metadata spill back onto block.slow before block.db filled up

metadata spill back onto block.slow before block.db filled up

[shasha lu]

Hi, Mark
We test bluestore with 12.2.1.
There are two host in our rgw cluster, each host contain 2 osds. The
rgw pool size is 2.  Using a 5GB partition for db.wal, a 50GB SSD
partition for block.db.

# ceph --admin-daemon ceph-osd.1.asok config get rocksdb_db_paths
{
    "rocksdb_db_paths": "db,51002736640 db.slow,284999998054"
}

After writing about 400W 4k rgw objects, using ceph-bluestore-tool to
export rocksdb file.

# ceph-bluestore-tool bluefs-export --path /var/lib/ceph/osd/osd1
--out-dir /tmp/osd1
# cd /tmp/osd1
# ls
db  db.slow  db.wal
# du -sh *
2.8G    db
809M    db.slow
439M    db.wal

block.db partition have 50GB space, but it only contains ~3GB files.
Then the metadata rolling over onto the db.slow.
It seems that only L0-L2 files located in block.db. (L0 256M; L1 256M;
L2 2.5GB), L3 and higher level file located in db.slow.

According to ceph docs, the metadata rolling over onto the db.slow
only when block.db filled up. But in our env the block.db partition is
far from filled up.
Did I make any mistakes?  Is there any additional options should be
set to rocksdb?


Thanks,
Shasha Lu

Re: metadata spill back onto block.slow before block.db filled up

[Sage Weil]

Hi Shasha,

On Tue, 28 Nov 2017, shasha lu wrote:
> Hi, Mark
> We test bluestore with 12.2.1.
> There are two host in our rgw cluster, each host contain 2 osds. The
> rgw pool size is 2.  Using a 5GB partition for db.wal, a 50GB SSD
> partition for block.db.
> 
> # ceph --admin-daemon ceph-osd.1.asok config get rocksdb_db_paths
> {
>     "rocksdb_db_paths": "db,51002736640 db.slow,284999998054"
> }
> 
> After writing about 400W 4k rgw objects, using ceph-bluestore-tool to
> export rocksdb file.
> 
> # ceph-bluestore-tool bluefs-export --path /var/lib/ceph/osd/osd1
> --out-dir /tmp/osd1
> # cd /tmp/osd1
> # ls
> db  db.slow  db.wal
> # du -sh *
> 2.8G    db
> 809M    db.slow
> 439M    db.wal
> 
> block.db partition have 50GB space, but it only contains ~3GB files.
> Then the metadata rolling over onto the db.slow.
> It seems that only L0-L2 files located in block.db. (L0 256M; L1 256M;
> L2 2.5GB), L3 and higher level file located in db.slow.
> 
> According to ceph docs, the metadata rolling over onto the db.slow
> only when block.db filled up. But in our env the block.db partition is
> far from filled up.
> Did I make any mistakes?  Is there any additional options should be
> set to rocksdb?

You didn't make any mistakes--this should happen automatically.  It looks 
like rocksdb isn't behaving as advertised.  I've opened 
http://tracker.ceph.com/issues/22264 to track this.  We need to start by 
reproducing the situation.

My guess is that rocksdb is deciding that deciding that all of L3 can't 
fit on db and so it's putting all of L3 on db.slow?

sage

Re: metadata spill back onto block.slow before block.db filled up

[Igor Fedotov]

Looks like I can easily reproduce that (note slow_used_bytes):

  "bluefs": {
         "gift_bytes": 105906176,
         "reclaim_bytes": 0,
         "db_total_bytes": 4294959104,
         "db_used_bytes": 76546048,
         "wal_total_bytes": 1073737728,
         "wal_used_bytes": 239075328,
         "slow_total_bytes": 1179648000,
         "slow_used_bytes": 63963136,
         "num_files": 13,
         "log_bytes": 2539520,
         "log_compactions": 3,
         "logged_bytes": 255176704,
         "files_written_wal": 3,
         "files_written_sst": 10,
         "bytes_written_wal": 1932165189,
         "bytes_written_sst": 340957748
     },


On 11/28/2017 4:17 PM, Sage Weil wrote:
> Hi Shasha,
>
> On Tue, 28 Nov 2017, shasha lu wrote:
>> Hi, Mark
>> We test bluestore with 12.2.1.
>> There are two host in our rgw cluster, each host contain 2 osds. The
>> rgw pool size is 2.  Using a 5GB partition for db.wal, a 50GB SSD
>> partition for block.db.
>>
>> # ceph --admin-daemon ceph-osd.1.asok config get rocksdb_db_paths
>> {
>>      "rocksdb_db_paths": "db,51002736640 db.slow,284999998054"
>> }
>>
>> After writing about 400W 4k rgw objects, using ceph-bluestore-tool to
>> export rocksdb file.
>>
>> # ceph-bluestore-tool bluefs-export --path /var/lib/ceph/osd/osd1
>> --out-dir /tmp/osd1
>> # cd /tmp/osd1
>> # ls
>> db  db.slow  db.wal
>> # du -sh *
>> 2.8G    db
>> 809M    db.slow
>> 439M    db.wal
>>
>> block.db partition have 50GB space, but it only contains ~3GB files.
>> Then the metadata rolling over onto the db.slow.
>> It seems that only L0-L2 files located in block.db. (L0 256M; L1 256M;
>> L2 2.5GB), L3 and higher level file located in db.slow.
>>
>> According to ceph docs, the metadata rolling over onto the db.slow
>> only when block.db filled up. But in our env the block.db partition is
>> far from filled up.
>> Did I make any mistakes?  Is there any additional options should be
>> set to rocksdb?
> You didn't make any mistakes--this should happen automatically.  It looks
> like rocksdb isn't behaving as advertised.  I've opened
> http://tracker.ceph.com/issues/22264 to track this.  We need to start by
> reproducing the situation.
>
> My guess is that rocksdb is deciding that deciding that all of L3 can't
> fit on db and so it's putting all of L3 on db.slow?
>
> sage
>
> --
> To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: metadata spill back onto block.slow before block.db filled up

[Mark Nelson]

Looks like a bug guys! :)  Mind making a ticket in the tracker?

Mark

On 11/28/2017 07:39 AM, Igor Fedotov wrote:
> Looks like I can easily reproduce that (note slow_used_bytes):
>
>  "bluefs": {
>         "gift_bytes": 105906176,
>         "reclaim_bytes": 0,
>         "db_total_bytes": 4294959104,
>         "db_used_bytes": 76546048,
>         "wal_total_bytes": 1073737728,
>         "wal_used_bytes": 239075328,
>         "slow_total_bytes": 1179648000,
>         "slow_used_bytes": 63963136,
>         "num_files": 13,
>         "log_bytes": 2539520,
>         "log_compactions": 3,
>         "logged_bytes": 255176704,
>         "files_written_wal": 3,
>         "files_written_sst": 10,
>         "bytes_written_wal": 1932165189,
>         "bytes_written_sst": 340957748
>     },
>
>
> On 11/28/2017 4:17 PM, Sage Weil wrote:
>> Hi Shasha,
>>
>> On Tue, 28 Nov 2017, shasha lu wrote:
>>> Hi, Mark
>>> We test bluestore with 12.2.1.
>>> There are two host in our rgw cluster, each host contain 2 osds. The
>>> rgw pool size is 2.  Using a 5GB partition for db.wal, a 50GB SSD
>>> partition for block.db.
>>>
>>> # ceph --admin-daemon ceph-osd.1.asok config get rocksdb_db_paths
>>> {
>>>      "rocksdb_db_paths": "db,51002736640 db.slow,284999998054"
>>> }
>>>
>>> After writing about 400W 4k rgw objects, using ceph-bluestore-tool to
>>> export rocksdb file.
>>>
>>> # ceph-bluestore-tool bluefs-export --path /var/lib/ceph/osd/osd1
>>> --out-dir /tmp/osd1
>>> # cd /tmp/osd1
>>> # ls
>>> db  db.slow  db.wal
>>> # du -sh *
>>> 2.8G    db
>>> 809M    db.slow
>>> 439M    db.wal
>>>
>>> block.db partition have 50GB space, but it only contains ~3GB files.
>>> Then the metadata rolling over onto the db.slow.
>>> It seems that only L0-L2 files located in block.db. (L0 256M; L1 256M;
>>> L2 2.5GB), L3 and higher level file located in db.slow.
>>>
>>> According to ceph docs, the metadata rolling over onto the db.slow
>>> only when block.db filled up. But in our env the block.db partition is
>>> far from filled up.
>>> Did I make any mistakes?  Is there any additional options should be
>>> set to rocksdb?
>> You didn't make any mistakes--this should happen automatically.  It looks
>> like rocksdb isn't behaving as advertised.  I've opened
>> http://tracker.ceph.com/issues/22264 to track this.  We need to start by
>> reproducing the situation.
>>
>> My guess is that rocksdb is deciding that deciding that all of L3 can't
>> fit on db and so it's putting all of L3 on db.slow?
>>
>> sage
>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>

Re: metadata spill back onto block.slow before block.db filled up

[Igor Fedotov]

here it is

http://tracker.ceph.com/issues/22264


On 11/28/2017 6:37 PM, Mark Nelson wrote:
> Looks like a bug guys! :)  Mind making a ticket in the tracker?
>
> Mark
>
> On 11/28/2017 07:39 AM, Igor Fedotov wrote:
>> Looks like I can easily reproduce that (note slow_used_bytes):
>>
>>  "bluefs": {
>>         "gift_bytes": 105906176,
>>         "reclaim_bytes": 0,
>>         "db_total_bytes": 4294959104,
>>         "db_used_bytes": 76546048,
>>         "wal_total_bytes": 1073737728,
>>         "wal_used_bytes": 239075328,
>>         "slow_total_bytes": 1179648000,
>>         "slow_used_bytes": 63963136,
>>         "num_files": 13,
>>         "log_bytes": 2539520,
>>         "log_compactions": 3,
>>         "logged_bytes": 255176704,
>>         "files_written_wal": 3,
>>         "files_written_sst": 10,
>>         "bytes_written_wal": 1932165189,
>>         "bytes_written_sst": 340957748
>>     },
>>
>>
>> On 11/28/2017 4:17 PM, Sage Weil wrote:
>>> Hi Shasha,
>>>
>>> On Tue, 28 Nov 2017, shasha lu wrote:
>>>> Hi, Mark
>>>> We test bluestore with 12.2.1.
>>>> There are two host in our rgw cluster, each host contain 2 osds. The
>>>> rgw pool size is 2.  Using a 5GB partition for db.wal, a 50GB SSD
>>>> partition for block.db.
>>>>
>>>> # ceph --admin-daemon ceph-osd.1.asok config get rocksdb_db_paths
>>>> {
>>>>      "rocksdb_db_paths": "db,51002736640 db.slow,284999998054"
>>>> }
>>>>
>>>> After writing about 400W 4k rgw objects, using ceph-bluestore-tool to
>>>> export rocksdb file.
>>>>
>>>> # ceph-bluestore-tool bluefs-export --path /var/lib/ceph/osd/osd1
>>>> --out-dir /tmp/osd1
>>>> # cd /tmp/osd1
>>>> # ls
>>>> db  db.slow  db.wal
>>>> # du -sh *
>>>> 2.8G    db
>>>> 809M    db.slow
>>>> 439M    db.wal
>>>>
>>>> block.db partition have 50GB space, but it only contains ~3GB files.
>>>> Then the metadata rolling over onto the db.slow.
>>>> It seems that only L0-L2 files located in block.db. (L0 256M; L1 256M;
>>>> L2 2.5GB), L3 and higher level file located in db.slow.
>>>>
>>>> According to ceph docs, the metadata rolling over onto the db.slow
>>>> only when block.db filled up. But in our env the block.db partition is
>>>> far from filled up.
>>>> Did I make any mistakes?  Is there any additional options should be
>>>> set to rocksdb?
>>> You didn't make any mistakes--this should happen automatically.  It 
>>> looks
>>> like rocksdb isn't behaving as advertised.  I've opened
>>> http://tracker.ceph.com/issues/22264 to track this.  We need to 
>>> start by
>>> reproducing the situation.
>>>
>>> My guess is that rocksdb is deciding that deciding that all of L3 can't
>>> fit on db and so it's putting all of L3 on db.slow?
>>>
>>> sage
>>>
>>> -- 
>>> To unsubscribe from this list: send the line "unsubscribe 
>>> ceph-devel" in
>>> the body of a message to majordomo@vger.kernel.org
>>> More majordomo info at http://vger.kernel.org/majordomo-info.html
>>

Re: metadata spill back onto block.slow before block.db filled up

[Igor Fedotov]

I've just updated the bug notes.

Most probably the issue is caused by an already fixed bug in RocksDB,

see 
https://github.com/facebook/rocksdb/commit/65a9cd616876c7a1204e1a50990400e4e1f61d7e

Hence the question is if we plan to backport the fix and how to arrange 
that.

