ceph/qa/suites/rbd/pwl-cache/tmpfs/4-cache-path.yaml
Ilya Dryomov 23759e0034 qa/suites/rbd: place cache file on tmpfs for xfstests
The RWL mode needs DAX and is dog slow otherwise -- qemu_xfstests.yaml
job always hits the 6 hour max_job_time limit.

As our tmpfs instance is limited and qemu_xfstests.yaml opens three
images at the same time, reduce the "big cache" size to 5G.  This facet
was added to iron out 32-bit head/tail pointer issues and 5G still does
the job there.

Going through the loop device is needed because tmpfs doesn't support
O_DIRECT.

Fixes: https://tracker.ceph.com/issues/55400
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
2022-06-03 19:20:37 +02:00

23 lines
868 B
YAML

overrides:
ceph:
conf:
client:
rbd_persistent_cache_path: /home/ubuntu/cephtest/rbd-pwl-cache
rbd_plugins: pwl_cache
tasks:
- exec:
client.0:
- "mkdir /home/ubuntu/cephtest/tmpfs"
- "mkdir /home/ubuntu/cephtest/rbd-pwl-cache"
- "sudo mount -t tmpfs -o size=20G tmpfs /home/ubuntu/cephtest/tmpfs"
- "truncate -s 20G /home/ubuntu/cephtest/tmpfs/loopfile"
- "mkfs.ext4 /home/ubuntu/cephtest/tmpfs/loopfile"
- "sudo mount -o loop /home/ubuntu/cephtest/tmpfs/loopfile /home/ubuntu/cephtest/rbd-pwl-cache"
- "sudo chmod 777 /home/ubuntu/cephtest/rbd-pwl-cache"
- exec_on_cleanup:
client.0:
- "sudo umount /home/ubuntu/cephtest/rbd-pwl-cache"
- "sudo umount /home/ubuntu/cephtest/tmpfs"
- "rm -rf /home/ubuntu/cephtest/rbd-pwl-cache"
- "rm -rf /home/ubuntu/cephtest/tmpfs"