Hiroyuki Yamada wrote:
> By the way, reading LogStandbySnapshot() and GetRunningTransactionLocks()
> raised following questions.
>
> * There is a window beween gathering lock information in GetRunningTransactionLocks()
> and writing WAL in LogAccessExclusiveLocks().
> * In current lock redo algorithm, locks are released when the transaction holding the lock
> are commited or aborted.
>
> ... then what happens if any transaction holding ACCESS EXCLUSIVE lock commits in the
> window ?
Hmm, when replying the locks record, we should be ignoring entries
belonging to already-committed transactions. But I don't actually see
that in the code. That needs to be fixed, I think.
> Similary,
>
> * There is a window beween writing COMMIT WAL in RecordTransactionCommit() and
> releasing locks in ResourceOwnerRelease()
>
> ... then what happens when GetRunningTransactionLocks() gathers ACCESS EXCLUSIVE
> locks whose holder has already written the COMMIT WAL ?
This is handled in standby in the same manner as the above case.
> Are there any chances of releasing locks which have no COMMIT WAL for releasing them ?
Yes. In normal operation, there should eventually be a commit or abort
WAL record for every transaction. But as a safety valve, when a
running-xacts record is replayed, we mark as aborted all transactions
that are older than the oldest still-running XID according to the
running-xacts record, and release their locks. See StandbyReleaseOldLocks().
-- Heikki Linnakangas EnterpriseDB http://www.enterprisedb.com