On Fri, Oct 1, 2021 at 11:55 AM Andrew Gunnerson <accounts.c...@chiller3.com> wrote: > > Thanks for the info. Do shadow roots affect that calculation at all?
They're not supposed to, but I haven't worked in this code...but the different names you get (with the ~ssd and ~hdd postfix) would indicate not. > > In the regular "ceph osd crush tree", I don't see any cycles and it doesn't > seem > like there are the same buckets under two roots (I only have one root): > > root=default > rack=rack1 > host=ssd-1 > <OSDs> > rack=rack2 > host=ssd-2 > <OSDs> > rack=rack3 > host=ssd-3 > <OSDs> > rack=rack4 > host=hdd-1 > <OSDs> > > If I use `--show-shadow` and pick `host=ssd-1` as an example, I see: > > root=default~ssd > rack=rack1~ssd > host=ssd-1~ssd > <OSDs> > root=default~hdd > rack=rack1~hdd > host=ssd-1~hdd > <Nothing/no OSDs> > > Are (eg.) `host=ssd-1~ssd` and `host=ssd-1~hdd` treated as the same bucket? > > On Fri, Oct 1, 2021, at 14:27, Gregory Farnum wrote: > > It generally means, in CS terms, that you have a graph rather than a tree. > > > > In other words, you have two roots, or other crush buckets, which > > contain some of the same buckets/items underneath themselves. > > > > On Fri, Oct 1, 2021 at 9:43 AM Harry G. Coin <hgc...@gmail.com> wrote: > >> > >> I asked as well, it seems nobody on the list knows so far. > >> > >> > >> On 9/30/21 10:34 AM, Andrew Gunnerson wrote: > >> > Hello, > >> > > >> > I'm trying to figure out what overlapping roots entails with the default > >> > scale-down autoscaling profile in Ceph Pacific. My test setup involves a > >> > CRUSH > >> > map that looks like: > >> > > >> > ID=-1 | root=default > >> > ID=-58 | rack=rack1 > >> > ID=-70 | host=ssd-1 > >> > | <OSDs> > >> > ID=-61 | rack=rack2 > >> > ID=-55 | host=ssd-2 > >> > | <OSDs> > >> > ID=-62 | rack=rack3 > >> > ID=-52 | host=ssd-3 > >> > | <OSDs> > >> > ID=-63 | rack=rack4 > >> > ID=-19 | host=hdd-1 > >> > | <OSDs> > >> > | <15 more hosts> > >> > > >> > The CRUSH rules I created are: > >> > > >> > # Rack failure domain for SSDs > >> > ceph osd crush rule create-replicated replicated_ssd default rack > >> > ssd > >> > # Host failure domain for HDDs > >> > ceph osd crush rule create-replicated replicated_hdd default host > >> > hdd > >> > ceph osd erasure-code-profile set erasure_hdd ruleset k=3 m=2 > >> > crush-device-class=hdd crush-failure-domain=host > >> > > >> > and the pools are: > >> > > >> > Pool | CRUSH rule/profile | Overlapped roots > >> > error > >> > > >> > ---------------------------|--------------------|----------------------- > >> > device_health_metrics | replicated_rule | -1 (root=default) > >> > cephfs_metadata | replicated_ssd | -51 > >> > (root=default~ssd) > >> > cephfs_data_replicated_ssd | replicated_ssd | -51 > >> > (root=default~ssd) > >> > cephfs_data_replicated_hdd | replicated_hdd | -2 > >> > (root=default~hdd) > >> > cephfs_data_erasure_hdd | erasure_hdd | -1 (root=default) > >> > > >> > With this setup, the autoscaler is getting disabled in every pool with > >> > the > >> > following error: > >> > > >> > [pg_autoscaler WARNING root] pool <num> contains an overlapping > >> > root -<id>... skipping scaling > >> > > >> > There doesn't seem to be much documentation about overlapped roots. I > >> > think I'm > >> > fundamentally not understanding what it means. Does it mean that the > >> > autoscaler > >> > can't handle two different pools using OSDs under the same (shadow?) > >> > root in the > >> > CRUSH map? > >> > > >> > Is this setup something that's not possible using the scale-down > >> > autoscaler > >> > profile? It seems that the scale-up profile doesn't have a concept of > >> > overlapped > >> > roots. > >> > > >> > Thank you, > >> > Andrew > >> > _______________________________________________ > >> > ceph-users mailing list -- ceph-users@ceph.io > >> > To unsubscribe send an email to ceph-users-le...@ceph.io > >> _______________________________________________ > >> ceph-users mailing list -- ceph-users@ceph.io > >> To unsubscribe send an email to ceph-users-le...@ceph.io > >> > > > > _______________________________________________ > > ceph-users mailing list -- ceph-users@ceph.io > > To unsubscribe send an email to ceph-users-le...@ceph.io > _______________________________________________ > ceph-users mailing list -- ceph-users@ceph.io > To unsubscribe send an email to ceph-users-le...@ceph.io > _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-le...@ceph.io
