Re: how to debug the postgres performance issue

[Ron <ronljohnsonjr@xxxxxxxxx>]

On 09/14/2018 04:34 AM, 范国腾 wrote:

Hi,

 

We are doing the performance test in two nodes system(active/standby). Now we find that the disk IO is very high but the CPU is low.

On both nodes or just one (and if so, which one)?

If you incrementally increase the workload, starting from the quiescent state, when do you see this happening?

Is there any method or tool to help us debug to find which function makes this issue?

 

We try to use the gprof but it does not help us. The result show the LWLockAssign（the shared buffer init） is called the most time. The other function is called no more that 33139. That seems no problem.

 

Could you please  give any suggestion how to debug the high IO/low CPU issue?

 

Thanks

Steven

 

===================================

no time accumulated

 

  %   cumulative   self              self     total          

 time   seconds   seconds    calls  Ts/call  Ts/call  name   

  0.00      0.00     0.00  9240928     0.00     0.00  LWLockAssign

  0.00      0.00     0.00    33139     0.00     0.00  ShmemAlloc

  0.00      0.00     0.00    33084     0.00     0.00  seg_alloc

….

--
Angular momentum makes the world go 'round.