task/: add tests for ec and rep mark_unfound_lost delete

Signed-off-by: Samuel Just <sam.just@inktank.com>

diff --git a/teuthology/task/ceph_manager.py b/teuthology/task/ceph_manager.py
index 3952950b2d2ed7c1efab13157d44fbee5b63317c..481c2173c96bd1a3dae6f313ec63a52a205c7754 100644 (file)
--- a/teuthology/task/ceph_manager.py
+++ b/teuthology/task/ceph_manager.py
@@ -681,7 +681,7 @@ class CephManager:
         self.log(status)
         return status['pgmap']['num_pgs']
 
-    def create_pool_with_unique_name(self, pg_num=16, ec_pool=False):
+    def create_pool_with_unique_name(self, pg_num=16, ec_pool=False, ec_m=1, ec_k=2):
         """
         Create a pool named unique_pool_X where X is unique.
         """
@@ -689,10 +689,15 @@ class CephManager:
         with self.lock:
             name = "unique_pool_%s" % (str(self.next_pool_id),)
             self.next_pool_id += 1
-            self.create_pool(name, pg_num, ec_pool=ec_pool)
+            self.create_pool(
+                name,
+                pg_num,
+                ec_pool=ec_pool,
+                ec_m=ec_m,
+                ec_k=ec_k)
         return name
 
-    def create_pool(self, pool_name, pg_num=16, ec_pool=False):
+    def create_pool(self, pool_name, pg_num=16, ec_pool=False, ec_m=1, ec_k=2):
         """
         Create a pool named from the pool_name parameter.
         :param pool_name: name of the pool being created.
@@ -705,7 +710,7 @@ class CephManager:
             self.log("creating pool_name %s"%(pool_name,))
             if ec_pool and not self.created_erasure_pool:
                 self.created_erasure_pool = True
-                self.raw_cluster_cmd('osd', 'erasure-code-profile', 'set', 'teuthologyprofile', 'ruleset-failure-domain=osd', 'm=1', 'k=2')
+                self.raw_cluster_cmd('osd', 'erasure-code-profile', 'set', 'teuthologyprofile', 'ruleset-failure-domain=osd', 'm='+str(ec_m), 'k='+str(ec_k))
 
             if ec_pool:
                 self.raw_cluster_cmd('osd', 'pool', 'create', pool_name, str(pg_num), str(pg_num), 'erasure', 'teuthologyprofile')

