Thread: [HACKERS] Logical replication and synchronous replication
Hi, As doc of logical decoding said as a note[1], logical replication can support the synchronous replication with some restriction. But In addition to this, IIUC in logical replication decoded data is sent to subscribers only when the commit WAL record is decoded (calls ReorderBufferCommit) . It means that the local SQL execution and applying the decoded data on subscriber side are always executed in a sequential order, and the response time can simply be doubled or even more (OTOH a good point is that decoded data of aborted transaction is never sent to subscriber). I think there will be a lot of peoples who want to use logical synchronous replication but this is a big restriction for such user. I think we should document it or deal with it. Thought? [1] https://www.postgresql.org/docs/devel/static/logicaldecoding-synchronous.html Regards, -- Masahiko Sawada NIPPON TELEGRAPH AND TELEPHONE CORPORATION NTT Open Source Software Center
On 18 April 2017 at 18:55, Masahiko Sawada <sawada.mshk@gmail.com> wrote: > Hi, > > As doc of logical decoding said as a note[1], logical replication can > support the synchronous replication with some restriction. But In > addition to this, IIUC in logical replication decoded data is sent to > subscribers only when the commit WAL record is decoded (calls > ReorderBufferCommit) . Correct. > It means that the local SQL execution and > applying the decoded data on subscriber side are always executed in a > sequential order, and the response time can simply be doubled or even > more (OTOH a good point is that decoded data of aborted transaction is > never sent to subscriber). I think there will be a lot of peoples who > want to use logical synchronous replication but this is a big > restriction for such user. I think we should document it or deal with > it. > Thought? Definitely should be documented. I think it's covered under logical decoding, but needs at least an xref. -- Craig Ringer http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training & Services
On Tue, Apr 18, 2017 at 9:05 PM, Craig Ringer <craig@2ndquadrant.com> wrote: > On 18 April 2017 at 18:55, Masahiko Sawada <sawada.mshk@gmail.com> wrote: >> Hi, >> >> As doc of logical decoding said as a note[1], logical replication can >> support the synchronous replication with some restriction. But In >> addition to this, IIUC in logical replication decoded data is sent to >> subscribers only when the commit WAL record is decoded (calls >> ReorderBufferCommit) . > > Correct. > >> It means that the local SQL execution and >> applying the decoded data on subscriber side are always executed in a >> sequential order, and the response time can simply be doubled or even >> more (OTOH a good point is that decoded data of aborted transaction is >> never sent to subscriber). I think there will be a lot of peoples who >> want to use logical synchronous replication but this is a big >> restriction for such user. I think we should document it or deal with >> it. >> Thought? > > Definitely should be documented. I think it's covered under logical > decoding, but needs at least an xref. Yes, I think so too. I'll send a patch for that this week, and maybe will propose a improvement patch for logical sync replication in the next release cycle. Regards, -- Masahiko Sawada NIPPON TELEGRAPH AND TELEPHONE CORPORATION NTT Open Source Software Center