The problem we are experiencing is described here:

https://bugzilla.redhat.com/show_bug.cgi?id=1497332

However, we are running 12.2.2.

Across our 6 ceph clusters, this one with the problem was first version
12.2.0, then upgraded to .1 and then to .2.

The other 5 ceph installations started as version 12.2.1 and then updated
to .2.

Karun Josy

On Mon, Jan 29, 2018 at 7:01 PM, Karun Josy <[email protected]> wrote:

> Thank you for your response.
>
> We don't think there is an issue with the cluster being behind snap
> trimming. We just don't think snaptrim is occurring at all.
>
> We have 6 individual ceph clusters. When we delete old snapshots for
> clients, we can see space being made available. In this particular one
> however, with 300 virtual machines, 28TBs of data (this is our largest
> ceph), I can delete hundreds of snapshots, and not a single gigabyte
> becomes available after doing that.
>
> In our other 5, smaller Ceph clusters, we can see hundreds of gigabytes
> becoming available again after doing massive deletions of snapshots.
>
> The Luminous gui also never shows "snaptrimming" occurring in the EC
> pool.  While the other 5 Luminous clusters, their GUI will show
> snaptrimming occurring for the EC pool. Within minutes we can see the
> additional space becoming available.
>
> This isn't an issue of the trimming queue behind schedule. The system
> shows there is no trimming scheduled in the queue, ever.
>
> However, when using ceph du on particular virtual machines, we can see
> that snapshots we delete are indeed no longer listed in ceph du's output.
>
> So, they seem to be deleting. But the space is not being reclaimed.
>
> All clusters are same hardware. Some have more disks and servers than
> others. The only major difference is that this particular Ceph with this
> problem, it had the noscrub and nodeep-scrub flags set for many weeks.
>
>
> Karun Josy
>
> On Mon, Jan 29, 2018 at 6:27 PM, David Turner <[email protected]>
> wrote:
>
>> I don't know why you keep asking the same question about snap trimming.
>> You haven't shown any evidence that your cluster is behind on that. Have
>> you looked into fstrim inside of your VMs?
>>
>> On Mon, Jan 29, 2018, 4:30 AM Karun Josy <[email protected]> wrote:
>>
>>> fast-diff map is not enabled for RBD images.
>>> Can it be a reason for Trimming not happening ?
>>>
>>> Karun Josy
>>>
>>> On Sat, Jan 27, 2018 at 10:19 PM, Karun Josy <[email protected]>
>>> wrote:
>>>
>>>> Hi David,
>>>>
>>>> Thank you for your reply! I really appreciate it.
>>>>
>>>> The images are in pool id 55. It is an erasure coded pool.
>>>>
>>>> ---------------
>>>> $ echo $(( $(ceph pg  55.58 query | grep snap_trimq | cut -d[ -f2 | cut
>>>> -d] -f1 | tr ',' '\n' | wc -l) - 1 ))
>>>> 0
>>>> $ echo $(( $(ceph pg  55.a query | grep snap_trimq | cut -d[ -f2 | cut
>>>> -d] -f1 | tr ',' '\n' | wc -l) - 1 ))
>>>> 0
>>>> $ echo $(( $(ceph pg  55.65 query | grep snap_trimq | cut -d[ -f2 | cut
>>>> -d] -f1 | tr ',' '\n' | wc -l) - 1 ))
>>>> 0
>>>> --------------
>>>>
>>>> Current snap_trim_sleep value is default.
>>>> "osd_snap_trim_sleep": "0.000000". I assume it means there is no delay.
>>>> (Can't find any documentation related to it)
>>>> Will changing its value initiate snaptrimming, like
