librbd: bug fixes for optional data pool support #11960
Conversation
| @@ -146,7 +146,7 @@ bool CopyupRequest::send_copyup() { | |||
| ldout(m_ictx->cct, 20) << __func__ << " " << this << " copyup with " | |||
| << "empty snapshot context" << dendl; | |||
| librados::AioCompletion *comp = util::create_rados_safe_callback(this); | |||
| r = m_ictx->md_ctx.aio_operate(m_oid, comp, ©up_op, 0, snaps); | |||
| r = m_ictx->data_ctx.aio_operate(m_oid, comp, ©up_op, 0, snaps); | |||
There was a problem hiding this comment.
Good catch, but not 100% correct. You will need to use a new "data_ctx" here -- it needs to be pointing to the correct pool but its snapshot context needs to be empty (like md_ctx).
There was a problem hiding this comment.
@dillaman I understand what you mean and see why it's required. I tried to see what's happens in the OSD when m_ictx->data_ctx is used and it seems that the SnapContext is empty in this case too (empty snaps during rbd.copyup and list of SnapContext for the actual (write) call):
2016-11-15 14:26:44.078037 7fc8807cc700 10 osd.2 pg_epoch: 44 pg[3.e( v 32'11 (0'0,32'11] local-les=35 n=5 ec=11 les/c/f 35/35/0 34/34/34) [2,0,1] r=0 lpr=34 crt=32'11 lcod 0'0 mlcod 0'0 active+clean] execute_ctx 0x560cf27f5c00
2016-11-15 14:26:44.078049 7fc8807cc700 10 osd.2 pg_epoch: 44 pg[3.e( v 32'11 (0'0,32'11] local-les=35 n=5 ec=11 les/c/f 35/35/0 34/34/34) [2,0,1] r=0 lpr=34 crt=32'11 lcod 0'0 mlcod 0'0 active+clean] do_op 3:71a2fc68:::rbd_data.85382ae8944a.0000000000000000:head [call rbd.copyup] ov 0'0 av 44'12 snapc 0=[] snapset 0=[]:[]
2016-11-15 14:26:44.220713 7fc8807cc700 10 osd.2 pg_epoch: 44 pg[3.e( v 44'12 (0'0,44'12] local-les=35 n=6 ec=11 les/c/f 35/35/0 34/34/34) [2,0,1] r=0 lpr=34 crt=32'11 lcod 32'11 mlcod 32'11 active+clean] execute_ctx 0x560cf2a8f500
2016-11-15 14:26:44.220726 7fc8807cc700 10 osd.2 pg_epoch: 44 pg[3.e( v 44'12 (0'0,44'12] local-les=35 n=6 ec=11 les/c/f 35/35/0 34/34/34) [2,0,1] r=0 lpr=34 crt=32'11 lcod 32'11 mlcod 32'11 active+clean] do_op 3:71a2fc68:::rbd_data.85382ae8944a.0000000000000000:head [set-alloc-hint object_size 4194304 write_size 4194304,write 0~0] ov 44'12 av 44'13 snapc 14=[14,13,12,11,10,f,e,d] snapset 0=[]:[]+head
Note that I've made the changes as suggested though.
vshankar
force-pushed
the
wip-librbd-ec-support
branch
2 times, most recently
from
6820fbc
to
9d1b605
Compare
|
Related ceph-qa-suite changes: ceph/ceph-qa-suite#1235 |
|
retest this please |
vshankar
changed the title
dillaman
changed the title
|
The python test needs to be updated to understand that the data pool feature bit will be enabled if the pool is specified: The import_export.sh test is directly scanning the "rbd" pool for objects and failing: Seg fault of ceph_test_librbd: |
|
jenkins test this please (eio ignored in master) |
|
jenkins test this please (osd-scrub-snaps.sh) |
|
jenkins test this please (eio now ignored in master) |
close() was never called for the passed in IoCtx which could probably result in an IoCtx leak if the original IoCtx was a valid pool context allocated earlier. Its kind of better to do it here rather than to leave the destruction on the caller for better (or cleaner) common case handling. Signed-off-by: Venky Shankar <vshankar@redhat.com>
Signed-off-by: Venky Shankar <vshankar@redhat.com>
Signed-off-by: Venky Shankar <vshankar@redhat.com>
Signed-off-by: Venky Shankar <vshankar@redhat.com>
Signed-off-by: Venky Shankar <vshankar@redhat.com>
Signed-off-by: Venky Shankar <vshankar@redhat.com>
Signed-off-by: Venky Shankar <vshankar@redhat.com>
Signed-off-by: Venky Shankar <vshankar@redhat.com>
vshankar
force-pushed
the
wip-librbd-ec-support
branch
from
9d1b605
to
3a91d7b
Compare
|
Changes in this patchset:
|
|
Re-running subset on smithi nodes to remove all the false failures due to VPS resource constraints: http://pulpito.ceph.com/jdillaman-2016-11-21_08:55:50-rbd-wip-ec-partial-overwrites---basic-smithi/ |
Thanks @dillaman |
This PR enables running librbd with an optional data pool. Relevant support was already added a while ago in the form of an additional "--datapool" option while creating RBD images, but there were places in librbd that were not fully functional to fully support using optional data pool. Furthermore, running teuthology tests revealed a bunch of failed tests which this PR aims to resolve.
NOTE: teuthology runs are being scheduled with this PR on top of Sam's ec-partial-overwrite branch.