def test_auto_load_balance(self):
"""
"""
log.info(f"start to install milvus")
release_name, host, port = install_milvus(
"test-auto-load-balance") # todo add release name
self.release_name = release_name
assert host is not None
conn = connections.connect("default", host=host, port=port)
assert conn is not None
self.health_checkers = {
Op.create: CreateChecker(),
Op.insert: InsertFlushChecker(),
Op.flush: InsertFlushChecker(flush=True),
Op.index: IndexChecker(),
Op.search: SearchChecker(),
Op.query: QueryChecker()
}
cc.start_monitor_threads(self.health_checkers)
# wait
sleep(constants.WAIT_PER_OP * 10)
all_collections = list_collections()
for c in all_collections:
seg_info = utility.get_query_segment_info(c)
seg_distribution = cf.get_segment_distribution(seg_info)
for k in seg_distribution.keys():
log.info(
f"collection {c}'s segment distribution in node {k} is {seg_distribution[k]['sealed']}"
)
# first assert
log.info("first assert")
assert_statistic(self.health_checkers)
# scale up
log.info("scale up milvus")
scale_up_milvus(self.release_name)
# reset counting
cc.reset_counting(self.health_checkers)
sleep(constants.WAIT_PER_OP * 10)
all_collections = list_collections()
for c in all_collections:
seg_info = utility.get_query_segment_info(c)
seg_distribution = cf.get_segment_distribution(seg_info)
for k in seg_distribution.keys():
log.info(
f"collection {c}'s sealed segment distribution in node {k} is {seg_distribution[k]['sealed']}"
)
# second assert
log.info("second assert")
assert_statistic(self.health_checkers)
# TODO assert segment distribution
# assert all expectations
assert_expectations()
def test_chaos(self, chaos_yaml):
# start the monitor threads to check the milvus ops
log.info("*********************Chaos Test Start**********************")
log.info(connections.get_connection_addr('default'))
cc.start_monitor_threads(self.health_checkers)
# parse chaos object
chaos_config = cc.gen_experiment_config(chaos_yaml)
meta_name = chaos_config.get('metadata', None).get('name', None)
release_name = meta_name
chaos_config_str = json.dumps(chaos_config)
chaos_config_str = chaos_config_str.replace("milvus-chaos", release_name)
chaos_config = json.loads(chaos_config_str)
self._chaos_config = chaos_config # cache the chaos config for tear down
log.info(f"chaos_config: {chaos_config}")
# parse the test expectations in testcases.yaml
if self.parser_testcase_config(chaos_yaml, chaos_config) is False:
log.error("Fail to get the testcase info in testcases.yaml")
assert False
# init report
dir_name = "./reports"
file_name = f"./reports/{meta_name}.log"
if not os.path.exists(dir_name):
os.makedirs(dir_name)
# wait 20s
sleep(constants.WAIT_PER_OP * 2)
# assert statistic:all ops 100% succ
log.info("******1st assert before chaos: ")
assert_statistic(self.health_checkers)
with open(file_name, "a+") as f:
ts = time.strftime("%Y-%m-%d %H:%M:%S")
f.write(f"{meta_name}-{ts}\n")
f.write("1st assert before chaos:\n")
f.write(record_results(self.health_checkers))
# apply chaos object
chaos_res = CusResource(kind=chaos_config['kind'],
group=constants.CHAOS_GROUP,
version=constants.CHAOS_VERSION,
namespace=constants.CHAOS_NAMESPACE)
chaos_res.create(chaos_config)
log.info("chaos injected")
log.info(f"chaos information: {chaos_res.get(meta_name)}")
sleep(constants.WAIT_PER_OP * 2)
# reset counting
cc.reset_counting(self.health_checkers)
# wait 40s
sleep(constants.CHAOS_DURATION)
log.info(f'Alive threads: {threading.enumerate()}')
# assert statistic
log.info("******2nd assert after chaos injected: ")
assert_statistic(self.health_checkers,
expectations={Op.create: self.expect_create,
Op.insert: self.expect_insert,
Op.flush: self.expect_flush,
Op.index: self.expect_index,
Op.search: self.expect_search,
Op.query: self.expect_query
})
with open(file_name, "a+") as f:
f.write("2nd assert after chaos injected:\n")
f.write(record_results(self.health_checkers))
# delete chaos
chaos_res.delete(meta_name)
log.info("chaos deleted")
log.info(f'Alive threads: {threading.enumerate()}')
sleep(2)