>>>> ceph tell osd.* injectargs '--osd_snap_trim_sleep 0.05'
>>>>
>>>> Also, we are using an rbd user with the below profile. It is used while
>>>> deleting snapshots
>>>> -------
>>>>         caps: [mon] profile rbd
>>>>         caps: [osd] profile rbd pool=ecpool, profile rbd pool=vm,
>>>> profile rbd-read-only pool=templates
>>>> -------
>>>>
>>>> Can it be a reason ?
>>>>
>>>> Also, can you let me know which all logs to check while deleting
>>>> snapshots to see if it is snaptrimming ?
>>>> I am sorry I feel like pestering you too much.
>>>> But in mailing lists, I can see you have dealt with similar issues with
>>>> Snapshots
>>>> So I think you can help me figure this mess out.
>>>>
>>>>
>>>> Karun Josy
>>>>
>>>> On Sat, Jan 27, 2018 at 7:15 PM, David Turner <[email protected]>
>>>> wrote:
>>>>
>>>>> Prove* a positive
>>>>>
>>>>> On Sat, Jan 27, 2018, 8:45 AM David Turner <[email protected]>
>>>>> wrote:
>>>>>
>>>>>> Unless you have things in your snap_trimq, your problem isn't snap
>>>>>> trimming. That is currently how you can check snap trimming and you say
>>>>>> you're caught up.
>>>>>>
>>>>>> Are you certain that you are querying the correct pool for the images
>>>>>> you are snapshotting. You showed that you tested 4 different pools. You
>>>>>> should only need to check the pool with the images you are dealing with.
>>>>>>
>>>>>> You can inversely price a positive by changing your snap_trim
>>>>>> settings to not do any cleanup and see if the appropriate PGs have 
>>>>>> anything
>>>>>> in their q.
>>>>>>
>>>>>> On Sat, Jan 27, 2018, 12:06 AM Karun Josy <[email protected]>
>>>>>> wrote:
>>>>>>
>>>>>>> Is scrubbing and deep scrubbing necessary for Snaptrim operation to
>>>>>>> happen ?
>>>>>>>
>>>>>>> Karun Josy
>>>>>>>
>>>>>>> On Fri, Jan 26, 2018 at 9:29 PM, Karun Josy <[email protected]>
>>>>>>> wrote:
>>>>>>>
>>>>>>>> Thank you for your quick response!
>>>>>>>>
>>>>>>>> I used the command to fetch the snap_trimq from many pgs, however
>>>>>>>> it seems they don't have any in queue ?
>>>>>>>>
>>>>>>>> For eg :
>>>>>>>> ====================
>>>>>>>> $ echo $(( $(ceph pg  55.4a query | grep snap_trimq | cut -d[ -f2 |
>>>>>>>> cut -d] -f1 | tr ',' '\n' | wc -l) - 1 ))
>>>>>>>> 0
>>>>>>>> $ echo $(( $(ceph pg  55.5a query | grep snap_trimq | cut -d[ -f2 |
>>>>>>>> cut -d] -f1 | tr ',' '\n' | wc -l) - 1 ))
>>>>>>>> 0
>>>>>>>> $ echo $(( $(ceph pg  55.88 query | grep snap_trimq | cut -d[ -f2 |
>>>>>>>> cut -d] -f1 | tr ',' '\n' | wc -l) - 1 ))
>>>>>>>> 0
>>>>>>>> $ echo $(( $(ceph pg  55.55 query | grep snap_trimq | cut -d[ -f2 |
>>>>>>>> cut -d] -f1 | tr ',' '\n' | wc -l) - 1 ))
>>>>>>>> 0
>>>>>>>> $ echo $(( $(ceph pg  54.a query | grep snap_trimq | cut -d[ -f2 |
>>>>>>>> cut -d] -f1 | tr ',' '\n' | wc -l) - 1 ))
>>>>>>>> 0
>>>>>>>> $ echo $(( $(ceph pg  34.1d query | grep snap_trimq | cut -d[ -f2 |
>>>>>>>> cut -d] -f1 | tr ',' '\n' | wc -l) - 1 ))
>>>>>>>> 0
>>>>>>>> $ echo $(( $(ceph pg  1.3f query | grep snap_trimq | cut -d[ -f2 |
>>>>>>>> cut -d] -f1 | tr ',' '\n' | wc -l) - 1 ))
>>>>>>>> 0
>>>>>>>> =====================
>>>>>>>>
>>>>>>>>
>>>>>>>> While going through the PG query, I find that these PGs have no
>>>>>>>> value in purged_snaps section too.
>>>>>>>> For eg :
>>>>>>>> ceph pg  55.80 query
>>>>>>>> --
>>>>>>>> ---
