forked from TrueCloudLab/frostfs-testcases
724 lines
31 KiB
Python
724 lines
31 KiB
Python
import logging
|
|
import random
|
|
from datetime import datetime
|
|
from time import sleep
|
|
|
|
import allure
|
|
import pytest
|
|
from frostfs_testlib import reporter
|
|
from frostfs_testlib.resources.common import MORPH_BLOCK_TIME
|
|
from frostfs_testlib.resources.wellknown_acl import PUBLIC_ACL
|
|
from frostfs_testlib.s3 import S3ClientWrapper, VersioningStatus
|
|
from frostfs_testlib.steps.cli.container import StorageContainer, StorageContainerInfo, create_container
|
|
from frostfs_testlib.steps.cli.object import get_object, put_object_to_random_node
|
|
from frostfs_testlib.steps.node_management import (
|
|
check_node_in_map,
|
|
check_node_not_in_map,
|
|
exclude_node_from_network_map,
|
|
include_node_to_network_map,
|
|
remove_nodes_from_map_morph,
|
|
wait_for_node_to_be_ready,
|
|
)
|
|
from frostfs_testlib.steps.s3 import s3_helper
|
|
from frostfs_testlib.steps.s3.s3_helper import search_nodes_with_bucket
|
|
from frostfs_testlib.storage.cluster import Cluster, ClusterNode, S3Gate, StorageNode
|
|
from frostfs_testlib.storage.controllers import ClusterStateController, ShardsWatcher
|
|
from frostfs_testlib.storage.dataclasses.object_size import ObjectSize
|
|
from frostfs_testlib.storage.dataclasses.storage_object_info import StorageObjectInfo
|
|
from frostfs_testlib.storage.dataclasses.wallet import WalletInfo
|
|
from frostfs_testlib.testing.cluster_test_base import ClusterTestBase
|
|
from frostfs_testlib.testing.test_control import expect_not_raises
|
|
from frostfs_testlib.utils import datetime_utils
|
|
from frostfs_testlib.utils.failover_utils import wait_object_replication
|
|
from frostfs_testlib.utils.file_keeper import FileKeeper
|
|
from frostfs_testlib.utils.file_utils import generate_file, get_file_hash
|
|
|
|
logger = logging.getLogger("NeoLogger")
|
|
stopped_nodes: list[StorageNode] = []
|
|
|
|
|
|
@pytest.fixture(scope="function")
|
|
@allure.title("Provide File Keeper")
|
|
def file_keeper():
|
|
keeper = FileKeeper()
|
|
yield keeper
|
|
keeper.restore_files()
|
|
|
|
|
|
@reporter.step("Return all stopped hosts")
|
|
@pytest.fixture(scope="function", autouse=True)
|
|
def after_run_return_all_stopped_hosts(cluster_state_controller: ClusterStateController) -> str:
|
|
yield
|
|
cluster_state_controller.start_stopped_hosts()
|
|
|
|
|
|
@reporter.step("Return all stopped services after test")
|
|
@pytest.fixture(scope="function")
|
|
def after_run_return_all_stopped_services(cluster_state_controller: ClusterStateController):
|
|
yield
|
|
cluster_state_controller.start_all_stopped_services()
|
|
|
|
|
|
@pytest.mark.failover
|
|
@pytest.mark.failover_storage
|
|
class TestFailoverStorage(ClusterTestBase):
|
|
@allure.title("Shutdown and start node (stop_mode={stop_mode})")
|
|
@pytest.mark.parametrize("stop_mode", ["hard", "soft"])
|
|
@pytest.mark.failover_reboot
|
|
def test_lose_storage_node_host(
|
|
self,
|
|
default_wallet,
|
|
stop_mode: str,
|
|
require_multiple_hosts,
|
|
simple_object_size: ObjectSize,
|
|
cluster: Cluster,
|
|
cluster_state_controller: ClusterStateController,
|
|
):
|
|
wallet = default_wallet
|
|
placement_rule = "REP 2 IN X CBF 2 SELECT 2 FROM * AS X"
|
|
source_file_path = generate_file(simple_object_size.value)
|
|
stopped_hosts_nodes = []
|
|
|
|
with reporter.step(f"Create container and put object"):
|
|
cid = create_container(
|
|
wallet,
|
|
shell=self.shell,
|
|
endpoint=self.cluster.default_rpc_endpoint,
|
|
rule=placement_rule,
|
|
basic_acl=PUBLIC_ACL,
|
|
)
|
|
oid = put_object_to_random_node(wallet, source_file_path, cid, shell=self.shell, cluster=self.cluster)
|
|
|
|
with reporter.step(f"Wait for replication and get nodes with object"):
|
|
nodes_with_object = wait_object_replication(cid, oid, 2, shell=self.shell, nodes=self.cluster.storage_nodes)
|
|
|
|
with reporter.step(f"Stop 2 nodes with object and wait replication one by one"):
|
|
for storage_node in random.sample(nodes_with_object, 2):
|
|
stopped_hosts_nodes.append(storage_node)
|
|
|
|
cluster_node = cluster.node(storage_node)
|
|
