Re: Feature request (or at least discussion): enable autovaccum on temp tables

2019-05-31 Thread Michael Lewis
> > So, a related question, since we have dozens of temp tables and a lot of > code, is there a way to look up what temp tables are being created by the > current session, so I can do a VACUUM or ANALYZE on all of them in bulk? I > know I can inspect pg_temp_* schema, but how to figure out which on

Re: Feature request (or at least discussion): enable autovaccum on temp tables

2019-05-31 Thread Christopher Browne
On Fri, May 31, 2019, 12:26 PM Ivan Voras wrote: > Hi, > On Fri, 31 May 2019 at 17:43, Andrew Gierth > wrote: > >> > "Ivan" == Ivan Voras writes: >> >> Ivan> Since AFAIK temp tables are very close to unlogged ordinary >> Ivan> tables, what would stop this feature from being implemented? >

Re: Feature request (or at least discussion): enable autovaccum on temp tables

2019-05-31 Thread Ivan Voras
Hi, On Fri, 31 May 2019 at 17:43, Andrew Gierth wrote: > > "Ivan" == Ivan Voras writes: > > Ivan> Since AFAIK temp tables are very close to unlogged ordinary > Ivan> tables, what would stop this feature from being implemented? > > The key difference between temp tables and other tables is

Re: Feature request (or at least discussion): enable autovaccum on temp tables

2019-05-31 Thread Ivan Voras
hi, On Fri, 31 May 2019 at 17:54, Christopher Browne wrote: > 2. Temp tables are only associated (and visible) in the session in which > you are doing the work. autovacuum operates inside an ordinary session > context, and in a separate connected session, so it can't see your temp > tables; t

Re: Feature request (or at least discussion): enable autovaccum on temp tables

2019-05-31 Thread Christopher Browne
On Fri, 31 May 2019 at 11:26, Ivan Voras wrote: > Hello, > > The reason why we are using temp tables is to allow concurrent runs on > some very large reports which involve creating and heavily churning dozens > of very large tables. > > The problem we're facing is that if we DON'T use temp tables

Re: Feature request (or at least discussion): enable autovaccum on temp tables

2019-05-31 Thread Andrew Gierth
> "Ivan" == Ivan Voras writes: Ivan> Since AFAIK temp tables are very close to unlogged ordinary Ivan> tables, what would stop this feature from being implemented? The key difference between temp tables and other tables is that temp table data does not use the shared_buffers but is buffere

Re: Feature request (or at least discussion): enable autovaccum on temp tables

2019-05-31 Thread Ivan Voras
On Fri, 31 May 2019 at 17:38, Adam Brusselback wrote: > Temp tables are not visibile outside of a single connection, so the > autovacuum worker connection isn't able to see it. > I can connect as a superuser via psql while temp tables are being used and I can examine tables in the pg_temp_* sche

Re: Feature request (or at least discussion): enable autovaccum on temp tables

2019-05-31 Thread Ivan Voras
On Fri, 31 May 2019 at 17:34, Pavel Stehule wrote: > Hi > > pá 31. 5. 2019 v 17:26 odesílatel Ivan Voras napsal: > >> Hello, >> >> >> autovacuum check tables once per minute, and working on closed > transactions. Lot of times temporary tables are filled inside some batch, > and immediately some

Re: Feature request (or at least discussion): enable autovaccum on temp tables

2019-05-31 Thread Adam Brusselback
Temp tables are not visibile outside of a single connection, so the autovacuum worker connection isn't able to see it. Are you sure that it's actually an issue with accumulating dead tuples, and not an issue with bad statistics? In my processes which are heavy on temp tables, I have to manually r

Re: Feature request (or at least discussion): enable autovaccum on temp tables

2019-05-31 Thread Pavel Stehule
Hi pá 31. 5. 2019 v 17:26 odesílatel Ivan Voras napsal: > Hello, > > The reason why we are using temp tables is to allow concurrent runs on > some very large reports which involve creating and heavily churning dozens > of very large tables. > > The problem we're facing is that if we DON'T use te