Обсуждение: very high CPU usage in "top", but not in "mpstat"
Hello all,
We are using Postgres 7.3 with JBoss 3.2.3 on a Linux Fedora 1.0 box.
When I am looking at CPU activity with “top”, I often see something like:
PID USER PRI NI SIZE RSS SHARE STAT %CPU %MEM TIME CPU COMMAND
14154 postgres 25 0 3592 3592 2924 R 99.1 0.3 93:53 0 postmaster
At the same time, “mpstat” gives me something like:
11:27:21 AM CPU %user %nice %system %idle intr/s
11:27:21 AM all 2.99 0.00 18.94 78.08 105.94
The system is not visibly slow and response time is satisfactory. Sometimes, the CPU usage drops to 1 or 2%, but not for long usually. Also I have checked the number of open connections to Postgres and there are only 5 (maximum is set to the default: 32).
Should I be worried that Postgres is eating up 99% of my CPU??? Or is this *expected* behaviour?
Please note that I am a developer, not a system administrator, so maybe I misunderstand the usage of “top” here.
Any help will be appreciated.
Cyrille.
"Cyrille Bonnet" <cyrille@3months.com> writes:
> Should I be worried that Postgres is eating up 99% of my CPU??? Or is this
> *expected* behaviour?
It's not expected, unless you are running some very long-running query.
The conflict between what top says and what mpstat says is strange; I
wonder if you might have a buggy version of one of them? You should
probably check some other tools (try "vmstat 1" for instance) to see if
you can get a consensus on whether the CPU is maxed out or not ...
regards, tom lane
I'm guessing you have a 4 cpu box: 1 99 percent busy process on a 4 way box == about 25% busy overall. On May 5, 2004, at 6:03 AM, Tom Lane wrote: > "Cyrille Bonnet" <cyrille@3months.com> writes: >> Should I be worried that Postgres is eating up 99% of my CPU??? Or is >> this >> *expected* behaviour? > > It's not expected, unless you are running some very long-running query. > > The conflict between what top says and what mpstat says is strange; I > wonder if you might have a buggy version of one of them? You should > probably check some other tools (try "vmstat 1" for instance) to see if > you can get a consensus on whether the CPU is maxed out or not ... > > regards, tom lane > > ---------------------------(end of > broadcast)--------------------------- > TIP 8: explain analyze is your friend >