Great!
Yes, behaviour exact as i described. So looks like it's root cause )
Thank you, Sam. Ilya!
2015-08-21 21:08 GMT+03:00 Samuel Just sj...@redhat.com:
I think I found the bug -- need to whiteout the snapset (or decache
it) upon evict.
http://tracker.ceph.com/issues/12748
-Sam
On Fri,
On Fri, Aug 21, 2015 at 5:59 PM, Samuel Just sj...@redhat.com wrote:
Odd, did you happen to capture osd logs?
No, but the reproducer is trivial to cut paste.
Thanks,
Ilya
___
ceph-users mailing list
ceph-users@lists.ceph.com
Odd, did you happen to capture osd logs?
-Sam
On Thu, Aug 20, 2015 at 8:10 PM, Ilya Dryomov idryo...@gmail.com wrote:
On Fri, Aug 21, 2015 at 2:02 AM, Samuel Just sj...@redhat.com wrote:
What's supposed to happen is that the client transparently directs all
requests to the cache pool rather
I think I found the bug -- need to whiteout the snapset (or decache
it) upon evict.
http://tracker.ceph.com/issues/12748
-Sam
On Fri, Aug 21, 2015 at 8:04 AM, Ilya Dryomov idryo...@gmail.com wrote:
On Fri, Aug 21, 2015 at 5:59 PM, Samuel Just sj...@redhat.com wrote:
Odd, did you happen to
Exact as in our case.
Ilya, same for images from our side. Headers opened from hot tier
пятница, 21 августа 2015 г. пользователь Ilya Dryomov написал:
On Fri, Aug 21, 2015 at 2:02 AM, Samuel Just sj...@redhat.com
javascript:; wrote:
What's supposed to happen is that the client transparently
On Fri, Aug 21, 2015 at 2:02 AM, Samuel Just sj...@redhat.com wrote:
What's supposed to happen is that the client transparently directs all
requests to the cache pool rather than the cold pool when there is a
cache pool. If the kernel is sending requests to the cold pool,
that's probably
Snapshotting with cache/tiering *is* supposed to work. Can you open a bug?
-Sam
On Thu, Aug 20, 2015 at 3:36 PM, Andrija Panic andrija.pa...@gmail.com wrote:
This was related to the caching layer, which doesnt support snapshooting per
docs...for sake of closing the thread.
On 17 August 2015
Good joke )
2015-08-21 2:06 GMT+03:00 Samuel Just sj...@redhat.com:
Certainly, don't reproduce this with a cluster you care about :).
-Sam
On Thu, Aug 20, 2015 at 4:02 PM, Samuel Just sj...@redhat.com wrote:
What's supposed to happen is that the client transparently directs all
We switch to forward mode as step to switch cache layer off.
Right now we have samsung 850 pro in cache layer (10 ssd, 2 per nodes)
and they show 2MB for 4K blocks... 250 IOPS... intead of 18-20K for intel
S3500 240G which we choose as replacement..
So with such good disks - cache layer - very
Created a ticket to improve our testing here -- this appears to be a hole.
http://tracker.ceph.com/issues/12742
-Sam
On Thu, Aug 20, 2015 at 4:09 PM, Samuel Just sj...@redhat.com wrote:
So you started draining the cache pool before you saw either the
inconsistent pgs or the anomalous snap
Right. But issues started...
2015-08-21 2:20 GMT+03:00 Samuel Just sj...@redhat.com:
But that was still in writeback mode, right?
-Sam
On Thu, Aug 20, 2015 at 4:18 PM, Voloshanenko Igor
igor.voloshane...@gmail.com wrote:
WE haven't set values for max_bytes / max_objects.. and all data
Yeah, I'm trying to confirm that the issues did happen in writeback mode.
-Sam
On Thu, Aug 20, 2015 at 4:21 PM, Voloshanenko Igor
igor.voloshane...@gmail.com wrote:
Right. But issues started...
2015-08-21 2:20 GMT+03:00 Samuel Just sj...@redhat.com:
But that was still in writeback mode,
As i we use journal collocation for journal now (because we want to utilize
cache layer ((( ) i use ceph-disk to create new OSD (changed journal size
on ceph.conf). I don;t prefer manual work))
So create very simple script to update journal size
2015-08-21 2:25 GMT+03:00 Voloshanenko Igor
Will do, Sam!
thank in advance for you help!