# wait all pods ready
log.info(f"wait for pods in namespace {constants.CHAOS_NAMESPACE} with label app.kubernetes.io/instance={meta_name}")
wait_pods_ready(constants.CHAOS_NAMESPACE, f"app.kubernetes.io/instance={meta_name}")
log.info(f"wait for pods in namespace {constants.CHAOS_NAMESPACE} with label release={meta_name}")
wait_pods_ready(constants.CHAOS_NAMESPACE, f"release={meta_name}")
log.info("all pods are ready")
# reconnect if needed
sleep(constants.WAIT_PER_OP * 2)
cc.reconnect(connections, alias='default')
# reset counting again
cc.reset_counting(self.health_checkers)
# wait 50s (varies by feature)
sleep(constants.WAIT_PER_OP * 5)
# assert statistic: all ops success again
log.info("******3rd assert after chaos deleted: ")
assert_statistic(self.health_checkers)
with open(file_name, "a+") as f:
f.write("3rd assert after chaos deleted:\n")
f.write(record_results(self.health_checkers))
# assert all expectations
assert_expectations()
log.info("*********************Chaos Test Completed**********************")
def test_multi_replicas_with_only_one_group_available(
self, chaos_type, failed_node_type, failed_group_scope,
is_streaming):
# start the monitor threads to check the milvus ops
log.info("*********************Chaos Test Start**********************")
log.info("Test config")
log.info(cc.gen_experiment_config(config_file_name))
# log.info(f"chaos_yaml: {chaos_yaml}")
log.info(connections.get_connection_addr('default'))
if is_streaming is False:
del self.health_checkers[Op.insert]
cc.start_monitor_threads(self.health_checkers)
# get replicas info
release_name = self.instance_name
querynode_id_pod_pair = get_querynode_info(release_name)
log.info(querynode_id_pod_pair)
group_list = []
shard_leader_list = []
replicas_info, _ = self.health_checkers[
Op.search].c_wrap.get_replicas()
for g in replicas_info.groups:
group_list.append(list(g.group_nodes))
for shard in g.shards:
shard_leader_list.append(shard.shard_leader)
# keep only one group in healthy status, other groups will be unhealthy by injecting pod failure chaos,
# In the effected groups, each group has one pod is in pod failure status
target_pod_list = []
target_group = []
group_list = sorted(group_list, key=lambda x: -len(x))
if failed_group_scope == "one":
target_group = random.sample(group_list, 1)
if failed_group_scope == "except_one":
target_group = random.sample(group_list, len(group_list) - 1)
if failed_group_scope == "all":
target_group = group_list[:]
for g in target_group:
target_nodes = []
if failed_node_type == "shard_leader":
target_nodes = list(set(g) & set(shard_leader_list))
if failed_node_type == "non_shard_leader":
target_nodes = list(set(g) - set(shard_leader_list))
if len(target_nodes) == 0:
log.info("there is no node satisfied, chose one randomly")
target_nodes = [random.choice(g)]
for target_node in target_nodes:
pod = querynode_id_pod_pair[target_node]
target_pod_list.append(pod)
log.info(f"target_pod_list: {target_pod_list}")
chaos_config = cc.gen_experiment_config(
f"{str(Path(__file__).absolute().parent)}/chaos_objects/template/{chaos_type}-by-pod-list.yaml"
)
chaos_config['metadata'][
'name'] = f"test-multi-replicase-{int(time.time())}"
meta_name = chaos_config.get('metadata', None).get('name', None)
chaos_config['spec']['selector']['pods'][
'chaos-testing'] = target_pod_list
self._chaos_config = chaos_config # cache the chaos config for tear down
log.info(f"chaos_config: {chaos_config}")
# wait 20s
sleep(constants.WAIT_PER_OP * 2)
# replicas info
replicas_info, _ = self.health_checkers[
Op.search].c_wrap.get_replicas()
log.info(
f"replicas_info for search collection {self.health_checkers[Op.search].c_wrap.name}: {replicas_info}"
)
# assert statistic:all ops 100% succ
log.info("******1st assert before chaos: ")
assert_statistic(self.health_checkers)
# apply chaos object
chaos_res = CusResource(kind=chaos_config['kind'],
group=constants.CHAOS_GROUP,
version=constants.CHAOS_VERSION,
namespace=constants.CHAOS_NAMESPACE)
chaos_res.create(chaos_config)
log.info("chaos injected")