diff --git a/teuthology/task/ec_lost_unfound.py b/teuthology/task/ec_lost_unfound.py
new file mode 100644 (file)
index 0000000..25bac6c
--- /dev/null
+++ b/teuthology/task/ec_lost_unfound.py
@@ -0,0 +1,134 @@
+"""
+Lost_unfound
+"""
+import logging
+import ceph_manager
+from teuthology import misc as teuthology
+from teuthology.task_util.rados import rados
+
+log = logging.getLogger(__name__)
+
+def task(ctx, config):
+    """
+    Test handling of lost objects on an ec pool.
+
+    A pretty rigid cluster is brought up andtested by this task
+    """
+    if config is None:
+        config = {}
+    assert isinstance(config, dict), \
+        'lost_unfound task only accepts a dict for configuration'
+    first_mon = teuthology.get_first_mon(ctx, config)
+    (mon,) = ctx.cluster.only(first_mon).remotes.iterkeys()
+
+    manager = ceph_manager.CephManager(
+        mon,
+        ctx=ctx,
+        logger=log.getChild('ceph_manager'),
+        )
+
+    manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
+    manager.raw_cluster_cmd('tell', 'osd.1', 'flush_pg_stats')
+    manager.raw_cluster_cmd('tell', 'osd.2', 'flush_pg_stats')
+    manager.raw_cluster_cmd('tell', 'osd.3', 'flush_pg_stats')
+    manager.wait_for_clean()
+
+
+    pool = manager.create_pool_with_unique_name(
+        ec_pool=True,
+        ec_m=2,
+        ec_k=2)
+
+    # something that is always there
+    dummyfile = '/etc/fstab'
+
+    # kludge to make sure they get a map
+    rados(ctx, mon, ['-p', pool, 'put', 'dummy', dummyfile])
+
+    manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
+    manager.raw_cluster_cmd('tell', 'osd.1', 'flush_pg_stats')
+    manager.wait_for_recovery()
+
+    # create old objects
+    for f in range(1, 10):
+        rados(ctx, mon, ['-p', pool, 'put', 'existing_%d' % f, dummyfile])
+        rados(ctx, mon, ['-p', pool, 'put', 'existed_%d' % f, dummyfile])
+        rados(ctx, mon, ['-p', pool, 'rm', 'existed_%d' % f])
+
+    # delay recovery, and make the pg log very long (to prevent backfill)
+    manager.raw_cluster_cmd(
+            'tell', 'osd.1',
+            'injectargs',
+            '--osd-recovery-delay-start 1000 --osd-min-pg-log-entries 100000000'
+            )
+
+    manager.kill_osd(0)
+    manager.mark_down_osd(0)
+    manager.kill_osd(3)
+    manager.mark_down_osd(3)
+    
+    for f in range(1, 10):
+        rados(ctx, mon, ['-p', pool, 'put', 'new_%d' % f, dummyfile])
+        rados(ctx, mon, ['-p', pool, 'put', 'existed_%d' % f, dummyfile])
+        rados(ctx, mon, ['-p', pool, 'put', 'existing_%d' % f, dummyfile])
+
+    # take out osd.1 and a necessary shard of those objects.
+    manager.kill_osd(1)
+    manager.mark_down_osd(1)
+    manager.raw_cluster_cmd('osd', 'lost', '1', '--yes-i-really-mean-it')
+    manager.revive_osd(0)
+    manager.wait_till_osd_is_up(0)
+    manager.revive_osd(3)
+    manager.wait_till_osd_is_up(3)
+
+    manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
+    manager.raw_cluster_cmd('tell', 'osd.2', 'flush_pg_stats')
+    manager.raw_cluster_cmd('tell', 'osd.3', 'flush_pg_stats')
+    manager.wait_till_active()
+    manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
+    manager.raw_cluster_cmd('tell', 'osd.2', 'flush_pg_stats')
+    manager.raw_cluster_cmd('tell', 'osd.3', 'flush_pg_stats')
+
+    # verify that there are unfound objects
+    unfound = manager.get_num_unfound_objects()
+    log.info("there are %d unfound objects" % unfound)
+    assert unfound
+
+    # mark stuff lost
+    pgs = manager.get_pg_stats()
+    for pg in pgs:
+        if pg['stat_sum']['num_objects_unfound'] > 0:
+            # verify that i can list them direct from the osd
+            log.info('listing missing/lost in %s state %s', pg['pgid'],
+                     pg['state']);
+            m = manager.list_pg_missing(pg['pgid'])
+            log.info('%s' % m)
+            assert m['num_unfound'] == pg['stat_sum']['num_objects_unfound']
+
+            log.info("reverting unfound in %s", pg['pgid'])
+            manager.raw_cluster_cmd('pg', pg['pgid'],
+                                    'mark_unfound_lost', 'delete')
+        else:
+            log.info("no unfound in %s", pg['pgid'])
+
+    manager.raw_cluster_cmd('tell', 'osd.0', 'debug', 'kick_recovery_wq', '5')
+    manager.raw_cluster_cmd('tell', 'osd.2', 'debug', 'kick_recovery_wq', '5')
+    manager.raw_cluster_cmd('tell', 'osd.3', 'debug', 'kick_recovery_wq', '5')
+    manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
+    manager.raw_cluster_cmd('tell', 'osd.2', 'flush_pg_stats')
+    manager.raw_cluster_cmd('tell', 'osd.3', 'flush_pg_stats')
+    manager.wait_for_recovery()
+
+    # verify result
+    for f in range(1, 10):
+        err = rados(ctx, mon, ['-p', pool, 'get', 'new_%d' % f, '-'])
+        assert err
+        err = rados(ctx, mon, ['-p', pool, 'get', 'existed_%d' % f, '-'])
+        assert err
+        err = rados(ctx, mon, ['-p', pool, 'get', 'existing_%d' % f, '-'])
+        assert err
+
+    # see if osd.1 can cope
+    manager.revive_osd(1)
+    manager.wait_till_osd_is_up(1)
+    manager.wait_for_clean()