2015-08-21 2:28 GMT+03:00 Samuel Just sj...@redhat.com:
Ok, create a ticket with a timeline and all of this information, I'll
try to look into it more tomorrow.
-Sam
On Thu, Aug 20, 2015 at 4:25 PM, Voloshanenko Igor
igor.voloshane...@gmail.com
WE haven't set values for max_bytes / max_objects.. and all data initially
writes only to cache layer and not flushed at all to cold layer.
Then we received notification from monitoring that we collect about 750GB
in hot pool ) So i changed values for max_object_bytes to be 0,9 of disk
size...
But that was still in writeback mode, right?
-Sam
On Thu, Aug 20, 2015 at 4:18 PM, Voloshanenko Igor
igor.voloshane...@gmail.com wrote:
WE haven't set values for max_bytes / max_objects.. and all data initially
writes only to cache layer and not flushed at all to cold layer.
Then we received
I already kill cache layer, but will try to reproduce on lab
2015-08-21 1:58 GMT+03:00 Samuel Just sj...@redhat.com:
Hmm, that might actually be client side. Can you attempt to reproduce
with rbd-fuse (different client side implementation from the kernel)?
-Sam
On Thu, Aug 20, 2015 at 3:56
So you started draining the cache pool before you saw either the
inconsistent pgs or the anomalous snap behavior? (That is, writeback
mode was working correctly?)
-Sam
On Thu, Aug 20, 2015 at 4:07 PM, Voloshanenko Igor
igor.voloshane...@gmail.com wrote:
Good joke )
2015-08-21 2:06
Specifically, the snap behavior (we already know that the pgs went
inconsistent while the pool was in writeback mode, right?).
-Sam
On Thu, Aug 20, 2015 at 4:22 PM, Samuel Just sj...@redhat.com wrote:
Yeah, I'm trying to confirm that the issues did happen in writeback mode.
-Sam
On Thu, Aug
I mean in forward mode - it;s permanent problem - snapshots not working.
And for writeback mode after we change max_bytes/object values, it;s around
30 by 70... 70% of time it;s works... 30% - not. Looks like for old images
- snapshots works fine (images which already exists before we change
What's supposed to happen is that the client transparently directs all
requests to the cache pool rather than the cold pool when there is a
cache pool. If the kernel is sending requests to the cold pool,
that's probably where the bug is. Odd. It could also be a bug
specific 'forward' mode
Our initial values for journal sizes was enough, but flush time was 5 secs,
so we increase journal side to fit flush timeframe min|max for 29/30
seconds.
I mean
filestore max sync interval = 30
filestore min sync interval = 29
when said flush time
2015-08-21 2:16 GMT+03:00 Samuel Just
It would help greatly if, on a disposable cluster, you could reproduce
the snapshot problem with
debug osd = 20
debug filestore = 20
debug ms = 1
on all of the osds and attach the logs to the bug report. That should
make it easier to work out what is going on.
-Sam
On Thu, Aug 20, 2015 at 4:40
Also, can you include the kernel version?
-Sam
On Thu, Aug 20, 2015 at 3:51 PM, Samuel Just sj...@redhat.com wrote:
Snapshotting with cache/tiering *is* supposed to work. Can you open a bug?
-Sam
On Thu, Aug 20, 2015 at 3:36 PM, Andrija Panic andrija.pa...@gmail.com
wrote:
This was
Yes, will do.
What we see. When cache tier in forward mod, if i did
rbd snap create - it's use rbd_header not from cold tier, but from
hot-tier, butm this 2 headers not synced
And can;t be evicted from hot-storage, as it;s locked by KVM (Qemu). If i
kill lock, evict header - all start to work..
Certainly, don't reproduce this with a cluster you care about :).
-Sam
On Thu, Aug 20, 2015 at 4:02 PM, Samuel Just sj...@redhat.com wrote:
What's supposed to happen is that the client transparently directs all
requests to the cache pool rather than the cold pool when there is a
cache pool.
Also, what do you mean by change journal side?
-Sam
On Thu, Aug 20, 2015 at 4:15 PM, Samuel Just sj...@redhat.com wrote:
Not sure what you mean by:
but it's stop to work in same moment, when cache layer fulfilled with
data and evict/flush started...
