Merge pull request #955 from shylesh/tracker13531

Rados: Tracker-13531-resolve stuck peering

Reviewed-by: Samuel Just <sjust@redhat.com>

suites/rados/singleton/all/resolve_stuck_peering.yaml

@@ -0,0 +1,10 @@
+roles:
+- [mon.0]
+- [osd.0, osd.1, osd.2, client.0]
+
+tasks:
+- install:
+- ceph:
+    fs: xfs
+- resolve_stuck_peering:
+

tasks/resolve_stuck_peering.py

@@ -0,0 +1,108 @@
+"""
+Resolve stuck peering
+"""
+import logging
+import time
+
+from teuthology import misc as teuthology
+from util.rados import rados
+
+log = logging.getLogger(__name__)
+
+def task(ctx, config):
+	"""
+	Test handling resolve stuck peering
+
+	requires 3 osds on a single test node
+	"""
+	if config is None:
+		config = {}
+	assert isinstance(config, dict), \
+	       'Resolve stuck peering only accepts a dict for config'
+
+	while len(ctx.manager.get_osd_status()['up']) < 3:
+		time.sleep(10)
+
+
+	ctx.manager.wait_for_clean()
+
+	dummyfile = '/etc/fstab'
+	dummyfile1 = '/etc/resolv.conf'
+
+	#create 1 PG pool
+	pool='foo'
+	log.info('creating pool foo')
+	ctx.manager.raw_cluster_cmd('osd', 'pool', 'create', '%s' % pool, '1')
+
+	#set min_size of the pool to 1
+	#so that we can continue with I/O
+	#when 2 osds are down
+	ctx.manager.set_pool_property(pool, "min_size", 1)
+
+	osds = [0, 1, 2]
+
+	primary = ctx.manager.get_pg_primary('foo', 0)
+	log.info("primary osd is %d", primary)
+
+	others = list(osds)
+	others.remove(primary)
+
+	log.info('writing initial objects')
+	first_mon = teuthology.get_first_mon(ctx, config)
+	(mon,) = ctx.cluster.only(first_mon).remotes.iterkeys()
+	#create few objects
+	for i in range(100):
+		rados(ctx, mon, ['-p', 'foo', 'put', 'existing_%d' % i, dummyfile])
+
+	ctx.manager.wait_for_clean()
+
+	#kill other osds except primary
+	log.info('killing other osds except primary')
+	for i in others:
+		ctx.manager.kill_osd(i)
+	for i in others:
+		ctx.manager.mark_down_osd(i)
+
+
+	for i in range(100):
+		rados(ctx, mon, ['-p', 'foo', 'put', 'new_%d' % i, dummyfile1])
+
+	#kill primary osd
+	ctx.manager.kill_osd(primary)
+	ctx.manager.mark_down_osd(primary)
+
+	#revive other 2 osds
+	for i in others:
+		ctx.manager.revive_osd(i)
+
+	#make sure that pg is down
+	#Assuming pg number for single pg pool will start from 0
+	pgnum=0
+	pgstr = ctx.manager.get_pgid(pool, pgnum)
+	stats = ctx.manager.get_single_pg_stats(pgstr)
+	print stats['state']
+
+	timeout=60
+	start=time.time()
+
+	while 'down' not in stats['state']:
+		assert time.time() - start < timeout, \
+			'failed to reach down state before timeout expired'
+
+	#mark primary as lost
+	ctx.manager.raw_cluster_cmd('osd', 'lost', '%d' % primary,\
+					 '--yes-i-really-mean-it')
+
+
+	#expect the pg status to be active+undersized+degraded
+	#pg should recover and become active+clean within timeout
+	stats = ctx.manager.get_single_pg_stats(pgstr)
+	print stats['state']
+
+	timeout=10
+	start=time.time()
+
+	while ctx.manager.get_num_down():
+		assert time.time() - start < timeout, \
+                        'failed to recover before timeout expired'
+