On 2021-Aug-17, Bossart, Nathan wrote: > On 8/17/21, 2:13 PM, "alvhe...@alvh.no-ip.org" <alvhe...@alvh.no-ip.org> > wrote: > > > So, why isn't it that we call Register in XLogInsertRecord, and > > Notify in XLogWrite? > > We do. However, we also call NotifySegmentsReadyForArchive() in > XLogInsertRecord() to handle the probably-unlikely scenario that the > flush pointer has already advanced past the to-be-registered boundary. > This ensures that the .ready files are created as soon as possible.
I see. I have two thoughts on that. First, why not do it outside the block that tests for crossing a segment boundary? If that's a good thing to do, then we should do it always. However, why do it in a WAL-producing client-connected backend? It strikes me as a bad thing to do, because you are possibly causing delays for client-connected backends. I suggest that we should give this task to the WAL writer process -- say, have XLogBackgroundFlush do it. -- Álvaro Herrera Valdivia, Chile — https://www.EnterpriseDB.com/ "All rings of power are equal, But some rings of power are more equal than others." (George Orwell's The Lord of the Rings)