Thanks,
Igor

On 11/28/2017 6:44 PM, Igor Fedotov wrote:
> here it is
>
> http://tracker.ceph.com/issues/22264
>
>
> On 11/28/2017 6:37 PM, Mark Nelson wrote:
>> Looks like a bug guys! :)  Mind making a ticket in the tracker?
>>
>> Mark
>>
>> On 11/28/2017 07:39 AM, Igor Fedotov wrote:
>>> Looks like I can easily reproduce that (note slow_used_bytes):
>>>
>>>  "bluefs": {
>>>         "gift_bytes": 105906176,
>>>         "reclaim_bytes": 0,
>>>         "db_total_bytes": 4294959104,
>>>         "db_used_bytes": 76546048,
>>>         "wal_total_bytes": 1073737728,
>>>         "wal_used_bytes": 239075328,
>>>         "slow_total_bytes": 1179648000,
>>>         "slow_used_bytes": 63963136,
>>>         "num_files": 13,
>>>         "log_bytes": 2539520,
>>>         "log_compactions": 3,
>>>         "logged_bytes": 255176704,
>>>         "files_written_wal": 3,
>>>         "files_written_sst": 10,
>>>         "bytes_written_wal": 1932165189,
>>>         "bytes_written_sst": 340957748
>>>     },
>>>
>>>
>>> On 11/28/2017 4:17 PM, Sage Weil wrote:
>>>> Hi Shasha,
>>>>
>>>> On Tue, 28 Nov 2017, shasha lu wrote:
>>>>> Hi, Mark
>>>>> We test bluestore with 12.2.1.
>>>>> There are two host in our rgw cluster, each host contain 2 osds. The
>>>>> rgw pool size is 2.  Using a 5GB partition for db.wal, a 50GB SSD
>>>>> partition for block.db.
>>>>>
>>>>> # ceph --admin-daemon ceph-osd.1.asok config get rocksdb_db_paths
>>>>> {
>>>>>      "rocksdb_db_paths": "db,51002736640 db.slow,284999998054"
>>>>> }
>>>>>
>>>>> After writing about 400W 4k rgw objects, using ceph-bluestore-tool to
>>>>> export rocksdb file.
>>>>>
>>>>> # ceph-bluestore-tool bluefs-export --path /var/lib/ceph/osd/osd1
>>>>> --out-dir /tmp/osd1
>>>>> # cd /tmp/osd1
>>>>> # ls
>>>>> db  db.slow  db.wal
>>>>> # du -sh *
>>>>> 2.8G    db
>>>>> 809M    db.slow
>>>>> 439M    db.wal
>>>>>
>>>>> block.db partition have 50GB space, but it only contains ~3GB files.
>>>>> Then the metadata rolling over onto the db.slow.
>>>>> It seems that only L0-L2 files located in block.db. (L0 256M; L1 
>>>>> 256M;
>>>>> L2 2.5GB), L3 and higher level file located in db.slow.
>>>>>
>>>>> According to ceph docs, the metadata rolling over onto the db.slow
>>>>> only when block.db filled up. But in our env the block.db 
>>>>> partition is
>>>>> far from filled up.
>>>>> Did I make any mistakes?  Is there any additional options should be
>>>>> set to rocksdb?
>>>> You didn't make any mistakes--this should happen automatically.  It 
>>>> looks
>>>> like rocksdb isn't behaving as advertised.  I've opened
>>>> http://tracker.ceph.com/issues/22264 to track this.  We need to 
>>>> start by
>>>> reproducing the situation.
>>>>
>>>> My guess is that rocksdb is deciding that deciding that all of L3 
>>>> can't
>>>> fit on db and so it's putting all of L3 on db.slow?
>>>>
>>>> sage
>>>>
>>>> -- 
>>>> To unsubscribe from this list: send the line "unsubscribe 
>>>> ceph-devel" in
>>>> the body of a message to majordomo@vger.kernel.org
>>>> More majordomo info at http://vger.kernel.org/majordomo-info.html
>>>
>
> -- 
> To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: metadata spill back onto block.slow before block.db filled up

[Sage Weil]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 4635 bytes --]

On Wed, 29 Nov 2017, Igor Fedotov wrote:
> I've just updated the bug notes.
> 
> Most probably the issue is caused by an already fixed bug in RocksDB,
> 
> see
> https://github.com/facebook/rocksdb/commit/65a9cd616876c7a1204e1a50990400e4e1f61d7e
> 
> Hence the question is if we plan to backport the fix and how to arrange that.

If you cna confirm the problem doesn't reproduce after cherry-picking that 
commit, we can just do that for the luminous branch.

