This was causing librados to unblock after the ACK on unwatch, which meant
that librbd users raced and tried to delete the image before the unwatch
change was committed..and got EBUSY. See #3958.
The watch operation has a similar problem.
Signed-off-by: Sage Weil <sage@inktank.com>
This ensures we release our in-progress recovery counters, which prevents
recovery from getting blocked indefinitely when a pool removal races with
recovery ops.
Fixes: #4217
Backport: bobtail
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Samuel Just <sam.just@inktank.com>
Use qi to parse a strictly formatted set of key/value pairs. Be picky
about whitespace. Any subset of recognized keys is allowed. Parse the
same set of keys as the ceph.*.layout.* vxattrs.
Signed-off-by: Sage Weil <sage@inktank.com>
Allow user to control the minimum level to go to syslog for the client-
and server-side submission paths for the cluster log, along with the syslog
'facility'. See syslog(3) man page.
Also move the level checks into a LogEntry method.
Closes: #3704
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Joao Luis <joao.luis@inktank.com>
Methods called by write_if_dirty() (get_osdmap()) assert that the pg
is locked.
Backport: bobtail
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Samuel Just <sam.just@inktank.com>
Fixes: #4177
Backport: bobtail
Listing multipart uploads had a typo, and was requiring the
wrong resource (uploadId instead of uploads).
Signed-off-by: Yehuda Sadeh <yehuda@inktank.com>
Reviewed-by: Greg Farnum <greg@inktank.com>
Fixes: #4150
Backport: bobtail
When object copied into itself, object will not be fully copied: tail
reference count stays the same, head part is rewritten.
Signed-off-by: Yehuda Sadeh <yehuda@inktank.com>
Reviewed-by: Greg Farnum <greg@inktank.com>
The existing code will overlay the placement of PGs from pools because
it simply adds the ps to the pool as the CRUSH input. That means that
the layout/placement for pg 0.10 == 1.9 == 2.8 == 3.7 == 4.6 == ...,
which is not optimal.
Instead, use hash(ps, poolid). The avoids the initial problem of
the sequence being adjacent to other pools. It also avoids the (small)
possibility that hash(poolid) will drop us somewhere in the output
number space where our sequence of outputs overlaps with some other
pool; instead, out output sequence will be a fully random (for a well-
behaved hash).
Use the multi-input hash functions used by CRUSH for this.
Default to the legacy behavior for now. We won't enable this until
deployed systems and kernel code catch up.
Fixes: #4128
Signed-off-by: Sage Weil <sage@inktank.com>
The '! command' doesn't fail properly, even with -e, in bash (wtf!).
Also, the last pool deletion command succeeds because the pool
'--yes-i-really-really-mean-it' doesn't exist. So drop that test.
Signed-off-by: Sage Weil <sage@inktank.com>
Small tweaks to the hadoop-internal test
to better use existing environment varaibles.
Signed-off-by: Joe Buck <jbbuck@gmail.com>
Reviewed-by: Noah Watkins <noahwatkins@gmail.com>
In the case of a divergent object prior to log.tail,
last_complete may end up before log.tail.
Backport: bobtail
Fixes#4174
Signed-off-by: Samuel Just <sam.just@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
We could have a sequence like:
- op1
- notify
- op2
in the finished queue. Op1 gets put on waiting_for_pg, the notify
creates the pg and requeues op1 (and the end), op2 is handled, and
finally op1 is handled. That breaks ordering; see #2947.
Instead, when we wake up a pg, queue the waiting messages at the front
of the dispatch queue.
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Samuel Just <sam.just@inktank.com>
Pull items off the finished queue on at a time. In certain cases, an
event may result in new items betting added to the finished queue that
will be put at the *front* instead of the back. See latest incarnation
of #2947.
Note that this is a significant changed in behavior in that we can
theoretically starve if an event keeps resulting in new events getting
generated. Beware!
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Samuel Just <sam.just@inktank.com>
warning: osd/osd_types.cc:1716:76: format '%lu' expects argument of type 'long unsigned int', but argument 5 has type 'version_t {aka long long unsigned int}' [-Wformat]
warning: osd/osd_types.cc:1716:76: format '%lu' expects argument of type 'long unsigned int', but argument 5 has type 'version_t {aka long long unsigned int}' [-Wformat]
Signed-off-by: Sage Weil <sage@inktank.com>
I'm not sure why we wouldn't. Also, this makes this test work without
annoying plumbing to pass the explicit path through.
Signed-off-by: Sage Weil <sage@inktank.com>
udev runs blkid on device close, and other such nonsense that can
make unmap fail with EBUSY. Settle before we unmap to avoid this if
possible. See #4183.
Closes: #4186
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Alex Elder <elder@inktank.com>
Changing the hadoop-internal tests to use the
newly added $TESTDIR environment variable.
Also, removed unneeded variables.
Signed-off-by: Joe Buck <jbbuck@gmail.com>
Reviewed-by: Sam Lang <sam.lang@inktank.com>
