ceph rbox test on passive compressed pool
I have been inserting 10790 exactly the same 64kb text message to a passive compressing enabled pool. I am still counting, but it looks like only half the objects are compressed. mail/b08c3218dbf1545ff430000052412a8e mtime 2020-09-06 16:27:39.000000, size 63580 mail/00f6043775f1545ff430000052412a8e mtime 2020-09-06 16:25:57.000000, size 525 mail/b875f40571f1545ff430000052412a8e mtime 2020-09-06 16:25:53.000000, size 63580 mail/e87c120b19f1545ff430000052412a8e mtime 2020-09-06 16:24:25.000000, size 525 I am not sure if this should be expected from passive, these docs[1] hint that passive 'compress if hinted COMPRESSIBLE'. From that I would conclude that all text messages should be compressed. A previous test with a 64kb gzip attachment seemed to not compress, although I did not look at all object sizes. on 14.2.11 [1] https://documentation.suse.com/ses/5.5/html/ses-all/ceph-pools.html#sec-ceph... https://docs.ceph.com/docs/mimic/rados/operations/pools/
The hints have to be given from the client side as far as I understand, can you share the client code too? Also,not seems that there's no guarantees that it will actually do anything (best effort I guess): https://docs.ceph.com/docs/mimic/rados/api/librados/#c.rados_set_alloc_hint Cheers On 6 September 2020 15:59:01 BST, Marc Roos <M.Roos@f1-outsourcing.eu> wrote:
I have been inserting 10790 exactly the same 64kb text message to a passive compressing enabled pool. I am still counting, but it looks like only half the objects are compressed.
mail/b08c3218dbf1545ff430000052412a8e mtime 2020-09-06 16:27:39.000000,
size 63580 mail/00f6043775f1545ff430000052412a8e mtime 2020-09-06 16:25:57.000000,
size 525 mail/b875f40571f1545ff430000052412a8e mtime 2020-09-06 16:25:53.000000,
size 63580 mail/e87c120b19f1545ff430000052412a8e mtime 2020-09-06 16:24:25.000000,
size 525
I am not sure if this should be expected from passive, these docs[1] hint that passive 'compress if hinted COMPRESSIBLE'. From that I would conclude that all text messages should be compressed. A previous test with a 64kb gzip attachment seemed to not compress, although I did not look at all object sizes.
on 14.2.11
[1] https://documentation.suse.com/ses/5.5/html/ses-all/ceph-pools.html#sec-ceph... https://docs.ceph.com/docs/mimic/rados/operations/pools/
_______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
-- Sent from my Android device with K-9 Mail. Please excuse my brevity.
Hi David, I suppose it is this part https://github.com/ceph-dovecot/dovecot-ceph-plugin/tree/master/src/storage-... -----Original Message----- To: ceph-users@ceph.io; Subject: Re: [ceph-users] ceph rbox test on passive compressed pool The hints have to be given from the client side as far as I understand, can you share the client code too? Also,not seems that there's no guarantees that it will actually do anything (best effort I guess): https://docs.ceph.com/docs/mimic/rados/api/librados/#c.rados_set_alloc_hint Cheers I have been inserting 10790 exactly the same 64kb text message to a passive compressing enabled pool. I am still counting, but it looks like only half the objects are compressed. mail/b08c3218dbf1545ff430000052412a8e mtime 2020-09-06 16:27:39.000000, size 63580 mail/00f6043775f1545ff430000052412a8e mtime 2020-09-06 16:25:57.000000, size 525 mail/b875f40571f1545ff430000052412a8e mtime 2020-09-06 16:25:53.000000, size 63580 mail/e87c120b19f1545ff430000052412a8e mtime 2020-09-06 16:24:25.000000, size 525 I am not sure if this should be expected from passive, these docs[1] hint that passive 'compress if hinted COMPRESSIBLE'. From that I would conclude that all text messages should be compressed. A previous test with a 64kb gzip attachment seemed to not compress, although I did not look at all object sizes. on 14.2.11 [1] https://documentation.suse.com/ses/5.5/html/ses-all/ceph-pools.html #sec-ceph-pool-compression https://docs.ceph.com/docs/mimic/rados/operations/pools/
Hi David, Just to let you know, this hint is being set, what is the reason for ceph of doing only half the objects? Can it be that there is some issue with my osd's? Like some maybe have an old fs (still using disk not volume)? Is this still to be expected or does ceph under pressure drop compressing? https://github.com/ceph-dovecot/dovecot-ceph-plugin/blob/56d6c900cc9ec07dfb9... Thanks, Marc -----Original Message----- Cc: jan.radon Subject: Re: [ceph-users] ceph rbox test on passive compressed pool The hints have to be given from the client side as far as I understand, can you share the client code too? Also,not seems that there's no guarantees that it will actually do anything (best effort I guess): https://docs.ceph.com/docs/mimic/rados/api/librados/#c.rados_set_alloc_hint Cheers On 6 September 2020 15:59:01 BST, Marc Roos <M.Roos@f1-outsourcing.eu> wrote: I have been inserting 10790 exactly the same 64kb text message to a passive compressing enabled pool. I am still counting, but it looks like only half the objects are compressed. mail/b08c3218dbf1545ff430000052412a8e mtime 2020-09-06 16:27:39.000000, size 63580 mail/00f6043775f1545ff430000052412a8e mtime 2020-09-06 16:25:57.000000, size 525 mail/b875f40571f1545ff430000052412a8e mtime 2020-09-06 16:25:53.000000, size 63580 mail/e87c120b19f1545ff430000052412a8e mtime 2020-09-06 16:24:25.000000, size 525 I am not sure if this should be expected from passive, these docs[1] hint that passive 'compress if hinted COMPRESSIBLE'. From that I would conclude that all text messages should be compressed. A previous test with a 64kb gzip attachment seemed to not compress, although I did not look at all object sizes. on 14.2.11 [1] https://documentation.suse.com/ses/5.5/html/ses-all/ceph-pools.html #sec-ceph-pool-compression https://docs.ceph.com/docs/mimic/rados/operations/pools/ ________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io -- Sent from my Android device with K-9 Mail. Please excuse my brevity.