-Sam
On Thu, Aug 20, 2015 at 4:11 PM,
Not sure what you mean by:
but it's stop to work in same moment, when cache layer fulfilled with
data and evict/flush started...
-Sam
On Thu, Aug 20, 2015 at 4:11 PM, Voloshanenko Igor
igor.voloshane...@gmail.com wrote:
No, when we start draining cache - bad pgs was in place...
We have big
Ok, create a ticket with a timeline and all of this information, I'll
try to look into it more tomorrow.
-Sam
On Thu, Aug 20, 2015 at 4:25 PM, Voloshanenko Igor
igor.voloshane...@gmail.com wrote:
Exactly
пятница, 21 августа 2015 г. пользователь Samuel Just написал:
And you adjusted the
Attachment blocked, so post as text...
root@zzz:~# cat update_osd.sh
#!/bin/bash
ID=$1
echo Process OSD# ${ID}
DEV=`mount | grep ceph-${ID} | cut -d -f 1`
echo OSD# ${ID} hosted on ${DEV::-1}
TYPE_RAW=`smartctl -a ${DEV} | grep Rota | cut -d -f 6`
if [ ${TYPE_RAW} == Solid ]
then
root@test:~# uname -a
Linux ix-s5 4.0.4-040004-generic #201505171336 SMP Sun May 17 17:37:22 UTC
2015 x86_64 x86_64 x86_64 GNU/Linux
2015-08-21 1:54 GMT+03:00 Samuel Just sj...@redhat.com:
Also, can you include the kernel version?
-Sam
On Thu, Aug 20, 2015 at 3:51 PM, Samuel Just
Right ( but also was rebalancing cycle 2 day before pgs corrupted)
2015-08-21 2:23 GMT+03:00 Samuel Just sj...@redhat.com:
Specifically, the snap behavior (we already know that the pgs went
inconsistent while the pool was in writeback mode, right?).
-Sam
On Thu, Aug 20, 2015 at 4:22 PM,
Exactly
пятница, 21 августа 2015 г. пользователь Samuel Just написал:
And you adjusted the journals by removing the osd, recreating it with
a larger journal, and reinserting it?
-Sam
On Thu, Aug 20, 2015 at 4:24 PM, Voloshanenko Igor
igor.voloshane...@gmail.com javascript:; wrote:
Right
And you adjusted the journals by removing the osd, recreating it with
a larger journal, and reinserting it?
-Sam
On Thu, Aug 20, 2015 at 4:24 PM, Voloshanenko Igor
igor.voloshane...@gmail.com wrote:
Right ( but also was rebalancing cycle 2 day before pgs corrupted)
2015-08-21 2:23 GMT+03:00
Hmm, that might actually be client side. Can you attempt to reproduce
with rbd-fuse (different client side implementation from the kernel)?
-Sam
On Thu, Aug 20, 2015 at 3:56 PM, Voloshanenko Igor
igor.voloshane...@gmail.com wrote:
root@test:~# uname -a
Linux ix-s5 4.0.4-040004-generic
We used 4.x branch, as we have very good Samsung 850 pro in production,
and they don;t support ncq_trim...
And 4,x first branch which include exceptions for this in libsata.c.
sure we can backport this 1 line to 3.x branch, but we prefer no to go
deeper if packege for new kernel exist.
No, when we start draining cache - bad pgs was in place...
We have big rebalance (disk by disk - to change journal side on both
hot/cold layers).. All was Ok, but after 2 days - arrived scrub errors and
2 pgs inconsistent...
In writeback - yes, looks like snapshot works good. but it's stop to
This was related to the caching layer, which doesnt support snapshooting
per docs...for sake of closing the thread.
On 17 August 2015 at 21:15, Voloshanenko Igor igor.voloshane...@gmail.com
wrote:
Hi all, can you please help me with unexplained situation...
All snapshot inside ceph broken...
Hi all, can you please help me with unexplained situation...
All snapshot inside ceph broken...
So, as example, we have VM template, as rbd inside ceph.
We can map it and mount to check that all ok with it
root@test:~# rbd map cold-storage/0e23c701-401d-4465-b9b4-c02939d57bb5
/dev/rbd0
39 matches
Mail list logo