def test_capi_with_checkpointing(self):
self.setup_xdcr()
self.src_cluster.pause_all_replications()
gen = DocumentGenerator('es', '{{"key":"value","mutated":0}}', range(100), start=0, end=self._num_items)
self.src_cluster.load_all_buckets_from_generator(gen)
self.src_cluster.resume_all_replications()
self._wait_for_replication_to_catchup()
self.sleep(120)
vb0_node = None
nodes = self.src_cluster.get_nodes()
ip = VBucketAwareMemcached(self.rest, 'default').vBucketMap[0].split(':')[0]
for node in nodes:
if ip == node.ip:
vb0_node = node
if not vb0_node:
raise XDCRCheckpointException("Error determining the node containing active vb0")
vb0_conn = RestConnection(vb0_node)
try:
repl = vb0_conn.get_replication_for_buckets('default', 'default')
checkpoint_record = vb0_conn.get_recent_xdcr_vb_ckpt(repl['id'])
self.log.info("Checkpoint record : {0}".format(checkpoint_record))
except Exception as e:
raise XDCRCheckpointException("Error retrieving last checkpoint document - {0}".format(e))
self._verify_es_results()
def validate_remote_failover_log(self, vb_uuid, high_seqno):
# TAP based validation
remote_uuid, remote_highseq = self.get_failover_log(self.dest_master)
self.log.info("Remote failover log = [{0},{1}]".format(remote_uuid, remote_highseq))
if int(remote_uuid) != int(vb_uuid):
raise XDCRCheckpointException("vb_uuid in commitopaque is {0} while actual remote vb_uuid is {1}"
.format(vb_uuid, remote_uuid))
def get_and_validate_latest_checkpoint(self):
rest_con = RestConnection(self.get_active_vb0_node(self.src_master))
repl = rest_con.get_replication_for_buckets('default', 'default')
try:
checkpoint_record = rest_con.get_recent_xdcr_vb_ckpt(repl['id'])
self.log.info("Checkpoint record : {0}".format(checkpoint_record))
self.chkpt_records.append(checkpoint_record)
except Exception as e:
raise XDCRCheckpointException("Error retrieving last checkpoint document - {0}".format(e))
failover_uuid = checkpoint_record["failover_uuid"]
seqno = checkpoint_record["seqno"]
self.log.info ("Verifying commitopaque/remote failover log ...")
if seqno != 0:
self.validate_remote_failover_log(checkpoint_record["target_vb_opaque"]["target_vb_uuid"], checkpoint_record["target_seqno"])
self.log.info ("Verifying local failover uuid ...")
local_vb_uuid, _ = self.get_failover_log(self.src_master)
self.assertTrue((int(failover_uuid) == int(local_vb_uuid)) or
(int(failover_uuid) == 0),
"local failover_uuid is wrong in checkpoint record! Expected: {0} seen: {1}".
format(local_vb_uuid, failover_uuid))
self.log.info("Checkpoint record verified")
else:
self.log.info("Skipping checkpoint record checks for checkpoint-0")
return True
def get_and_validate_latest_checkpoint(self):
rest_con = RestConnection(self.src_master)
try:
checkpoint_record = rest_con.get_recent_xdcr_vb_ckpt('default')
self.log.info("Checkpoint record : {}".format(checkpoint_record))
self.chkpt_records.append(checkpoint_record)
except Exception as e:
raise XDCRCheckpointException(
"Error retrieving last checkpoint document - {}".format(e))
commit_opaque = checkpoint_record["commitopaque"]
failover_uuid = checkpoint_record["failover_uuid"]
seqno = checkpoint_record["seqno"]
self.log.info("Verifying commitopaque/remote failover log ...")
if seqno != 0:
if rest_con.is_goxdcr_enabled():
self.validate_remote_failover_log(
checkpoint_record["target_vb_uuid"],
checkpoint_record["commitopaque"])
else:
self.validate_remote_failover_log(commit_opaque[0],
commit_opaque[1])
self.log.info("Verifying local failover uuid ...")
local_vb_uuid, _ = self.get_failover_log(self.src_master)
self.assertTrue(
int(local_vb_uuid) == int(failover_uuid),
"local failover_uuid is wrong in checkpoint record! Expected: {0} seen: {1}"
.format(local_vb_uuid, failover_uuid))
self.log.info("Checkpoint record verified")
else:
self.log.info("Skipping checkpoint record checks for checkpoint-0")
return True
def wait_for_checkpoint_to_happen(self, timeout=180):
"""
Keeps checking if num_checkpoints stat for the replication
was incremented, every 10 sec, times out after 2 mins
"""
end_time = time.time() + timeout
while time.time() < end_time:
num_success_ckpts =self.get_stat_successful_checkpoints()
if num_success_ckpts > self.stat_num_success_ckpts:
return
else:
self.sleep(10)
else:
raise XDCRCheckpointException("Timed-out waiting for checkpoint to happen")
