Re: [HACKERS] SERIALIZABLE on standby servers

On Thu, Nov 16, 2017 at 5:52 PM, Simon Riggs <simon(at)2ndquadrant(dot)com> wrote:
>> I haven't looked at this again yet but a nearby thread reminded me of
>> another problem with this which I wanted to restate explicitly here in
>> the context of this patch. Even without replication in the picture,
>> there is a race to reach ProcArrayEndTransaction() after
>> RecordTransactionCommit() runs, which means that the DO history
>> (normal primary server) and REDO history (recovery) don't always agree
>> on the order that transactions become visible. With this patch, this
>> kind of diverging DO and REDO could allow undetectable read only
>> serialization anomalies. I think that ProcArrayEndTransaction() and
>> RecordTransactionCommit() need to be made atomic in the simple case so
>> that DO and REDO agree.
>
> Not atomic, we just need to make ProcArrayEndTransaction() apply
> changes in the order of commits.
>
> I think that is more easily possible by reusing the
> SyncRepWaitForLSN() code, since that already orders things by LSN.
>
> So make all committers wait and then get WALwriter to wake people
> after ProcArrayEndTransaction() has been applied.

That doesn't solve any problem we have. Making committers wait after
ProcArrayEndTransaction() wouldn't keep the effects of those
transactions from being visible to other transactions in the system.
And that's precisely the issue: on the primary, transactions become
visible to other transactions in the order in which they perform
ProcArrayEndTransaction(), but on the standby, they become visible in
the order in which they RecordTransactionCommit(). That difference is
a source of serialization anomalies.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company