sleep(constants.WAIT_PER_OP * 2)
# reset counting
cc.reset_counting(self.health_checkers)
# wait 120s
sleep(constants.CHAOS_DURATION)
log.info(f'Alive threads: {threading.enumerate()}')
# node info
querynode_id_pod_pair = get_querynode_info(release_name)
log.info(querynode_id_pod_pair)
# replicas info
replicas_info, _ = self.health_checkers[
Op.search].c_wrap.get_replicas()
log.info(
f"replicas_info for search collection {self.health_checkers[Op.search].c_wrap.name}: {replicas_info}"
)
replicas_info, _ = self.health_checkers[Op.query].c_wrap.get_replicas()
log.info(
f"replicas_info for query collection {self.health_checkers[Op.query].c_wrap.name}: {replicas_info}"
)
# assert statistic
log.info("******2nd assert after chaos injected: ")
expectations = {Op.search: constants.SUCC, Op.query: constants.SUCC}
if failed_group_scope == "all":
expectations = {
Op.search: constants.FAIL,
Op.query: constants.FAIL
}
assert_statistic(self.health_checkers, expectations=expectations)
# delete chaos
chaos_res.delete(meta_name)
log.info("chaos deleted")
sleep(2)
# wait all pods ready
log.info(
f"wait for pods in namespace {constants.CHAOS_NAMESPACE} with label app.kubernetes.io/instance={release_name}"
)
ready_1 = wait_pods_ready(
constants.CHAOS_NAMESPACE,
f"app.kubernetes.io/instance={release_name}")
log.info(
f"wait for pods in namespace {constants.CHAOS_NAMESPACE} with label release={release_name}"
)
ready_2 = wait_pods_ready(constants.CHAOS_NAMESPACE,
f"release={release_name}")
if ready_1 and ready_2:
log.info("all pods are ready")
# reconnect if needed
sleep(constants.WAIT_PER_OP * 2)
# cc.reconnect(connections, alias='default')
# reset counting again
cc.reset_counting(self.health_checkers)
# wait 50s (varies by feature)
sleep(constants.WAIT_PER_OP * 5)
# node info
querynode_id_pod_pair = get_querynode_info(release_name)
log.info(querynode_id_pod_pair)
sleep(30)
# replicas info
replicas_info, _ = self.health_checkers[
Op.search].c_wrap.get_replicas()
log.info(
f"replicas_info for collection {self.health_checkers[Op.search].c_wrap.name}: {replicas_info}"
)
replicas_info, _ = self.health_checkers[Op.query].c_wrap.get_replicas()
log.info(
f"replicas_info for collection {self.health_checkers[Op.query].c_wrap.name}: {replicas_info}"
)
# assert statistic: all ops success again
log.info("******3rd assert after chaos deleted: ")
assert_statistic(self.health_checkers)
# assert all expectations
assert_expectations()
log.info(
"*********************Chaos Test Completed**********************")
def test_chaos(self, chaos_yaml):
# start the monitor threads to check the milvus ops
log.info("*********************Chaos Test Start**********************")
log.info(connections.get_connection_addr('default'))
self.checker_threads = cc.start_monitor_threads(self.health_checkers)
# parse chaos object
chaos_config = cc.gen_experiment_config(chaos_yaml)
self._chaos_config = chaos_config # cache the chaos config for tear down
log.info(f"chaos_config: {chaos_config}")
# parse the test expectations in testcases.yaml
if self.parser_testcase_config(chaos_yaml) is False:
log.error("Fail to get the testcase info in testcases.yaml")
assert False
# init report
meta_name = chaos_config.get('metadata', None).get('name', None)
dir_name = "./reports"
file_name = f"./reports/{meta_name}.log"
if not os.path.exists(dir_name):
os.makedirs(dir_name)
# wait 20s
sleep(constants.WAIT_PER_OP * 2)
# assert statistic:all ops 100% succ
log.info("******1st assert before chaos: ")
assert_statistic(self.health_checkers)
with open(file_name, "a+") as f:
f.write("1st assert before chaos: ")
f.write(f"{self.health_checkers}\n")
# apply chaos object
chaos_res = CusResource(kind=chaos_config['kind'],
group=constants.CHAOS_GROUP,
version=constants.CHAOS_VERSION,
namespace=constants.CHAOS_NAMESPACE)
chaos_res.create(chaos_config)
log.info("chaos injected")
log.info(f"chaos information: {chaos_res.get(meta_name)}")
sleep(constants.WAIT_PER_OP * 2.1)
# reset counting
cc.reset_counting(self.health_checkers)
# wait 40s
sleep(constants.CHAOS_DURATION)
