Thanks, I make a summary of the issue, no connection storm(fork) either, just suddenly many session waiting on "extend" and "DataFileRead", it last 2 seconds, this server has 64 vcpu and running there long time without issue, only last weekend, we patch from 14.8 to 14.14. We checked with Infra and VMWARE, and OS monitoring, except very high %sys cpu, not found other abnormal events. so, we are thinking if Postgresql buffer cache or OS page cache level contention leading to that(if that can contribute to IO wait, DataFileRead, extend).
attached please find details, just masking dbname, username,and ip addresses.
On Thu, 2025-06-26 at 10:32 +0800, James Pang wrote: > thans for you suggestions, we have iowait from sar command too, copy here, checking with infra team not found abnormal IO activities either. > 02:00:01 PM CPU %usr %nice %sys %iowait %irq %soft %steal %guest %gnice %idle > 02:00:03 PM all 15.92 0.00 43.02 0.65 0.76 2.56 0.00 0.00 0.00 37.09
Crazy high system time.
If it is not transparent hugepages, the other suspect is fork(). Turn on "log_connections" for a minute or two and see if there are lots of connections established. That can easily use your CPU.
Other than that, I am out of guesses.
> Frits Hoogland <frits.hoogland@gmail.com> 於 2025年6月25日週三 下午10:27寫道： > > > On 25 Jun 2025, at 07:59, Laurenz Albe <laurenz.albe@cybertec.at> wrote: > > > > > > Run "sar -P all 1" and see if "%iowait" is high. > > > > I would (strongly) advise against the use of iowait as an indicator.
I have heard that before, and I am sure you are right.
I grant that if it is low, it may just mean that the CPU is under load. But if %iowait is high, my experience is that that indicates an I/O problem.
