Re: 2x Time difference between first and subsequent run of the same query on fresh established connection (on linux, with perf data included, all query data in the shared buffers) on postgresql 9.3.10. Any explanation? - Mailing list pgsql-general

Hi,

On 2016-02-25 13:50:11 +1100, Maxim Boguk wrote:
> The first run (something fishy with kernel calls):
>   19,60%  postgres  [kernel.kallsyms]  [k] filemap_map_pages
>   15,86%  postgres  postgres           [.] hash_search_with_hash_value
>    8,20%  postgres  postgres           [.] heap_hot_search_buffer
>    8,20%  postgres  postgres           [.] heap_page_prune_opt
>    5,72%  postgres  postgres           [.] PinBuffer
>    4,38%  postgres  [kernel.kallsyms]  [k] page_fault
>    4,04%  postgres  [kernel.kallsyms]  [k] page_waitqueue
>    3,55%  postgres  [kernel.kallsyms]  [k] __wake_up_bit
>    2,95%  postgres  postgres           [.] LWLockAcquire
>    2,31%  postgres  [kernel.kallsyms]  [k] unlock_page
>    1,96%  postgres  [vdso]             [.] __vdso_gettimeofday
>    1,83%  postgres  [kernel.kallsyms]  [k] radix_tree_next_chunk
>    1,77%  postgres  [kernel.kallsyms]  [k] page_add_file_rmap
>    1,66%  postgres  postgres           [.] _bt_checkkeys
>    1,27%  postgres  postgres           [.] LWLockRelease

Looks like page faults are part of the proble here. Upgrade to 9.4 and
use them unfortunately is the best suggestion I have.

Andres