Merge back what we have in the (open)SUSE ceph spec file for JUnit.
Add missing Requires and the package is named junit4 on some SUSE
versions.
Signed-off-by: Danny Al-Gaaf <danny.al-gaaf@bisect.de>
This patch adds two buildrequires to the ceph.spec file, that are needed
to build the rpms under Fedora. Danny Al-Gaaf commented that the
snappy-devel dependency should actually be added to the leveldb-devel
package. I will try to get that fixed too, in the mean time, this patch
does make sure Ceph builds on Fedora.
Signed-off-by: Erik Logtenberg <erik@logtenberg.eu>
The iterator should be compared against the end() function of
the same iter() from region_conn_map.
CID 1058791 (#1 of 1): Invalid iterator comparison (MISMATCHED_ITERATOR)
mismatched_comparison: Comparing "iter" from "this->region_conn_map" to
"this->zone_conn_map.end()" from "this->zone_conn_map".
Signed-off-by: Danny Al-Gaaf <danny.al-gaaf@bisect.de>
[Also move into a separatate test script; validate result -sage]
Signed-off-by: Dan Mick <dan.mick@inktank.com>
Signed-off-by: Sage Weil <sage@inktank.com>
Fixes regression in daemon commands with arguments; also resolves
reported bug with existing daemon code for arguments with
embedded spaces.
Fixes: #5503Fixes: #5800
Signed-off-by: Dan Mick <dan.mick@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Fixes: 5808
We cannot call get_bucket_instance_info() at that point,
as the bucket structure wasn't initialized, so we don't
have the bucket instance location information. Just calling
get_bucket_info().
Signed-off-by: Yehuda Sadeh <yehuda@inktank.com>
Reviewed-by: Greg Farnum <greg@inktank.com>
Otherwise, we might leave a session attached to the
CommandOp for an down OSD. handle_osd_map will then
delete the session for the down OSD. tick() will then
attempt to follow the invalid pointer to find a
connection over which to send a MPing.
Fixes: #5798
Signed-off-by: Samuel Just <sam.just@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
If someone is syncing from us and there is an election, they currently get
reset and have to restart their sync. This can lead to situations where
they can never finish, e.g., when the load from them syncing makes us time
out commits and call elections.
There is nothing that changes during bootstrap that would prevent a sync
from proceeding. The only time we need to stop providing is when we
ourselves decide to sync from someone else; modify that reset call to
reset provider state. All other resets become requester resets.
Signed-off-by: Sage Weil <sage@inktank.com>
Fixes: #5793
Beforehand all remote copies were going to the master region
which was awfully wrong.
Signed-off-by: Yehuda Sadeh <yehuda@inktank.com>
Reviewed-by: Greg Farnum <greg@inktank.com>
Fixes: #5789
This was fixed before, however, might have been lost due to
recent refactoring + merge.
Signed-off-by: Yehuda Sadeh <yehuda@inktank.com>
Reviewed-by: Greg Farnum <greg@inktank.com>
Need to do the action through the bucket instance handler
and not through the bucket handler, otherwise it's wrongly
recorded (and wrongly replayed, ouch).
Fixes: #5791
Signed-off-by: Yehuda Sadeh <yehuda@inktank.com>
Reviewed-by: Greg Farnum <greg@inktank.com>
We now keep the bucket instance oid in rgw_bucket. The reason
we need it is that the bucket might have been created before
the entrypoint / bucket instance separation.
Fixes: #5790
Signed-off-by: Yehuda Sadeh <yehuda@inktank.com>
Reviewed-by: Greg Farnum <greg@inktank.com>
Use the correct type for the dumpcontents arg. Fixes the dump_pools_json
output and avoids these errors:
2013-07-29 13:09:14.089188 7fa0c5d21700 -1 0x7fa0c5d1e7a8
2013-07-29 13:09:16.306560 7fa0c5d21700 -1 bad boost::get: key dumpcontents is not type std::vector<std::string, std::allocator<std::string> >
2013-07-29 13:09:16.317104 7fa0c5d21700 -1 0x7fa0c5d1e7a8
2013-07-29 13:09:16.317136 7fa0c5d21700 -1 bad boost::get: key dumpcontents is not type std::vector<std::string, std::allocator<std::string> >
Fixes: #5786
Signed-off-by: Sage Weil <sage@inktank.com>
When trying to establish if the old acting set is either empty or
smaller than the min_size of the osdmap,
pg_interval_t::check_new_interval compares with the min_size of the
new osdmap. Since the goal is to try to determine if the previous
interval may have been writeable, it should not enter the if when
there were not enough osds in the acting set ( i.e. < min_size ). But
it may enter it anyway if min_size was decremented in the new osdmap.
A complete set of unit tests were added to cover the logic of
check_new_interval. The parameters are prepared to describe a
situation where the function returns false (i.e. no new
interval). Each case is described in a separate bloc that introduces
the minimal changes to demonstrate the intended test case.
Because a number of cases have the same output while implementing a
different logic, the debug output is parsed to differentiate between them.
A test case demonstrating the problem ( check_new_interval must
compare old acting with old osdmap ) is added, with a link to the bug
number for future reference. The problem is fixed. The text of two
debug messages are slightly changed to make the maintenance of the
test that match them easier.
http://tracker.ceph.com/issues/5780 refs #5780
Signed-off-by: Loic Dachary <loic@dachary.org>
Reviewed-by: Sage Weil <sage@inktank.com>
Reviewed-by: Samuel Just <sam.just@inktank.com>
If N op_tp threads are configured, and recovery_max_active
is set to a sufficiently large number, all N op_tp threads
might grab a MOSDPGPush op off of the queue for the same PG.
The last thread to get the lock will have waited
N*time_to_handle_push before completing its item and pinging
the heartbeat timeout. If that time exceeds the timeout
and there are enough ops waiting, each thread subsequently
will end up exceeding the timeout before completeing an
item preventing the OSD from heartbeating indefinitely.
We prevent this by suspending the timeout while we try to
get the PG lock. Even if we do block for an excessive
period of time attempting to get the lock, hopefully,
the thread holding the lock will cause the threadpool
to time out.
Signed-off-by: Samuel Just <sam.just@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
If there is a gap in our map history, get the full range of maps that
the mon has. Make sure the first one is a full map.
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Samuel Just <sam.just@inktank.com>
If we have map 250, and monitor's first is 251, but sends 260, we can
request the intervening range.
Fixes: #5784
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Samuel Just <sam.just@inktank.com>
