Re: Understanding Postgres Memory Usage - Mailing list pgsql-general
| From | Theron Luhn |
|---|---|
| Subject | Re: Understanding Postgres Memory Usage |
| Date | |
| Msg-id | CAHYFdT-7UqVyjb2V+5p_vewyJGhpSy9+ofB3BseHKoLLPaXFQA@mail.gmail.com Whole thread Raw |
| In response to | Re: Understanding Postgres Memory Usage ("Ilya Kazakevich" <Ilya.Kazakevich@JetBrains.com>) |
| Responses |
Re: Understanding Postgres Memory Usage
(John R Pierce <pierce@hogranch.com>)
|
| List | pgsql-general |
$ free -h # Before the query
total used free shared buffers cached
Mem: 7.8G 5.2G 2.6G 212M 90M 4.9G
-/+ buffers/cache: 248M 7.6G
Swap: 0B 0B 0B
$ free -h # After the query
total used free shared buffers cached
Mem: 7.8G 5.3G 2.5G 212M 90M 4.9G
-/+ buffers/cache: 312M 7.5G
Swap: 0B 0B 0B
[I.K >> ] Are you talking about buffers/cache increased? AFAIK this memory is used by kernel as buffer before any block device (HDD for example).
Postgres does not use this memory directly, it simply reads data from block device, and kernel caches it. Process can’t be OOMed because of it.
I am sure you should configure your Postgres to NEVER exceed available RAM. You may use tools like (http://pgtune.leopard.in.ua/) or calculate it manually.
I do not remember exact formula, but it should be something like “work_mem*max_connections + shared_buffers” and it should be around 80% of your machine RAM (minus RAM used by other processes and kernel).
It will save you from OOM.
If you face performance bottleneck after it, you fix it using tools like “log_min_duration_statement”, “track_io_timing” and system-provided tools.
Ilya Kazakevich
JetBrains
The Drive to Develop
pgsql-general by date: