Not a resolution, but an idea that you've probably thought of. Disabling logging on any affected OSDs (possibly just all of them) seems like a needed step to be able to keep working with this cluster to finish the upgrade and get it healthier.
On Wed, Oct 10, 2018 at 6:37 PM Wido den Hollander <wido@xxxxxxxx> wrote:
On 10/11/2018 12:08 AM, Wido den Hollander wrote:
> Hi,
>
> On a Luminous cluster running a mix of 12.2.4, 12.2.5 and 12.2.8 I'm
> seeing OSDs writing heavily to their logfiles spitting out these lines:
>
>
> 2018-10-10 21:52:04.019037 7f90c2f0f700 0 stupidalloc 0x0x55828ae047d0
> dump 0x15cd2078000~34000
> 2018-10-10 21:52:04.019038 7f90c2f0f700 0 stupidalloc 0x0x55828ae047d0
> dump 0x15cd22cc000~24000
> 2018-10-10 21:52:04.019038 7f90c2f0f700 0 stupidalloc 0x0x55828ae047d0
> dump 0x15cd2300000~20000
> 2018-10-10 21:52:04.019039 7f90c2f0f700 0 stupidalloc 0x0x55828ae047d0
> dump 0x15cd2324000~24000
> 2018-10-10 21:52:04.019040 7f90c2f0f700 0 stupidalloc 0x0x55828ae047d0
> dump 0x15cd26c0000~24000
> 2018-10-10 21:52:04.019041 7f90c2f0f700 0 stupidalloc 0x0x55828ae047d0
> dump 0x15cd2704000~30000
>
> It goes so fast that the OS-disk in this case can't keep up and become
> 100% util.
>
> This causes the OSD to slow down and cause slow requests and starts to flap.
>
> It seems that this is *only* happening on OSDs which are the fullest
> (~85%) on this cluster and they have about ~400 PGs each (Yes, I know,
> that's high).
>
After some searching I stumbled upon this Bugzilla report:
https://bugzilla.redhat.com/show_bug.cgi?id=1600138
That seems to be the same issue, although I'm not 100% sure.
Wido
> Looking at StupidAllocator.cc I see this piece of code:
>
> void StupidAllocator::dump()
> {
> std::lock_guard<std::mutex> l(lock);
> for (unsigned bin = 0; bin < free.size(); ++bin) {
> ldout(cct, 0) << __func__ << " free bin " << bin << ": "
> << free[bin].num_intervals() << " extents" << dendl;
> for (auto p = free[bin].begin();
> p != free[bin].end();
> ++p) {
> ldout(cct, 0) << __func__ << " 0x" << std::hex << p.get_start()
> << "~"
> << p.get_len() << std::dec << dendl;
> }
> }
> }
>
> I'm just wondering why it would spit out these lines and what's causing it.
>
> Has anybody seen this before?
>
> Wido
> _______________________________________________
> ceph-users mailing list
> ceph-users@xxxxxxxxxxxxxx
> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
>
_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
_______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