On 09/11 09:36, Marc Roos wrote:
Hi David,
Just to let you know, this hint is being set, what is the reason for ceph of doing only half the objects? Can it be that there is some issue with my osd's? Like some maybe have an old fs (still using disk not volume)? Is this still to be expected or does ceph under pressure drop compressing?
https://github.com/ceph-dovecot/dovecot-ceph-plugin/blob/56d6c900cc9ec07dfb9...
I was trying to look into this a bit :), can you give me more info about the OSDs that you are using? What filesystem are they on? Cheers!
Thanks, Marc
-----Original Message----- Cc: jan.radon Subject: Re: [ceph-users] ceph rbox test on passive compressed pool
The hints have to be given from the client side as far as I understand, can you share the client code too?
Also,not seems that there's no guarantees that it will actually do anything (best effort I guess): https://docs.ceph.com/docs/mimic/rados/api/librados/#c.rados_set_alloc_hint
Cheers
On 6 September 2020 15:59:01 BST, Marc Roos <M.Roos@f1-outsourcing.eu> wrote:
I have been inserting 10790 exactly the same 64kb text message to a
passive compressing enabled pool. I am still counting, but it looks like only half the objects are compressed.
mail/b08c3218dbf1545ff430000052412a8e mtime 2020-09-06 16:27:39.000000, size 63580 mail/00f6043775f1545ff430000052412a8e mtime 2020-09-06 16:25:57.000000, size 525 mail/b875f40571f1545ff430000052412a8e mtime 2020-09-06 16:25:53.000000, size 63580 mail/e87c120b19f1545ff430000052412a8e mtime 2020-09-06 16:24:25.000000, size 525
I am not sure if this should be expected from passive, these docs[1] hint that passive 'compress if hinted COMPRESSIBLE'. From that I would conclude that all text messages should be compressed. A previous test with a 64kb gzip attachment seemed to not compress,
although I did not look at all object sizes.
on 14.2.11
[1] https://documentation.suse.com/ses/5.5/html/ses-all/ceph-pools.html #sec-ceph-pool-compression https://docs.ceph.com/docs/mimic/rados/operations/pools/ ________________________________
ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
-- Sent from my Android device with K-9 Mail. Please excuse my brevity.
-- David Caro
It is a hdd pool, all bluestore, configured with ceph-disk. Upgrades seem not to have 'updated' bluefs, some osds report like this: { "/dev/sdb2": { "osd_uuid": "xxxxxxxxxxx", "size": 4000681103360, "btime": "2019-01-08 13:45:59.488533", "description": "main", "bluefs": "1", "ceph_fsid": "xxxxxxxxx", "kv_backend": "rocksdb", "magic": "ceph osd volume v026", "mkfs_done": "yes", "ready": "ready", "require_osd_release": "14", "whoami": "3" } } And some like this: { "/dev/sdh2": { "osd_uuid": "xxxxxxxxxxx", "size": 3000487051264, "btime": "2017-07-14 14:45:59.212792", "description": "main", "require_osd_release": "14" } } -----Original Message----- Cc: ceph-users Subject: Re: [ceph-users] ceph rbox test on passive compressed pool On 09/11 09:36, Marc Roos wrote:
Hi David,
Just to let you know, this hint is being set, what is the reason for ceph of doing only half the objects? Can it be that there is some issue with my osd's? Like some maybe have an old fs (still using disk not volume)? Is this still to be expected or does ceph under pressure drop compressing?
https://github.com/ceph-dovecot/dovecot-ceph-plugin/blob/56d6c900cc9ec 07dfb98ef2abac07aae466b7610/src/librmb/rados-storage-impl.cpp#L75
I was trying to look into this a bit :), can you give me more info about the OSDs that you are using? What filesystem are they on? Cheers!
Thanks, Marc
-----Original Message----- Cc: jan.radon Subject: Re: [ceph-users] ceph rbox test on passive compressed pool
The hints have to be given from the client side as far as I understand, can you share the client code too?