diff --git a/teuthology/task/rep_lost_unfound_delete.py b/teuthology/task/rep_lost_unfound_delete.py
new file mode 100644 (file)
index 0000000..f75a4d2
--- /dev/null
+++ b/teuthology/task/rep_lost_unfound_delete.py
@@ -0,0 +1,153 @@
+"""
+Lost_unfound
+"""
+import logging
+import ceph_manager
+from teuthology import misc as teuthology
+from teuthology.task_util.rados import rados
+
+log = logging.getLogger(__name__)
+
+def task(ctx, config):
+    """
+    Test handling of lost objects.
+
+    A pretty rigid cluseter is brought up andtested by this task
+    """
+    if config is None:
+        config = {}
+    assert isinstance(config, dict), \
+        'lost_unfound task only accepts a dict for configuration'
+    first_mon = teuthology.get_first_mon(ctx, config)
+    (mon,) = ctx.cluster.only(first_mon).remotes.iterkeys()
+
+    manager = ceph_manager.CephManager(
+        mon,
+        ctx=ctx,
+        logger=log.getChild('ceph_manager'),
+        )
+
+    while len(manager.get_osd_status()['up']) < 3:
+        manager.sleep(10)
+    manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
+    manager.raw_cluster_cmd('tell', 'osd.1', 'flush_pg_stats')
+    manager.raw_cluster_cmd('tell', 'osd.2', 'flush_pg_stats')
+    manager.wait_for_clean()
+
+    # something that is always there
+    dummyfile = '/etc/fstab'
+
+    # take an osd out until the very end
+    manager.kill_osd(2)
+    manager.mark_down_osd(2)
+    manager.mark_out_osd(2)
+
+    # kludge to make sure they get a map
+    rados(ctx, mon, ['-p', 'data', 'put', 'dummy', dummyfile])
+
+    manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
+    manager.raw_cluster_cmd('tell', 'osd.1', 'flush_pg_stats')
+    manager.wait_for_recovery()
+
+    # create old objects
+    for f in range(1, 10):
+        rados(ctx, mon, ['-p', 'data', 'put', 'existing_%d' % f, dummyfile])
+        rados(ctx, mon, ['-p', 'data', 'put', 'existed_%d' % f, dummyfile])
+        rados(ctx, mon, ['-p', 'data', 'rm', 'existed_%d' % f])
+
+    # delay recovery, and make the pg log very long (to prevent backfill)
+    manager.raw_cluster_cmd(
+            'tell', 'osd.1',
+            'injectargs',
+            '--osd-recovery-delay-start 1000 --osd-min-pg-log-entries 100000000'
+            )
+
+    manager.kill_osd(0)
+    manager.mark_down_osd(0)
+    
+    for f in range(1, 10):
+        rados(ctx, mon, ['-p', 'data', 'put', 'new_%d' % f, dummyfile])
+        rados(ctx, mon, ['-p', 'data', 'put', 'existed_%d' % f, dummyfile])
+        rados(ctx, mon, ['-p', 'data', 'put', 'existing_%d' % f, dummyfile])
+
+    # bring osd.0 back up, let it peer, but don't replicate the new
+    # objects...
+    log.info('osd.0 command_args is %s' % 'foo')
+    log.info(ctx.daemons.get_daemon('osd', 0).command_args)
+    ctx.daemons.get_daemon('osd', 0).command_kwargs['args'].extend([
+            '--osd-recovery-delay-start', '1000'
+            ])
+    manager.revive_osd(0)
+    manager.mark_in_osd(0)
+    manager.wait_till_osd_is_up(0)
+
+    manager.raw_cluster_cmd('tell', 'osd.1', 'flush_pg_stats')
+    manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
+    manager.wait_till_active()
+
+    # take out osd.1 and the only copy of those objects.
+    manager.kill_osd(1)
+    manager.mark_down_osd(1)
+    manager.mark_out_osd(1)
+    manager.raw_cluster_cmd('osd', 'lost', '1', '--yes-i-really-mean-it')
+
+    # bring up osd.2 so that things would otherwise, in theory, recovery fully
+    manager.revive_osd(2)
+    manager.mark_in_osd(2)
+    manager.wait_till_osd_is_up(2)
+
+    manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
+    manager.raw_cluster_cmd('tell', 'osd.2', 'flush_pg_stats')
+    manager.wait_till_active()
+    manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
+    manager.raw_cluster_cmd('tell', 'osd.2', 'flush_pg_stats')
+
+    # verify that there are unfound objects
+    unfound = manager.get_num_unfound_objects()
+    log.info("there are %d unfound objects" % unfound)
+    assert unfound
+
+    # mark stuff lost
+    pgs = manager.get_pg_stats()
+    for pg in pgs:
+        if pg['stat_sum']['num_objects_unfound'] > 0:
+            primary = 'osd.%d' % pg['acting'][0]
+
+            # verify that i can list them direct from the osd
+            log.info('listing missing/lost in %s state %s', pg['pgid'],
+                     pg['state']);
+            m = manager.list_pg_missing(pg['pgid'])
+            #log.info('%s' % m)
+            assert m['num_unfound'] == pg['stat_sum']['num_objects_unfound']
+            num_unfound=0
+            for o in m['objects']:
+                if len(o['locations']) == 0:
+                    num_unfound += 1
+            assert m['num_unfound'] == num_unfound
+
+            log.info("reverting unfound in %s on %s", pg['pgid'], primary)
+            manager.raw_cluster_cmd('pg', pg['pgid'],
+                                    'mark_unfound_lost', 'delete')
+        else:
+            log.info("no unfound in %s", pg['pgid'])
+
+    manager.raw_cluster_cmd('tell', 'osd.0', 'debug', 'kick_recovery_wq', '5')
+    manager.raw_cluster_cmd('tell', 'osd.2', 'debug', 'kick_recovery_wq', '5')
+    manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
+    manager.raw_cluster_cmd('tell', 'osd.2', 'flush_pg_stats')
+    manager.wait_for_recovery()
+
+    # verify result
+    for f in range(1, 10):
+        err = rados(ctx, mon, ['-p', 'data', 'get', 'new_%d' % f, '-'])
+        assert err
+        err = rados(ctx, mon, ['-p', 'data', 'get', 'existed_%d' % f, '-'])
+        assert err
+        err = rados(ctx, mon, ['-p', 'data', 'get', 'existing_%d' % f, '-'])
+        assert err
+
+    # see if osd.1 can cope
+    manager.revive_osd(1)
+    manager.mark_in_osd(1)
+    manager.wait_till_osd_is_up(1)
+    manager.wait_for_clean()