>>>>>>>> ---
>>>>>>>>  {
>>>>>>>>             "peer": "83(3)",
>>>>>>>>             "pgid": "55.80s3",
>>>>>>>>             "last_update": "43360'15121927",
>>>>>>>>             "last_complete": "43345'15073146",
>>>>>>>>             "log_tail": "43335'15064480",
>>>>>>>>             "last_user_version": 15066124,
>>>>>>>>             "last_backfill": "MAX",
>>>>>>>>             "last_backfill_bitwise": 1,
>>>>>>>>             "purged_snaps": [],
>>>>>>>>             "history": {
>>>>>>>>                 "epoch_created": 5950,
>>>>>>>>                 "epoch_pool_created": 5950,
>>>>>>>>                 "last_epoch_started": 43339,
>>>>>>>>                 "last_interval_started": 43338,
>>>>>>>>                 "last_epoch_clean": 43340,
>>>>>>>>                 "last_interval_clean": 43338,
>>>>>>>>                 "last_epoch_split": 0,
>>>>>>>>                 "last_epoch_marked_full": 42032,
>>>>>>>>                 "same_up_since": 43338,
>>>>>>>>                 "same_interval_since": 43338,
>>>>>>>>                 "same_primary_since": 43276,
>>>>>>>>                 "last_scrub": "35299'13072533",
>>>>>>>>                 "last_scrub_stamp": "2018-01-18 14:01:19.557972",
>>>>>>>>                 "last_deep_scrub": "31372'12176860",
>>>>>>>>                 "last_deep_scrub_stamp": "2018-01-15
>>>>>>>> 12:21:17.025305",
>>>>>>>>                 "last_clean_scrub_stamp": "2018-01-18
>>>>>>>> 14:01:19.557972"
>>>>>>>>             },
>>>>>>>>
>>>>>>>> Not sure if it is related.
>>>>>>>>
>>>>>>>> The cluster is not open to any new clients. However we see a steady
>>>>>>>> growth of  space usage every day.
>>>>>>>> And worst case scenario, it might grow faster than we can add more
>>>>>>>> space, which will be dangerous.
>>>>>>>>
>>>>>>>> Any help is really appreciated.
>>>>>>>>
>>>>>>>> Karun Josy
>>>>>>>>
>>>>>>>> On Fri, Jan 26, 2018 at 8:23 PM, David Turner <
>>>>>>>> [email protected]> wrote:
>>>>>>>>
>>>>>>>>> "snap_trimq": "[]",
>>>>>>>>>
>>>>>>>>> That is exactly what you're looking for to see how many objects a
>>>>>>>>> PG still had that need to be cleaned up. I think something like this 
>>>>>>>>> should
>>>>>>>>> give you the number of objects in the snap_trimq for a PG.
>>>>>>>>>
>>>>>>>>> echo $(( $(ceph pg $pg query | grep snap_trimq | cut -d[ -f2 | cut
>>>>>>>>> -d] -f1 | tr ',' '\n' | wc -l) - 1 ))
>>>>>>>>>
>>>>>>>>> Note, I'm not at a computer and topping this from my phone so it's
>>>>>>>>> not pretty and I know of a few ways to do that better, but that 
>>>>>>>>> should work
>>>>>>>>> all the same.
>>>>>>>>>
>>>>>>>>> For your needs a visual inspection of several PGs should be
>>>>>>>>> sufficient to see if there is anything in the snap_trimq to begin 
>>>>>>>>> with.
>>>>>>>>>
>>>>>>>>> On Fri, Jan 26, 2018, 9:18 AM Karun Josy <[email protected]>
>>>>>>>>> wrote:
>>>>>>>>>
>>>>>>>>>>  Hi David,
>>>>>>>>>>
>>>>>>>>>> Thank you for the response. To be honest, I am afraid it is going
>>>>>>>>>> to be a issue in our cluster.
>>>>>>>>>> It seems snaptrim has not been going on for sometime now , maybe
>>>>>>>>>> because we were expanding the cluster adding nodes for the past few 
>>>>>>>>>> weeks.
>>>>>>>>>>
>>>>>>>>>> I would be really glad if you can guide me how to overcome this.
