The new monitor store does not create the data directory on --mkfs. We
must create it instead, much like what happens with the osds.
Signed-off-by: Joao Eduardo Luis <joao.luis@inktank.com>
We now add a new option 'thrash-many' that by being set to true will break
the default behaviour of killing only one monitor at a time. Instead,
this option will select up to the maximum number of killable monitors to
kill in each round.
We also add a new 'maintain-quorum' option that will limit the amount of
monitors that can be killed in each thrashing round. If set to true, this
option will limit the amount of killable monitors up to (n/2-1). This
means that if we are running a configuration that only has up to two
configured monitors, if 'maintain-quorum' is set to true, this task won't
run as there are no killable monitors -- in such a scenario, this option
should be set to false.
Furthermore, if 'store-thrash' is set to true, then 'maintain-quorum' must
also be set to true, as we cannot let the task to thrash all the monitor
stores, or we wouldn't be able to sync from other monitors, nor can we
let quorum be dropped, or we won't be able to resync our way into quorum.
Signed-off-by: Joao Eduardo Luis <joao.luis@inktank.com>
This patch introduces an option to thrash a monitor store when we thrash
the monitors, as well as a 'store-thrash-probability' option (defaulting
to 50%).
We also took this opportunity to introduce a new 'seed' option, that ought
to allow a given run of this task to be reproducible. This might come in
hand when attempting to reproduce a given behavior that would otherwise
be randomly triggered.
You should note that while the 'seed' option will indeed mimic past
behaviors, this only applies to a past behavior of this task: other tasks
are not affected by this value, nor are any workunits or even ceph daemons.
Signed-off-by: Joao Eduardo Luis <joao.luis@inktank.com>
- call apt separately for each package; it will error out annoyingly if
there is one in the list not in the APT sources.
- use dpkg with appropriate force to clean up broken half-installs.
Signed-off-by: Sage Weil <sage@inktank.com>
Otherwise we get stuck in a loop if an osd crashes unexpectedly, the
task never fails, and we don't collect all the evidence.
Signed-off-by: Sage Weil <sage@inktank.com>
This is probably temporary. It's simpler than adding the task to every
job in the suite. We'll want to do that later when we want to test
alternative install methods (like ceph-deploy's install function).
Signed-off-by: Sage Weil <sage@inktank.com>
Some command-line tools need to reference the path
to the test directory, which is created at run-time.
We export this as TESTDIR
Signed-off-by: Joe Buck <jbbuck@gmail.com>
Reviewed-by: Sam Lang <sam.lang@inktank.com>
We need to switch around how these are compressed and pulled, since they
aren't in the regular archive dir anymore.
Signed-off-by: Sage Weil <sage@inktank.com>
This required reordering the cluster setup so that we do the ceph-osd
--mkfs --mkkey prior to gathering keys and initializing the monitors.
Also, run daemons as root.
Signed-off-by: Sage Weil <sage@inktank.com>