For master, let's fast forward rocksdb past it?

sage


> 
> Thanks,
> Igor
> 
> On 11/28/2017 6:44 PM, Igor Fedotov wrote:
> > here it is
> > 
> > http://tracker.ceph.com/issues/22264
> > 
> > 
> > On 11/28/2017 6:37 PM, Mark Nelson wrote:
> > > Looks like a bug guys! :)  Mind making a ticket in the tracker?
> > > 
> > > Mark
> > > 
> > > On 11/28/2017 07:39 AM, Igor Fedotov wrote:
> > > > Looks like I can easily reproduce that (note slow_used_bytes):
> > > > 
> > > >  "bluefs": {
> > > >         "gift_bytes": 105906176,
> > > >         "reclaim_bytes": 0,
> > > >         "db_total_bytes": 4294959104,
> > > >         "db_used_bytes": 76546048,
> > > >         "wal_total_bytes": 1073737728,
> > > >         "wal_used_bytes": 239075328,
> > > >         "slow_total_bytes": 1179648000,
> > > >         "slow_used_bytes": 63963136,
> > > >         "num_files": 13,
> > > >         "log_bytes": 2539520,
> > > >         "log_compactions": 3,
> > > >         "logged_bytes": 255176704,
> > > >         "files_written_wal": 3,
> > > >         "files_written_sst": 10,
> > > >         "bytes_written_wal": 1932165189,
> > > >         "bytes_written_sst": 340957748
> > > >     },
> > > > 
> > > > 
> > > > On 11/28/2017 4:17 PM, Sage Weil wrote:
> > > > > Hi Shasha,
> > > > > 
> > > > > On Tue, 28 Nov 2017, shasha lu wrote:
> > > > > > Hi, Mark
> > > > > > We test bluestore with 12.2.1.
> > > > > > There are two host in our rgw cluster, each host contain 2 osds. The
> > > > > > rgw pool size is 2.  Using a 5GB partition for db.wal, a 50GB SSD
> > > > > > partition for block.db.
> > > > > > 
> > > > > > # ceph --admin-daemon ceph-osd.1.asok config get rocksdb_db_paths
> > > > > > {
> > > > > >      "rocksdb_db_paths": "db,51002736640 db.slow,284999998054"
> > > > > > }
> > > > > > 
> > > > > > After writing about 400W 4k rgw objects, using ceph-bluestore-tool
> > > > > > to
> > > > > > export rocksdb file.
> > > > > > 
> > > > > > # ceph-bluestore-tool bluefs-export --path /var/lib/ceph/osd/osd1
> > > > > > --out-dir /tmp/osd1
> > > > > > # cd /tmp/osd1
> > > > > > # ls
> > > > > > db  db.slow  db.wal
> > > > > > # du -sh *
> > > > > > 2.8G    db
> > > > > > 809M    db.slow
> > > > > > 439M    db.wal
> > > > > > 
> > > > > > block.db partition have 50GB space, but it only contains ~3GB files.
> > > > > > Then the metadata rolling over onto the db.slow.
> > > > > > It seems that only L0-L2 files located in block.db. (L0 256M; L1
> > > > > > 256M;
> > > > > > L2 2.5GB), L3 and higher level file located in db.slow.
> > > > > > 
> > > > > > According to ceph docs, the metadata rolling over onto the db.slow
> > > > > > only when block.db filled up. But in our env the block.db partition
> > > > > > is
> > > > > > far from filled up.
> > > > > > Did I make any mistakes?  Is there any additional options should be
> > > > > > set to rocksdb?
> > > > > You didn't make any mistakes--this should happen automatically.  It
> > > > > looks
> > > > > like rocksdb isn't behaving as advertised.  I've opened
> > > > > http://tracker.ceph.com/issues/22264 to track this.  We need to start
> > > > > by
> > > > > reproducing the situation.
> > > > > 
> > > > > My guess is that rocksdb is deciding that deciding that all of L3
> > > > > can't
> > > > > fit on db and so it's putting all of L3 on db.slow?
> > > > > 
> > > > > sage
> > > > > 
> > > > > -- 
> > > > > To unsubscribe from this list: send the line "unsubscribe ceph-devel"
> > > > > in
> > > > > the body of a message to majordomo@vger.kernel.org
> > > > > More majordomo info at http://vger.kernel.org/majordomo-info.html
> > > > 
> > 
> > -- 
> > To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
> > the body of a message to majordomo@vger.kernel.org
> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 
> --
> To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 
> 

Re: metadata spill back onto block.slow before block.db filled up

[shasha lu]

After cherry-picking that commit, the problem doesn't reproduce in our env.
All L0-L3 files located in db now.  (L0 256M; L1 256M; L2 2.5G;  L3 25G)
L4 250G, all of L4 can't fit on db, so it will put all of L4 on db.slow.

ceph daemon osd.4 perf dump | grep bluefs -A 17
    "bluefs": {
        "gift_bytes": 0,
        "reclaim_bytes": 0,
        "db_total_bytes": 53687083008,
        "db_used_bytes": 5341446144,
        "wal_total_bytes": 5368705024,
        "wal_used_bytes": 524288000,
        "slow_total_bytes": 12000952320,
        "slow_used_bytes": 0,
        "num_files": 94,
        "log_bytes": 3330048,
        "log_compactions": 2,
        "logged_bytes": 556756992,
        "files_written_wal": 2,
        "files_written_sst": 620,
        "bytes_written_wal": 32516096174,
        "bytes_written_sst": 38724761218
    },

The rocksdb of ceph master already cherry-picking that. But v12.2.1
still have this problem.

So the size of db should be well planed.  The size of db is 30GB or
100GB make no difference. It will put all of L4 on db.slow.

Shasha

2017-11-29 21:07 GMT+08:00 Sage Weil <sage@newdream.net>:
> On Wed, 29 Nov 2017, Igor Fedotov wrote:
>> I've just updated the bug notes.
>>
>> Most probably the issue is caused by an already fixed bug in RocksDB,
>>
>> see
>> https://github.com/facebook/rocksdb/commit/65a9cd616876c7a1204e1a50990400e4e1f61d7e
>>
>> Hence the question is if we plan to backport the fix and how to arrange that.
>
> If you cna confirm the problem doesn't reproduce after cherry-picking that
> commit, we can just do that for the luminous branch.
>
> For master, let's fast forward rocksdb past it?
>
> sage
>
>
>>
>> Thanks,
>> Igor
>>
>> On 11/28/2017 6:44 PM, Igor Fedotov wrote:
>> > here it is
>> >
>> > http://tracker.ceph.com/issues/22264
>> >
>> >
>> > On 11/28/2017 6:37 PM, Mark Nelson wrote:
>> > > Looks like a bug guys! :)  Mind making a ticket in the tracker?
>> > >
>> > > Mark
>> > >
>> > > On 11/28/2017 07:39 AM, Igor Fedotov wrote:
>> > > > Looks like I can easily reproduce that (note slow_used_bytes):
>> > > >
>> > > >  "bluefs": {
>> > > >         "gift_bytes": 105906176,
>> > > >         "reclaim_bytes": 0,
>> > > >         "db_total_bytes": 4294959104,
>> > > >         "db_used_bytes": 76546048,
>> > > >         "wal_total_bytes": 1073737728,
>> > > >         "wal_used_bytes": 239075328,
>> > > >         "slow_total_bytes": 1179648000,
>> > > >         "slow_used_bytes": 63963136,
>> > > >         "num_files": 13,
>> > > >         "log_bytes": 2539520,
>> > > >         "log_compactions": 3,
>> > > >         "logged_bytes": 255176704,
>> > > >         "files_written_wal": 3,
>> > > >         "files_written_sst": 10,
>> > > >         "bytes_written_wal": 1932165189,
>> > > >         "bytes_written_sst": 340957748
>> > > >     },
>> > > >
>> > > >
>> > > > On 11/28/2017 4:17 PM, Sage Weil wrote:
>> > > > > Hi Shasha,
>> > > > >
>> > > > > On Tue, 28 Nov 2017, shasha lu wrote:
>> > > > > > Hi, Mark
>> > > > > > We test bluestore with 12.2.1.
>> > > > > > There are two host in our rgw cluster, each host contain 2 osds. The
>> > > > > > rgw pool size is 2.  Using a 5GB partition for db.wal, a 50GB SSD
>> > > > > > partition for block.db.
>> > > > > >
>> > > > > > # ceph --admin-daemon ceph-osd.1.asok config get rocksdb_db_paths
>> > > > > > {
>> > > > > >      "rocksdb_db_paths": "db,51002736640 db.slow,284999998054"
>> > > > > > }
>> > > > > >
>> > > > > > After writing about 400W 4k rgw objects, using ceph-bluestore-tool
>> > > > > > to
>> > > > > > export rocksdb file.
>> > > > > >
>> > > > > > # ceph-bluestore-tool bluefs-export --path /var/lib/ceph/osd/osd1
>> > > > > > --out-dir /tmp/osd1
>> > > > > > # cd /tmp/osd1
>> > > > > > # ls
>> > > > > > db  db.slow  db.wal
>> > > > > > # du -sh *
>> > > > > > 2.8G    db
>> > > > > > 809M    db.slow
>> > > > > > 439M    db.wal
>> > > > > >
>> > > > > > block.db partition have 50GB space, but it only contains ~3GB files.
>> > > > > > Then the metadata rolling over onto the db.slow.
>> > > > > > It seems that only L0-L2 files located in block.db. (L0 256M; L1
>> > > > > > 256M;
>> > > > > > L2 2.5GB), L3 and higher level file located in db.slow.
>> > > > > >
>> > > > > > According to ceph docs, the metadata rolling over onto the db.slow
>> > > > > > only when block.db filled up. But in our env the block.db partition
>> > > > > > is
>> > > > > > far from filled up.
>> > > > > > Did I make any mistakes?  Is there any additional options should be
>> > > > > > set to rocksdb?
>> > > > > You didn't make any mistakes--this should happen automatically.  It
>> > > > > looks
>> > > > > like rocksdb isn't behaving as advertised.  I've opened
>> > > > > http://tracker.ceph.com/issues/22264 to track this.  We need to start
>> > > > > by
>> > > > > reproducing the situation.
>> > > > >
>> > > > > My guess is that rocksdb is deciding that deciding that all of L3
>> > > > > can't
>> > > > > fit on db and so it's putting all of L3 on db.slow?
>> > > > >
>> > > > > sage
>> > > > >
>> > > > > --
>> > > > > To unsubscribe from this list: send the line "unsubscribe ceph-devel"
>> > > > > in
>> > > > > the body of a message to majordomo@vger.kernel.org
>> > > > > More majordomo info at http://vger.kernel.org/majordomo-info.html
>> > > >
>> >
>> > --
>> > To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
>> > the body of a message to majordomo@vger.kernel.org
>> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>
>>

Re: metadata spill back onto block.slow before block.db filled up

[Igor Fedotov]

Submitted PR https://github.com/ceph/ceph/pull/19257


On 11/30/2017 12:40 PM, shasha lu wrote:
> After cherry-picking that commit, the problem doesn't reproduce in our env.
> All L0-L3 files located in db now.  (L0 256M; L1 256M; L2 2.5G;  L3 25G)
> L4 250G, all of L4 can't fit on db, so it will put all of L4 on db.slow.
>
> ceph daemon osd.4 perf dump | grep bluefs -A 17
>      "bluefs": {
>          "gift_bytes": 0,
>          "reclaim_bytes": 0,
>          "db_total_bytes": 53687083008,
>          "db_used_bytes": 5341446144,
>          "wal_total_bytes": 5368705024,
>          "wal_used_bytes": 524288000,
>          "slow_total_bytes": 12000952320,
>          "slow_used_bytes": 0,
>          "num_files": 94,
>          "log_bytes": 3330048,
>          "log_compactions": 2,
>          "logged_bytes": 556756992,
>          "files_written_wal": 2,
>          "files_written_sst": 620,
>          "bytes_written_wal": 32516096174,
>          "bytes_written_sst": 38724761218
>      },
>
> The rocksdb of ceph master already cherry-picking that. But v12.2.1
> still have this problem.
>
> So the size of db should be well planed.  The size of db is 30GB or
> 100GB make no difference. It will put all of L4 on db.slow.
>
> Shasha
>
> 2017-11-29 21:07 GMT+08:00 Sage Weil <sage@newdream.net>:
>> On Wed, 29 Nov 2017, Igor Fedotov wrote:
>>> I've just updated the bug notes.
>>>
>>> Most probably the issue is caused by an already fixed bug in RocksDB,
>>>
>>> see
>>> https://github.com/facebook/rocksdb/commit/65a9cd616876c7a1204e1a50990400e4e1f61d7e
>>>
>>> Hence the question is if we plan to backport the fix and how to arrange that.
>> If you cna confirm the problem doesn't reproduce after cherry-picking that
>> commit, we can just do that for the luminous branch.
>>
>> For master, let's fast forward rocksdb past it?
>>
>> sage
>>
>>
>>> Thanks,
>>> Igor
>>>
>>> On 11/28/2017 6:44 PM, Igor Fedotov wrote:
>>>> here it is
>>>>
>>>> http://tracker.ceph.com/issues/22264
>>>>
>>>>
>>>> On 11/28/2017 6:37 PM, Mark Nelson wrote:
>>>>> Looks like a bug guys! :)  Mind making a ticket in the tracker?
>>>>>
>>>>> Mark
>>>>>
>>>>> On 11/28/2017 07:39 AM, Igor Fedotov wrote:
>>>>>> Looks like I can easily reproduce that (note slow_used_bytes):
>>>>>>
>>>>>>   "bluefs": {
>>>>>>          "gift_bytes": 105906176,
>>>>>>          "reclaim_bytes": 0,
>>>>>>          "db_total_bytes": 4294959104,
>>>>>>          "db_used_bytes": 76546048,
>>>>>>          "wal_total_bytes": 1073737728,
>>>>>>          "wal_used_bytes": 239075328,
>>>>>>          "slow_total_bytes": 1179648000,
>>>>>>          "slow_used_bytes": 63963136,
>>>>>>          "num_files": 13,
>>>>>>          "log_bytes": 2539520,