>>>>>>>>>> Cluster has about 30TB data and 11 million objects. With about
>>>>>>>>>> 100 disks spread across 16 nodes. Version is 12.2.2
>>>>>>>>>> Searching through the mailing lists I can see many cases where
>>>>>>>>>> the performance were affected while snaptrimming.
>>>>>>>>>>
>>>>>>>>>> Can you help me figure out these :
>>>>>>>>>>
>>>>>>>>>> - How to find snaptrim queue of a PG.
>>>>>>>>>> - Can snaptrim be started just on 1 PG
>>>>>>>>>> - How can I make sure cluster IO performance is not affected ?
>>>>>>>>>> I read about osd_snap_trim_sleep , how can it be changed ?
>>>>>>>>>> Is this the command : ceph tell osd.* injectargs
>>>>>>>>>> '--osd_snap_trim_sleep 0.005'
>>>>>>>>>>
>>>>>>>>>> If yes what is the recommended value that we can use ?
>>>>>>>>>>
>>>>>>>>>> Also, what all parameters should we be concerned about? I would
>>>>>>>>>> really appreciate any suggestions.
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Below is a brief extract of a PG queried
>>>>>>>>>> ----------------------------
>>>>>>>>>> ceph pg  55.77 query
>>>>>>>>>> {
>>>>>>>>>>     "state": "active+clean",
>>>>>>>>>>     "snap_trimq": "[]",
>>>>>>>>>> ---
>>>>>>>>>> ----
>>>>>>>>>>
>>>>>>>>>> "pgid": "55.77s7",
>>>>>>>>>>             "last_update": "43353'17222404",
>>>>>>>>>>             "last_complete": "42773'16814984",
>>>>>>>>>>             "log_tail": "42763'16812644",
>>>>>>>>>>             "last_user_version": 16814144,
>>>>>>>>>>             "last_backfill": "MAX",
>>>>>>>>>>             "last_backfill_bitwise": 1,
>>>>>>>>>>             "purged_snaps": [],
>>>>>>>>>>             "history": {
>>>>>>>>>>                 "epoch_created": 5950,
>>>>>>>>>> ---
>>>>>>>>>> ---
>>>>>>>>>> ---
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Karun Josy
>>>>>>>>>>
>>>>>>>>>> On Fri, Jan 26, 2018 at 6:36 PM, David Turner <
>>>>>>>>>> [email protected]> wrote:
>>>>>>>>>>
>>>>>>>>>>> You may find the information in this ML thread useful.
>>>>>>>>>>> https://www.spinics.net/lists/ceph-users/msg41279.html
>>>>>>>>>>>
>>>>>>>>>>> It talks about a couple ways to track your snaptrim queue.
>>>>>>>>>>>
>>>>>>>>>>> On Fri, Jan 26, 2018 at 2:09 AM Karun Josy <[email protected]>
>>>>>>>>>>> wrote:
>>>>>>>>>>>
>>>>>>>>>>>> Hi,
>>>>>>>>>>>>
>>>>>>>>>>>> We have set no scrub , no deep scrub flag on a ceph cluster.
>>>>>>>>>>>> When we are deleting snapshots we are not seeing any change in
>>>>>>>>>>>> usage space.
>>>>>>>>>>>>
>>>>>>>>>>>> I understand that Ceph OSDs delete data asynchronously, so
>>>>>>>>>>>> deleting a snapshot doesn’t free up the disk space immediately. 
>>>>>>>>>>>> But we are
>>>>>>>>>>>> not seeing any change for sometime.
>>>>>>>>>>>>
>>>>>>>>>>>> What can be possible reason ? Any suggestions would be really
>>>>>>>>>>>> helpful as the cluster size seems to be growing each day even 
>>>>>>>>>>>> though
>>>>>>>>>>>> snapshots are deleted.
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> Karun
>>>>>>>>>>>> _______________________________________________
>>>>>>>>>>>> ceph-users mailing list
>>>>>>>>>>>> [email protected]
>>>>>>>>>>>> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>
>>>
>
_______________________________________________
ceph-users mailing list
[email protected]
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

Reply via email to