Re: Supporting localise Read for RGW

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

 



For userspace clients I think we can implement a normal config option (rados_localize_reads) and set it to true in the mon config store, and then clients will pick it up. But that would be global to all ops — for it to be pool-based, we’d have to extend the OSDMap with client information and make some more extensive Objecter changes (and kclient ones). That’s not impossible but feels a bit weird. :/
-Greg

On Mon, Jul 3, 2023 at 11:36 AM Matt Benjamin <mbenjami@xxxxxxxxxx> wrote:
is there a global/cluster-wide version of such a flag, or to automate applying it?  it seems like the scenarios I hear about are stretched clusters, and that it would be attractive to not have to manage that by exception?

Matt

On Mon, Jul 3, 2023 at 12:25 PM Casey Bodley <cbodley@xxxxxxxxxx> wrote:
On Tue, Jun 27, 2023 at 7:12 PM Sam Just <sjust@xxxxxxxxxx> wrote:
>
> It would probably make sense to add support in the objecter for
> picking up certain flags (such as LOCALIZE_READS) from the pool
> config.
> -Sam

thanks Sam. Jiffin, would you be willing to look into this Objecter
part? i think it would make your project easier if the feature was
transparent to rgw

>
> On Sun, Jun 18, 2023 at 1:34 AM Josh Salomon <jsalomon@xxxxxxxxxx> wrote:
> >
> > One more point - from a business perspective, this flag is per pool and not global so the API should be per pool and not global.
> > Regards,
> >
> > Josh
> >
> >
> > On Sat, Jun 17, 2023 at 4:22 PM Ilya Dryomov <idryomov@xxxxxxxxx> wrote:
> >>
> >> On Fri, Jun 16, 2023 at 5:13 PM Casey Bodley <cbodley@xxxxxxxxxx> wrote:
> >> >
> >> > On Fri, Jun 16, 2023 at 3:21 AM Jiffin Thottan <jthottan@xxxxxxxxxx> wrote:
> >> > >
> >> > > Hi,
> >> > >
> >> > > I am planning to support read localise feature for RGW servers similar to what RBD volumes support. From code reading, it looks to pass "librados::OPERATION_LOCALIZE_READS" before sending the request to RADOS.  I have created a tracker issue for this feature https://tracker.ceph.com/issues/61701, this will be per server config option. I don't know about any other technical hurdles in implementing this feature. Please share your thoughts on the same.
> >> > >
> >> > > Thanks and regards,
> >> > > Jiffin
> >> >
> >> > thanks Jiffin,
> >> >
> >> > that sounds simple enough, though rgw issues librados reads in many
> >> > different places that would need to manage this flag
> >> >
> >> > i'm curious why this is a per-op flag, rather than a global librados
> >> > setting. is there a reason that rgw would only want *some* of its
> >> > reads localized? are there any implications for read-after-write
> >> > consistency that we need to worry about here?
> >>
> >> Hi Casey,
> >>
> >> I'm not aware of any read-after-write or other consistency
> >> implications.  Read-from-replica is safe for general use since Octopus
> >> (OSD-side issues were fixed in [1]).
> >>
> >> Ultimately all flags are per-op.  The reason that there is no global
> >> librados setting for BALANCE_READS and LOCALIZE_READS seems to be
> >> a combination of a) the fact that these flags could only be used in
> >> special circumstances in the past and b) the fact that librados doesn't
> >> have a generic API for applying a given flag globally like it does for
> >> applying a given snapshot ID/context to all reads/writes.
> >>
> >> [1] https://github.com/ceph/ceph/pull/32381
> >>
> >> Thanks,
> >>
> >>                 Ilya
> >> _______________________________________________
> >> Dev mailing list -- dev@xxxxxxx
> >> To unsubscribe send an email to dev-leave@xxxxxxx
> >
> > _______________________________________________
> > Dev mailing list -- dev@xxxxxxx
> > To unsubscribe send an email to dev-leave@xxxxxxx
>
_______________________________________________
Dev mailing list -- dev@xxxxxxx
To unsubscribe send an email to dev-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
_______________________________________________
Dev mailing list -- dev@xxxxxxx
To unsubscribe send an email to dev-leave@xxxxxxx
_______________________________________________
Dev mailing list -- dev@xxxxxxx
To unsubscribe send an email to dev-leave@xxxxxxx

[Index of Archives]     [CEPH Users]     [Ceph Devel]     [Ceph Large]     [Information on CEPH]     [Linux BTRFS]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux