You can give it a try only on that partition just to see if your query plan gets better. I prefer defining partitioning over ranging attributes like, for example: cid between 123 and 456. It makes more sense, especially when there are attributes which value strictly depends on the check attribute. Btw, dozens of millions is not a problem on modern systems. I remember of reading about a recommended 20 millions per partition but I usually work with 60 millions per partition without any problem.
Do you autovacuum? How frequently do the updates and insert operations occur?
Give us your configuration about work_mem, shared_buffers, max_connections etc. Kernel version? If possible avoid 3.2 and 3.8-3.13. Also think to upgrade your OS version.
From today I'm on vacancy, so others could help :)
Pietro Pugni
Il 12/ago/2015 03:49, "Rural Hunter" <
ruralhunter@gmail.com> ha scritto:
article_729 has about 0.8 million rows. The rows of the children tables are variance from several thousands to dozens of millions. How can it help to create index on the partition key?