I've noticed sometimes try_push_job_info() fails because of server load
issues. It should try more than once (and now does).
Signed-off-by: Zack Cerza <zack.cerza@inktank.com>
I didn't love the way safe_while was encouraged to be used and it didn't
fit right with the new no-raising behavior. Now it's encouraged to be
used like this:
with safe_while() as proceed:
while proceed():
do_things()
Signed-off-by: Zack Cerza <zack.cerza@inktank.com>
If we're going to embark on a mission to rid ourselves of
infinitely-looping while loops, it seems smart to start marking the ones
we've fixed in order to make grepping for unfixed loops easier.
Signed-off-by: Zack Cerza <zack.cerza@inktank.com>
The + means we should concatenate everything in the directory. Do that.
This was totally broken before (and unused until now).
Signed-off-by: Sage Weil <sage@inktank.com>
For ops that default to 0, only add arguments for them if they are
specified in the task config. This lets us use the same task across
ceph versions, even if the older version does not support new op
types, like append on dumpling.
Signed-off-by: Josh Durgin <josh.durgin@inktank.com>
The timeout is configurable, but defaults to six hours. It's implemented
by using the 'timeout' command on the remote host.
Signed-off-by: Zack Cerza <zack.cerza@inktank.com>
* the crush ruleset and the pool create parameters must be identicals
k=2 m= 1
* the --property argument is invalid
* the failure domain is ignored on pool create
Signed-off-by: Loic Dachary <loic@dachary.org>
Some kernels (primarily Debian distro kernels) do not support
setting kdb. Rather than having the entire test fail.
Signed-off-by: Sandon Van Ness <sandon@inktank.com>
Use --limit to limit the number of jobs being scheduled during
teuthology-suite. Also can be used with schedule_suite.sh via the
10th argument.
Signed-off-by: Sandon Van Ness <sandon@inktank.com>
Signed-off-by: Yuri Weinstein <yuri.weinstein@inktank.com>
get_mds_status returns None for things it can't see,
so have to check for Noneness on all its outputs.
Signed-off-by: John Spray <john.spray@inktank.com>
This simplifies the code to make MdsTrash be a greenlet
(as it logically is) rather than encapsulating one that
gets started in __init__ (spawning threads in constructors
is evil).
With this done, do_thrash is called from _run inside an
exception handler that will give us full tracebacks
if something bad happens.
Signed-off-by: John Spray <john.spray@inktank.com>
