Yes, I am trying to do all the tests today including SSD Journal + All SSD
pools.
Thanks, I will have a try.
Best Regards
-- Ray
On Mon, Jul 6, 2015 at 7:54 AM, Somnath Roy wrote:
> Are you using SSD journal ?
>
> I would say try tuning with the following parameters first..
>
>
>
> jo
Are you using SSD journal ?
I would say try tuning with the following parameters first..
journal_max_write_entries
journal_max_write_bytes
journal_queue_max_ops
journal_queue_max_bytes
filestore_max_sync_interval
filestore_min_sync_interval
f
Roy,
I try to use Grafana to make a trend image of apply_latency of my osd 0,
and it seems grow all the time. So as my understanding, both of the
keys(avgcount, sum) is accumulation values. So the correct arithmetic
should be:
current apply latency = (avgcount current - avgcount previous)/(sum cur
Yes, similar values are reported in ‘ceph osd perf’..But, here the
commit_latency meaning the journal commit latency in admin socket perf dump..
Also, remember, these values (latencies) are not stable (that’s why you will be
seeing spiky write performance) and very difficult to correlate I guess
Roy,
This is really helpful. So as your description, for each ops/seconds, I can
use avgcount/sum. Is this the same value as "ceph osd perf"?
osd fs_commit_latency(ms) fs_apply_latency(ms)
023 85
1 22
2
The virtualized MON will be on a completely different platform and won't
use any of the ceph cluster resources (either compute or disk). I was
thinking of putting the 'master' MON (the one with the lowest IP) on a
dedicated server because I read the load is heavier on this one (lots of
logs in part
I would say this is a problem of ntfs mount.. I found another way to do this,
so that’s that.
Thanks for noticing.
Br,T
-Original Message-
From: Ilya Dryomov [mailto:idryo...@gmail.com]
Sent: 5. heinäkuuta 2015 20:45
To: Tuomas Juntunen
Cc: ceph-users
Subject: Re: [ceph-users] RBD mo
Hi Ray,
Here is the description of the different latencies under filestore perf
counters.
Journal_latency :
--
This is the latency of putting the ops in journal. Write is acknowledged after
that (well a bit after that, there is one context switch after this).
commitcycle_la
On Sun, Jul 5, 2015 at 7:57 PM, Tuomas Juntunen
wrote:
> Hi
>
> Is there any other kernel that would work? Anyone else had this kind of
> problem with rbd map?
Well, 4.1 is the latest and therefore the easiest to debug, assuming
this is a kernel client problem.
What is the output of /sys/kernel
Hi
Is there any other kernel that would work? Anyone else had this kind of problem
with rbd map?
Br, T
-Original Message-
From: Ilya Dryomov [mailto:idryo...@gmail.com]
Sent: 5. heinäkuuta 2015 19:42
To: Tuomas Juntunen
Cc: ceph-users
Subject: Re: [ceph-users] RBD mounted image on lin
On Sun, Jul 5, 2015 at 7:37 PM, Tuomas Juntunen
wrote:
> Couple of times the same and and the whole rbd mount is hanged
>
> can't df, or ls.
>
> umount -l and rbd unmap takes 10-20mins to get rid of it and then I can mount
> again and try the transfer
>
> I have 20TB of stuff that needs to be cop
Couple of times the same and and the whole rbd mount is hanged
can't df, or ls.
umount -l and rbd unmap takes 10-20mins to get rid of it and then I can mount
again and try the transfer
I have 20TB of stuff that needs to be copied to that partition.
Br, T
-Original Message-
From: Ilya
On Sun, Jul 5, 2015 at 6:58 PM, Tuomas Juntunen
wrote:
> Hi
>
> That's the only error what comes from this, there's nothing else.
Is it repeated?
Thanks,
Ilya
___
ceph-users mailing list
ceph-users@lists.ceph.com
http://lists.ceph.com/
On Jul 4, 2015, at 2:44 PM, Adrien Gillard
mailto:gillard.adr...@gmail.com>> wrote:
Lastly, regarding Cluster Throughput: EC seems to require a bit more CPU and
memory than straight replication, which begs the question of how much RAM and
CPU are you putting into the chassis? With proper amou
Hi
That's the only error what comes from this, there's nothing else.
Br,
T
-Original Message-
From: Ilya Dryomov [mailto:idryo...@gmail.com]
Sent: 5. heinäkuuta 2015 18:30
To: Tuomas Juntunen
Cc: ceph-users
Subject: Re: [ceph-users] RBD mounted image on linux server kernel error and
On Sun, Jul 5, 2015 at 6:18 PM, Tuomas Juntunen
wrote:
> Hi
>
>
>
> We are experiencing the following
>
>
>
> - Hammer 0.94.2
>
> - Ubuntu 14.04.1
>
> - Kernel 3.16.0-37-generic
>
> - 40TB NTFS disk mounted through RBD
>
>
>
>
>
> First 50GB goes fine, but then
Hi
We are experiencing the following
- Hammer 0.94.2
- Ubuntu 14.04.1
- Kernel 3.16.0-37-generic
- 40TB NTFS disk mounted through RBD
First 50GB goes fine, but then this happens
Jul 5 16:56:01 cephclient kernel: [110581.046141] kworker/u65:
Cephers,
Is there any documents or code definition to explain ceph perf dump? I am a
little confusing about the output, for example, under filestore, there's
journal_latency and apply_latency and each of them has avgcount and sum. I
am not quite sure what's the unit and meaning of the numbers? How
I'm quite clear now.
> this is my test setup - that's why I'm trying to break it and fix it
(best way to learn)
Thanks for your feedback!!
Kinjo
On Sun, Jul 5, 2015 at 8:51 PM, Jacek Jarosiewicz <
jjarosiew...@supermedia.pl> wrote:
> Well, the docs say that when your osds get full you should
Well, the docs say that when your osds get full you should add another
osd - and the cluster should redistribute data by it self:
http://ceph.com/docs/master/rados/troubleshooting/troubleshooting-osd/#no-free-drive-space
this is my test setup - that's why I'm trying to break it and fix it
(bes
That's good!
So was the root cause is because the osd was full? What's your thought
about that?
Was there any reason to delete any files?
Kinjo
On Sun, Jul 5, 2015 at 6:51 PM, Jacek Jarosiewicz <
jjarosiew...@supermedia.pl> wrote:
> ok, I got it working...
>
> first i manually deleted some fi
ok, I got it working...
first i manually deleted some files from the full osd, set the flag
noout and restarted the osd daemon.
then i waited a while for the cluster to backfill pgs, and after that
the rados -p cache cache-try-flush-evict-all command went OK.
I'm wondering though, because t
22 matches
Mail list logo
