Re: [PERFORM] Re: Planner chooses slow index heap scan despite accurate row estimates

2016-05-27 Thread Tom Lane
Jake Magner writes: > I tried without doing an INSERT at all, just running the SELECT queries and > the result is the same. Nested loop is chosen but is much slower. FWIW, I just noticed that the comparisons you're using are plain equality of the arrays. While a GIN array index supports that, it

[PERFORM] Re: Planner chooses slow index heap scan despite accurate row estimates

2016-05-27 Thread Jake Magner
I tried without doing an INSERT at all, just running the SELECT queries and the result is the same. Nested loop is chosen but is much slower. -- View this message in context: http://postgresql.nabble.com/Planner-chooses-slow-index-heap-scan-despite-accurate-row-estimates-tp5905357p5905383.html

Re: [PERFORM] Planner chooses slow index heap scan despite accurate row estimates

2016-05-27 Thread Tom Lane
Jake Magner writes: > I am trying to insert rows that don't already exist from a temp table into > another table. I am using a LEFT JOIN on all the columns and checking for > nulls in the base table to know which rows to insert. The problem is that > the planner is choosing a nested loop plan whic

[PERFORM] Planner chooses slow index heap scan despite accurate row estimates

2016-05-27 Thread Jake Magner
I am trying to insert rows that don't already exist from a temp table into another table. I am using a LEFT JOIN on all the columns and checking for nulls in the base table to know which rows to insert. The problem is that the planner is choosing a nested loop plan which is very slow over the much

Re: [PERFORM] Performance problems with 9.2.15

2016-05-27 Thread Steve Crawford
> > ...(BTW, I wonder why you are moving only to 9.2 and not something more >> recent.) >> > > Well, 9.2.15 is what comes bundled with RHEL 7, so I decided to go with > that to avoid dependency issues. But I could install a more fresh version > from scratch if that would solve my problem. > Genera

Re: [PERFORM] Performance problems with 9.2.15

2016-05-27 Thread Johan Fredriksson
The rowcount estimates from 9.2 seem greatly different from the 8.4 plan. Did you remember to ANALYZE all the tables after migrating? Maybe there were some table-specific statistics targets that you forgot to transfer over? No, I did not. Honestly I though everything would be transfered with a

Re: [PERFORM] Performance problems with 9.2.15

2016-05-27 Thread Tom Lane
Johan Fredriksson writes: > I am just about to upgrade from PostgreSQL 8.4.20 to 9.2.15, but I'v run > into some huge performance issues. The rowcount estimates from 9.2 seem greatly different from the 8.4 plan. Did you remember to ANALYZE all the tables after migrating? Maybe there were some ta

[PERFORM] Performance problems with 9.2.15

2016-05-27 Thread Johan Fredriksson
Hello! I am just about to upgrade from PostgreSQL 8.4.20 to 9.2.15, but I'v run into some huge performance issues. Both databases are configured the same way (shared_buffers = 2GB, temp_buffers = 32MB). I have increased work_mem on the 9.2 from 4MB to 64MB, but to no avail. Now, the query on 8.4: