Re: Downside of many rgw bucket shards?

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Thanks for your input:

There are buckets with over 15m files and >300 shards, but yesterday a
customer with 2.5m files and 101 shards complained about the slowness of
listing files.
We do not have indexless buckets. I am not sure if a customer can create
such a bucket on their own via the usual tooling, but we do not create such
buckets.

I am happy to know that the default settings go up to 1999 shards, and I
reshard on my own (because we have a multi zonegroup setting, without
replication of buckets). So I tent to "overshard" buckets a bit (jump from
11 to 101 to 263) because I don't want to monitor the logfile continuously.


Am Mo., 29. Aug. 2022 um 19:49 Uhr schrieb Matt Benjamin <
mbenjami@xxxxxxxxxx>:

> We choose prime number shard counts, yes.
> Indexless buckets do increase insert-delete performance, but by
> definition, though, an indexless bucket cannot be listed.
>
> Matt
>
> On Mon, Aug 29, 2022 at 1:46 PM Anthony D'Atri <anthony.datri@xxxxxxxxx>
> wrote:
>
>> Do I recall that the number of shards is ideally odd, or even prime?
>> Performance might be increased by indexless buckets if the application can
>> handle
>>
>> > On Aug 29, 2022, at 10:06 AM, J. Eric Ivancich <ivancich@xxxxxxxxxx>
>> wrote:
>> >
>> > Generally it’s a good thing. There’s less contention for bucket index
>> updates when, for example, lots of writes are happening together. Dynamic
>> resharding will take things up to 1999 shards on its own with the default
>> config.
>> >
>> > Given that we use hashing of objet names to determine which shard they
>> go to, the most complicated operation is bucket listing, which has to
>> retrieve entries from each shard, order them, and return them to the
>> client. And it has to do this in batches of about 1000 at a time.
>> >
>> > It looks like you’re expecting on the order of 10,000,000 objects in
>> these buckets, so I imagine you’re not going to be listing them with any
>> regularity.
>> >
>> > Eric
>> > (he/him)
>> >
>> >> On Aug 29, 2022, at 12:06 PM, Boris Behrens <bb@xxxxxxxxx> wrote:
>> >>
>> >> Hi there,
>> >>
>> >> I have some buckets that would require >100 shards and I would like to
>> ask
>> >> if there are any downsides to have these many shards on a bucket?
>> >>
>> >> Cheers
>> >> Boris
>> >> _______________________________________________
>> >> ceph-users mailing list -- ceph-users@xxxxxxx
>> >> To unsubscribe send an email to ceph-users-leave@xxxxxxx
>> >>
>> >
>> > _______________________________________________
>> > ceph-users mailing list -- ceph-users@xxxxxxx
>> > To unsubscribe send an email to ceph-users-leave@xxxxxxx
>> _______________________________________________
>> ceph-users mailing list -- ceph-users@xxxxxxx
>> To unsubscribe send an email to ceph-users-leave@xxxxxxx
>>
>
>
> --
>
> Matt Benjamin
> Red Hat, Inc.
> 315 West Huron Street, Suite 140A
> Ann Arbor, Michigan 48103
>
> http://www.redhat.com/en/technologies/storage
>
> tel.  734-821-5101
> fax.  734-769-8938
> cel.  734-216-5309
>


-- 
Die Selbsthilfegruppe "UTF-8-Probleme" trifft sich diesmal abweichend im
groüen Saal.
_______________________________________________
ceph-users mailing list -- ceph-users@xxxxxxx
To unsubscribe send an email to ceph-users-leave@xxxxxxx




[Index of Archives]     [Information on CEPH]     [Linux Filesystem Development]     [Ceph Development]     [Ceph Large]     [Ceph Dev]     [Linux USB Development]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [xfs]


  Powered by Linux