| From: | Robert Haas <robertmhaas(at)gmail(dot)com> |
|---|---|
| To: | David Gould <daveg(at)sonic(dot)net> |
| Cc: | Heikki Linnakangas <hlinnakangas(at)vmware(dot)com>, Andres Freund <andres(at)2ndquadrant(dot)com>, pgsql-hackers(at)postgresql(dot)org, Anand Ranganathan <arangana(at)adobe(dot)com>, Alex Eulenberg <aeulenbe(at)adobe(dot)com>, Ashokraj M <ashokraj(at)adobe(dot)com>, Hari <hari(at)adobe(dot)com>, Elein Mustain <mustain(at)adobe(dot)com> |
| Subject: | Re: Re: bulk_multi_insert infinite loops with large rows and small fill factors |
| Date: | 2012-12-14 20:39:44 |
| Message-ID: | CA+TgmoY+zKLqBuWqV3Bej5XDtxFDE28CJhs34v9CYKfQNBM_mw@mail.gmail.com |
| Views: | Raw Message | Whole Thread | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-hackers |
On Wed, Dec 12, 2012 at 8:29 AM, David Gould <daveg(at)sonic(dot)net> wrote:
> We lose noticable performance when we raise fill-factor above 10. Even 20 is
> slower.
Whoa.
> During busy times these hosts sometimes fall into a stable state
> with very high cpu use mostly in s_lock() and LWLockAcquire() and I think
> PinBuffer plus very high system cpu in the scheduler (I don't have the perf
> trace in front of me so take this with a grain of salt). In this mode they
> fall from the normal 7000 queries per second to below 3000.
I have seen signs of something similar to this when running pgbench -S
tests at high concurrency. I've never been able to track down where
the problem is happening. My belief is that once a spinlock starts to
be contended, there's some kind of death spiral that can't be arrested
until the workload eases up. But I haven't had much luck identifying
exactly which spinlock is the problem or if it even is just one...
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Robert Haas | 2012-12-14 20:45:42 | Re: PageIsAllVisible()'s trustworthiness in Hot Standby |
| Previous Message | Robert Haas | 2012-12-14 20:33:28 | Re: Use gcc built-in atomic inc/dec in lock.c |