>>>>>>          "log_compactions": 3,
>>>>>>          "logged_bytes": 255176704,
>>>>>>          "files_written_wal": 3,
>>>>>>          "files_written_sst": 10,
>>>>>>          "bytes_written_wal": 1932165189,
>>>>>>          "bytes_written_sst": 340957748
>>>>>>      },
>>>>>>
>>>>>>
>>>>>> On 11/28/2017 4:17 PM, Sage Weil wrote:
>>>>>>> Hi Shasha,
>>>>>>>
>>>>>>> On Tue, 28 Nov 2017, shasha lu wrote:
>>>>>>>> Hi, Mark
>>>>>>>> We test bluestore with 12.2.1.
>>>>>>>> There are two host in our rgw cluster, each host contain 2 osds. The
>>>>>>>> rgw pool size is 2.  Using a 5GB partition for db.wal, a 50GB SSD
>>>>>>>> partition for block.db.
>>>>>>>>
>>>>>>>> # ceph --admin-daemon ceph-osd.1.asok config get rocksdb_db_paths
>>>>>>>> {
>>>>>>>>       "rocksdb_db_paths": "db,51002736640 db.slow,284999998054"
>>>>>>>> }
>>>>>>>>
>>>>>>>> After writing about 400W 4k rgw objects, using ceph-bluestore-tool
>>>>>>>> to
>>>>>>>> export rocksdb file.
>>>>>>>>
>>>>>>>> # ceph-bluestore-tool bluefs-export --path /var/lib/ceph/osd/osd1
>>>>>>>> --out-dir /tmp/osd1
>>>>>>>> # cd /tmp/osd1
>>>>>>>> # ls
>>>>>>>> db  db.slow  db.wal
>>>>>>>> # du -sh *
>>>>>>>> 2.8G    db
>>>>>>>> 809M    db.slow
>>>>>>>> 439M    db.wal
>>>>>>>>
>>>>>>>> block.db partition have 50GB space, but it only contains ~3GB files.
>>>>>>>> Then the metadata rolling over onto the db.slow.
>>>>>>>> It seems that only L0-L2 files located in block.db. (L0 256M; L1
>>>>>>>> 256M;
>>>>>>>> L2 2.5GB), L3 and higher level file located in db.slow.
>>>>>>>>
>>>>>>>> According to ceph docs, the metadata rolling over onto the db.slow
>>>>>>>> only when block.db filled up. But in our env the block.db partition
>>>>>>>> is
>>>>>>>> far from filled up.
>>>>>>>> Did I make any mistakes?  Is there any additional options should be
>>>>>>>> set to rocksdb?
>>>>>>> You didn't make any mistakes--this should happen automatically.  It
>>>>>>> looks
>>>>>>> like rocksdb isn't behaving as advertised.  I've opened
>>>>>>> http://tracker.ceph.com/issues/22264 to track this.  We need to start
>>>>>>> by
>>>>>>> reproducing the situation.
>>>>>>>
>>>>>>> My guess is that rocksdb is deciding that deciding that all of L3
>>>>>>> can't
>>>>>>> fit on db and so it's putting all of L3 on db.slow?
>>>>>>>
>>>>>>> sage
>>>>>>>
>>>>>>> --
>>>>>>> To unsubscribe from this list: send the line "unsubscribe ceph-devel"
>>>>>>> in
>>>>>>> the body of a message to majordomo@vger.kernel.org
>>>>>>> More majordomo info at http://vger.kernel.org/majordomo-info.html
>>>> --
>>>> To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
>>>> the body of a message to majordomo@vger.kernel.org
>>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
>>> the body of a message to majordomo@vger.kernel.org
>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>>
>>>