during cleanup.
3) DeleteTabletListAndSendRequests executes delete on tablet before commiting tablet info changes.
As a result tablet could be deleted before and callback called, before info changes in memory.
So table would hang in delete state. Because callback would think that tablet is not being deleted.
4) Decreased log flooding when compactions are being enabled in RocksDB.
When compactions are being enabled we call SetOptions twice for each RocksDB, and each of them dumps all current options values.
So while we have regular and intents DB we have 4 dumps of all rocksdb options.
Also added debug logging to `RWCLock::WriteLock()`, when it takes a too long time to acquire this lock, it would log the stack trace of the successful write lock.
Test Plan: ybd --gtest_filter YbAdminSnapshotScheduleTest.UndeleteIndex -n 20
Reviewers: bogdan
Reviewed By: bogdan
Subscribers: amitanand, ybase
Differential Revision: https://phabricator.dev.yugabyte.com/D11614
0 commit comments