Also,not seems that there's no guarantees that it will actually do anything (best effort I guess): https://docs.ceph.com/docs/mimic/rados/api/librados/#c.rados_set_alloc _hint
Cheers
On 6 September 2020 15:59:01 BST, Marc Roos <M.Roos@f1-outsourcing.eu> wrote:
I have been inserting 10790 exactly the same 64kb text message to
a
passive compressing enabled pool. I am still counting, but it
looks
like only half the objects are compressed.
mail/b08c3218dbf1545ff430000052412a8e mtime 2020-09-06 16:27:39.000000, size 63580 mail/00f6043775f1545ff430000052412a8e mtime 2020-09-06 16:25:57.000000, size 525 mail/b875f40571f1545ff430000052412a8e mtime 2020-09-06 16:25:53.000000, size 63580 mail/e87c120b19f1545ff430000052412a8e mtime 2020-09-06 16:24:25.000000, size 525
I am not sure if this should be expected from passive, these docs[1] hint that passive 'compress if hinted COMPRESSIBLE'. From that I would conclude that all text messages should be compressed. A previous test with a 64kb gzip attachment seemed to not compress,
although I did not look at all object sizes.
on 14.2.11
[1]
https://documentation.suse.com/ses/5.5/html/ses-all/ceph-pools.html
#sec-ceph-pool-compression https://docs.ceph.com/docs/mimic/rados/operations/pools/ ________________________________
ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
-- Sent from my Android device with K-9 Mail. Please excuse my brevity.
-- David Caro
mail/b875f40571f1545ff430000052412a8e mtime 2020-09-06 16:25:53.000000, size 63580 mail/e87c120b19f1545ff430000052412a8e mtime 2020-09-06 16:24:25.000000, size 525
Hi David, How is this going. To me this looks more like deduplication than compression. This is a link[1] to the 62kb text I used [1] https://pastebin.pl/view/e45ac998
mail/b875f40571f1545ff430000052412a8e mtime 2020-09-06 16:25:53.000000, size 63580 mail/e87c120b19f1545ff430000052412a8e mtime 2020-09-06 16:24:25.000000, size 525
Hi David, How is this going? To me this looks more like deduplication than compression. This is a link[1] to the 62kb text I used. I cannot really believe that this compresses to 526 bytes. If I compress this with gzip, it is already 7189 bytes. [1] https://pastebin.pl/view/e45ac998
Hi David, Jan, Terribly terribly sorry, I just noticed that the imaptest program used this 'DON'T DELETE THIS MESSAGE -- FOLDER INTERNAL DATA' to test with, which is 537 bytes. This explains at least why half the messages were so small. So conclusion is that. - either the hint is not correctly set to do the passive compression. - or the passive compression is not working when this hint is set. Thanks, Marc -----Original Message----- Cc: ceph-users Subject: Re: [ceph-users] ceph rbox test on passive compressed pool On 09/11 09:36, Marc Roos wrote:
Hi David,
Just to let you know, this hint is being set, what is the reason for ceph of doing only half the objects? Can it be that there is some issue with my osd's? Like some maybe have an old fs (still using disk not volume)? Is this still to be expected or does ceph under pressure drop compressing?
https://github.com/ceph-dovecot/dovecot-ceph-plugin/blob/56d6c900cc9ec 07dfb98ef2abac07aae466b7610/src/librmb/rados-storage-impl.cpp#L75
I was trying to look into this a bit :), can you give me more info about the OSDs that you are using? What filesystem are they on? Cheers!
Thanks, Marc
-----Original Message----- Cc: jan.radon Subject: Re: [ceph-users] ceph rbox test on passive compressed pool
The hints have to be given from the client side as far as I understand, can you share the client code too?
Also,not seems that there's no guarantees that it will actually do anything (best effort I guess): https://docs.ceph.com/docs/mimic/rados/api/librados/#c.rados_set_alloc _hint
Cheers
On 6 September 2020 15:59:01 BST, Marc Roos <M.Roos@f1-outsourcing.eu> wrote:
I have been inserting 10790 exactly the same 64kb text message to
a
passive compressing enabled pool. I am still counting, but it
looks
like only half the objects are compressed.
mail/b08c3218dbf1545ff430000052412a8e mtime 2020-09-06 16:27:39.000000, size 63580 mail/00f6043775f1545ff430000052412a8e mtime 2020-09-06 16:25:57.000000, size 525 mail/b875f40571f1545ff430000052412a8e mtime 2020-09-06 16:25:53.000000, size 63580 mail/e87c120b19f1545ff430000052412a8e mtime 2020-09-06 16:24:25.000000, size 525
I am not sure if this should be expected from passive, these docs[1] hint that passive 'compress if hinted COMPRESSIBLE'. From that I would conclude that all text messages should be compressed. A previous test with a 64kb gzip attachment seemed to not compress,
although I did not look at all object sizes.
on 14.2.11
[1]
https://documentation.suse.com/ses/5.5/html/ses-all/ceph-pools.html
#sec-ceph-pool-compression https://docs.ceph.com/docs/mimic/rados/operations/pools/ ________________________________
ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
-- Sent from my Android device with K-9 Mail. Please excuse my brevity.
-- David Caro
participants (3)
-
david
-
David Caro
-
Marc Roos