Re: Re: BUG #14680: startup process on standby encounter a deadlock of TwoPhaseStateLock when redo 2PC xlog

On Tue, Jun 13, 2017 at 7:49 AM, Alvaro Herrera
<alvherre(at)2ndquadrant(dot)com> wrote:
> However, I found out that this rationale is likely not true, because the
> checkpointer may be running concurrently with this code from startup
> process, and checkpointer does process 2PC data. Maybe there are other
> reasons why there's no live bug here, but it looks wrong (I didn't try
> to reproduce a problem).

The current coding is actually safe because the checkpointer does not
remove or add any 2PC entry in the array while holding
TwoPhaseStateLock, it just updates some values that need to be read
and/or written while holding the lock. Well, to be honest, HEAD is
wrong because it can read a flag value while the checkpointer updates
it, and the patch is careful to change that to be correct. The wrong
part is when calling ProcessTwoPhaseBuffer() in
RecoverPreparedTransactions() which accesses gxact->ondisk and
prepare_start_lsn without locking things.
--
Michael