BDR, wal sender, high system cpu, mutex_lock_common

[milist ujang <ujang.milist@xxxxxxxxx>]

Hi all,

I've an environment 9.4 + bdr: 

PostgreSQL 9.4.4 on x86_64-unknown-linux-gnu, compiled by gcc (Debian 4.7.2-5) 4.7.2, 64-bit

kernel version:

3.2.0-4-amd64 #1 SMP Debian 3.2.65-1 x86_64 GNU/Linux

This is consolidation databases, in this machine there are around 250+ wal sender processes.

top output revealed high system cpu:

%Cpu(s):  1.4 us, 49.7 sy,  0.0 ni, 48.8 id,  0.0 wa,  0.0 hi,  0.0 si,  0.0 st

profiling cpu with perf:

perf top -e cpu-clock

Events: 142K cpu-clock

 82.37%  [kernel]            [k] __mutex_lock_common.isra.5

  4.49%  [kernel]            [k] do_raw_spin_lock

  2.23%  [kernel]            [k] mutex_lock

  2.16%  [kernel]            [k] mutex_unlock

  2.12%  [kernel]            [k] arch_local_irq_restore

  1.73%  postgres            [.] ValidXLogRecord

  0.87%  [kernel]            [k] __mutex_unlock_slowpath

  0.78%  [kernel]            [k] arch_local_irq_enable

  0.63%  [kernel]            [k] sys_recvfrom

finally get which processes (wal senders) that are using mutexes:

perf top -e task-clock -p 55382

Events: 697  task-clock

 88.08%  [kernel]  [k] __mutex_lock_common.isra.5

  3.27%  [kernel]  [k] do_raw_spin_lock

  2.34%  [kernel]  [k] arch_local_irq_restore

  2.10%  postgres  [.] ValidXLogRecord

  1.87%  [kernel]  [k] mutex_unlock

  1.87%  [kernel]  [k] mutex_lock

  0.47%  [kernel]  [k] sys_recvfrom

I think bdr is only reading wal file (current state is we behind current wal lsn),

so why reading wal file needs mutex?

I wonder, is there kernel version has better handling mutexes? 

--

regards

ujang jaenudin | DBA Consultant (Freelancer)
http://ora62.wordpress.com
http://id.linkedin.com/pub/ujang-jaenudin/12/64/bab