for k, t in self.checker_threads.items():
log.info(f"10s later: Thread {k} is_alive(): {t.is_alive()}")
# assert statistic
log.info("******2nd assert after chaos injected: ")
assert_statistic(self.health_checkers,
expectations={
Op.create: self.expect_create,
Op.insert: self.expect_insert,
Op.flush: self.expect_flush,
Op.index: self.expect_index,
Op.search: self.expect_search,
Op.query: self.expect_query
})
with open(file_name, "a+") as f:
f.write("2nd assert after chaos injected:")
f.write(f"{self.health_checkers}\n")
# delete chaos
chaos_res.delete(meta_name)
log.info("chaos deleted")
for k, t in self.checker_threads.items():
log.info(f"Thread {k} is_alive(): {t.is_alive()}")
sleep(2)
# reconnect if needed
sleep(constants.WAIT_PER_OP * 2)
cc.reconnect(connections, alias='default')
# reset counting again
cc.reset_counting(self.health_checkers)
# wait 50s (varies by feature)
sleep(constants.WAIT_PER_OP * 5)
# assert statistic: all ops success again
log.info("******3rd assert after chaos deleted: ")
assert_statistic(self.health_checkers)
with open(file_name, "a+") as f:
f.write("3rd assert after chaos deleted:")
f.write(f"{self.health_checkers}\n")
# assert all expectations
assert_expectations()
log.info(
"*********************Chaos Test Completed**********************")
def test_bulk_load(self, chaos_type, target_component):
# start the monitor threads to check the milvus ops
log.info("*********************Chaos Test Start**********************")
log.info(connections.get_connection_addr('default'))
release_name = self.instance_name
cc.start_monitor_threads(self.health_checkers)
chaos_config = cc.gen_experiment_config(
f"{str(Path(__file__).absolute().parent)}/chaos_objects/{chaos_type}/chaos_{target_component}_{chaos_type}.yaml"
)
chaos_config['metadata']['name'] = f"test-bulk-load-{int(time.time())}"
kind = chaos_config['kind']
meta_name = chaos_config.get('metadata', None).get('name', None)
update_key_value(chaos_config, "release", release_name)
update_key_value(chaos_config, "app.kubernetes.io/instance",
release_name)
self._chaos_config = chaos_config # cache the chaos config for tear down
log.info(f"chaos_config: {chaos_config}")
# wait 20s
sleep(constants.WAIT_PER_OP * 10)
# assert statistic:all ops 100% succ
log.info("******1st assert before chaos: ")
assert_statistic(self.health_checkers)
# apply chaos object
chaos_res = CusResource(kind=chaos_config['kind'],
group=constants.CHAOS_GROUP,
version=constants.CHAOS_VERSION,
namespace=constants.CHAOS_NAMESPACE)
chaos_res.create(chaos_config)
log.info("chaos injected")
sleep(constants.WAIT_PER_OP * 10)
# reset counting
cc.reset_counting(self.health_checkers)
# wait 120s
sleep(constants.CHAOS_DURATION)
log.info(f'Alive threads: {threading.enumerate()}')
# assert statistic
log.info("******2nd assert after chaos injected: ")
assert_statistic(self.health_checkers,
expectations={
Op.bulk_load: constants.FAIL,
})
# delete chaos
chaos_res.delete(meta_name)
log.info("chaos deleted")
sleep(2)
# wait all pods ready
log.info(
f"wait for pods in namespace {constants.CHAOS_NAMESPACE} with label app.kubernetes.io/instance={release_name}"
)
wait_pods_ready(constants.CHAOS_NAMESPACE,
f"app.kubernetes.io/instance={release_name}")
log.info(
f"wait for pods in namespace {constants.CHAOS_NAMESPACE} with label release={release_name}"
)
wait_pods_ready(constants.CHAOS_NAMESPACE, f"release={release_name}")
log.info("all pods are ready")
# reconnect if needed
sleep(constants.WAIT_PER_OP * 2)
cc.reconnect(connections, alias='default')
# recheck failed tasks in third assert
self.health_checkers[Op.bulk_load].recheck_failed_task = True
# reset counting again
cc.reset_counting(self.health_checkers)
# wait 50s (varies by feature)
sleep(constants.WAIT_PER_OP * 10)
# assert statistic: all ops success again
log.info("******3rd assert after chaos deleted: ")
assert_statistic(self.health_checkers)
# assert all expectations
assert_expectations()
log.info(
"*********************Chaos Test Completed**********************")