RepoMirrors/ceph
1
0
Fork 0
mirror of https://github.com/ceph/ceph synced 2024-12-18 09:25:49 +00:00
Code Issues Packages Projects Releases Wiki Activity

qa: simplify TestJournalRepair

Browse Source 
This was sending lots of metadata ops to MDSs to persuade
them to migrate some subtrees, but that was flaky.  Use
the shiny new rank pinning functionality instead.

Signed-off-by: John Spray <john.spray@redhat.com>
This commit is contained in:
John Spray 2017-05-15 17:23:42 -04:00
parent 880dde8e47
commit 2350555fe5
2 changed files with 22 additions and 55 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
qa/tasks/cephfs/mount.py
View File

@ -436,36 +436,6 @@ class CephFSMount(object):
self._kill_background(p)
self.background_procs.remove(p)
def spam_dir_background(self, path):
"""
Create directory `path` and do lots of metadata operations
in it until further notice.
"""
assert(self.is_mounted())
abs_path = os.path.join(self.mountpoint, path)
pyscript = dedent("""
import sys
import time
import os
abs_path = "{abs_path}"
if not os.path.exists(abs_path):
os.makedirs(abs_path)
n = 0
while True:
file_path = os.path.join(abs_path, "tmp%d" % n)
f = open(file_path, 'w')
f.close()
n = n + 1
""").format(abs_path=abs_path)
rproc = self._run_python(pyscript)
self.background_procs.append(rproc)
return rproc
def get_global_id(self):
raise NotImplementedError()

47
qa/tasks/cephfs/test_journal_repair.py
View File

@ -173,26 +173,33 @@ class TestJournalRepair(CephFSTestCase):
self.mds_cluster.mds_stop(unneeded_mds)
self.mds_cluster.mds_fail(unneeded_mds)
# Do a bunch of I/O such that at least some will hit the second MDS: create
# lots of directories so that the balancer should find it easy to make a decision
# to allocate some of them to the second mds.
spammers = []
for n in range(0, 16):
dir_name = "spam_{0}".format(n)
spammers.append(self.mount_a.spam_dir_background(dir_name))
# Create a dir on each rank
self.mount_a.run_shell(["mkdir", "alpha"])
self.mount_a.run_shell(["mkdir", "bravo"])
self.mount_a.setfattr("alpha/", "ceph.dir.pin", "0")
self.mount_a.setfattr("bravo/", "ceph.dir.pin", "1")
def subtrees_assigned():
got_subtrees = self.fs.mds_asok(["get", "subtrees"], mds_id=active_mds_names[0])
rank_1_count = len([s for s in got_subtrees if s['auth_first'] == 1])
# Greater than 1, because there is typically 1 for ~mds1, and once it
# has been assigned something in addition to that it means it has been
# assigned a "real" subtree.
return rank_1_count > 1
for s in got_subtrees:
if s['dir']['path'] == '/bravo':
if s['auth_first'] == 1:
return True
else:
# Should not happen
raise RuntimeError("/bravo is subtree but not rank 1!")
# We are waiting for the MDS to respond to hot directories, which
# is not guaranteed to happen at a particular time, so a lengthy timeout here.
self.wait_until_true(subtrees_assigned, 600)
return False
# Ensure the pinning has taken effect and the /bravo dir is now
# migrated to rank 1.
self.wait_until_true(subtrees_assigned, 30)
# Do some IO (this should be split across ranks according to
# the rank-pinned dirs)
self.mount_a.create_n_files("alpha/file", 1000)
self.mount_a.create_n_files("bravo/file", 1000)
# Flush the journals so that we have some backing store data
# belonging to one MDS, and some to the other MDS.
@ -229,16 +236,6 @@ class TestJournalRepair(CephFSTestCase):
# killing the mount also means killing the node.
pass
log.info("Terminating spammer processes...")
for spammer_proc in spammers:
spammer_proc.stdin.close()
try:
spammer_proc.wait()
except (CommandFailedError, ConnectionLostError):
# The ConnectionLostError case is for kernel client, where
# killing the mount also means killing the node.
pass
# See that the second MDS will crash when it starts and tries to
# acquire rank 1
damaged_id = active_mds_names[1]