Immediately after the Firefly installation, create an erasure code pool
behind a replicated cache pool. Run deep-scrub on all OSD while a rados
task runs. After upgrading half of the cluster (MON and OSD), run a
rados task again also deep-scrub in parallel.
http://tracker.ceph.com/issues/11054Fixes: #11054
Signed-off-by: Loic Dachary <loic@dachary.org>
Rename sub-directories of firefly-x/stress-split-erasure-code make room
for a workload immediately after the installation, i.e. number 2.
Signed-off-by: Loic Dachary <loic@dachary.org>
Mimic upgrade/firefly-x/stress-split only with thrashosd configured with
min_in: 4 for k=3,m=1 . All erasure code workloads are moved in this
dedicated suite.
http://tracker.ceph.com/issues/9652Fixes: #9652
Signed-off-by: Loic Dachary <loic-201408@dachary.org>
The ec-rados-default.yaml was linked to suites as if it was a task
although it is a workload intended for parallel upgrade tests.
The ec-rados-plugin=jerasure-k=2-m=1.yaml task is defined and used
instead, where relevant.
http://tracker.ceph.com/issues/9549Fixes: #9549
Signed-off-by: Loic Dachary <loic-201408@dachary.org>
And replace the ec-readwrite.yaml custom workloads with links to the
workloads found in the erasure-code directory.
Signed-off-by: Loic Dachary <loic-201408@dachary.org>
Sage found:
"... supposed to have half dumpling, half x osds. but the steps that upgrade and restart the mons upgrade the packages on the second host (which should remain dumpling w/ osd 3,4,5)"
Signed-off-by: Yuri Weinstein <yuri.weinstein@inktank.com>
Because rgw.py iterates over it to run the rgw server. If it is removed
the rgw servers are not run and all fails.
Signed-off-by: Loic Dachary <loic@dachary.org>
Globally overriding the rgw idle_timeout is not possible because it it
needs to be done on a per client.0, client.1, etc. basis. Add the
default_idle_timeout key to the rgw config : it defaults to the
previously hardcoded default (30) and can be changed via the override.
The existing tasks that were previously overriding the idle_timeout on a
per client basis are changed to use the default_idle_timeout instead for
consistency and to allow a global override.
Signed-off-by: Loic Dachary <loic@dachary.org>
It is a s/dumpling/firefly/ mostly. With the exception of
rados/test-upgrade-firefly.sh which is dumpling specific.
The 'mon warn on legacy crush tunables: false' are dropped because they
no longer apply to firefly.
Signed-off-by: Loic Dachary <loic@dachary.org>