cluster_state_controller.stop_node_host(cluster_node, stop_mode)
|
|
|
|
replicated_nodes = wait_object_replication(
|
|
cid,
|
|
oid,
|
|
2,
|
|
shell=self.shell,
|
|
nodes=list(set(self.cluster.storage_nodes) - {*stopped_hosts_nodes}),
|
|
)
|
|
|
|
with reporter.step("Check object data is not corrupted"):
|
|
got_file_path = get_object(
|
|
wallet, cid, oid, endpoint=replicated_nodes[0].get_rpc_endpoint(), shell=self.shell
|
|
)
|
|
assert get_file_hash(source_file_path) == get_file_hash(got_file_path)
|
|
|
|
with reporter.step("Return all hosts"):
|
|
cluster_state_controller.start_stopped_hosts()
|
|
|
|
with reporter.step("Check object data is not corrupted"):
|
|
replicated_nodes = wait_object_replication(cid, oid, 2, shell=self.shell, nodes=self.cluster.storage_nodes)
|
|
got_file_path = get_object(
|
|
wallet, cid, oid, shell=self.shell, endpoint=replicated_nodes[0].get_rpc_endpoint()
|
|
)
|
|
assert get_file_hash(source_file_path) == get_file_hash(got_file_path)
|
|
|
|
@allure.title("Do not ignore unhealthy tree endpoints (s3_client={s3_client})")
|
|
def test_unhealthy_tree(
|
|
self,
|
|
s3_client: S3ClientWrapper,
|
|
default_wallet: str,
|
|
simple_object_size: ObjectSize,
|
|
cluster_state_controller: ClusterStateController,
|
|
after_run_return_all_stopped_services,
|
|
):
|
|
default_node = self.cluster.cluster_nodes[0]
|
|
|
|
with reporter.step("Turn S3 GW off on default node"):
|
|
cluster_state_controller.stop_service_of_type(default_node, S3Gate)
|
|
|
|
with reporter.step("Turn off storage on default node"):
|
|
cluster_state_controller.stop_service_of_type(default_node, StorageNode)
|
|
|
|
with reporter.step("Turn on S3 GW on default node"):
|
|
cluster_state_controller.start_service_of_type(default_node, S3Gate)
|
|
|
|
with reporter.step("Turn on storage on default node"):
|
|
cluster_state_controller.start_service_of_type(default_node, StorageNode)
|
|
|
|
with reporter.step("Create bucket with REP 1 SELECT 1 policy"):
|
|
bucket = s3_client.create_bucket(
|
|
location_constraint="load-1-1",
|
|
)
|
|
|
|
file_path = generate_file(simple_object_size.value)
|
|
file_name = s3_helper.object_key_from_file_path(file_path)
|
|
with reporter.step("Put object into bucket"):
|
|
put_object = s3_client.put_object(bucket, file_path)
|
|
s3_helper.check_objects_in_bucket(s3_client, bucket, expected_objects=[file_name])
|
|
|
|
node_bucket = search_nodes_with_bucket(
|
|
cluster=self.cluster,
|
|
bucket_name=bucket,
|
|
wallet=default_wallet,
|
|
shell=self.shell,
|
|
endpoint=self.cluster.storage_nodes[0].get_rpc_endpoint(),
|
|
)[0]
|
|
|
|
with reporter.step("Turn off all storage nodes except bucket node"):
|
|
for node in [node_to_stop for node_to_stop in self.cluster.cluster_nodes if node_to_stop != node_bucket]:
|
|
with reporter.step(f"Stop storage service on node: {node}"):
|
|
cluster_state_controller.stop_service_of_type(node, StorageNode)
|
|
|
|
with reporter.step(f"Change s3 endpoint to bucket node"):
|
|
s3_client.set_endpoint(node_bucket.s3_gate.get_endpoint())
|
|
|
|
with reporter.step("Check that object is available"):
|
|
s3_helper.check_objects_in_bucket(s3_client, bucket, expected_objects=[file_name])
|
|
|
|
with reporter.step("Start storage nodes"):
|
|
cluster_state_controller.start_all_stopped_services()
|
|
|
|
|
|
@pytest.mark.failover
|
|
@pytest.mark.failover_empty_map
|
|
class TestEmptyMap(ClusterTestBase):
|
|
"""
|
|
A set of tests for makes map empty and verify that we can read objects after that
|
|
"""
|
|
|
|
@reporter.step("Teardown after EmptyMap offline test")
|
|
@pytest.fixture()
|
|
def empty_map_offline_teardown(self):
|
|
yield
|
|
with reporter.step("Return all storage nodes to network map"):
|
|
for node in stopped_nodes:
|
|
include_node_to_network_map(node, node, shell=self.shell, cluster=self.cluster)
|
|
stopped_nodes.remove(node)
|
|
|
|
@pytest.mark.failover_empty_map_offlne
|
|
@allure.title("Empty network map via offline all storage nodes (s3_client={s3_client})")
|
|
def test_offline_all_storage_nodes(
|
|
self,
|
|
s3_client: S3ClientWrapper,
|
|
bucket: str,
|
|
simple_object_size: ObjectSize,
|
|
empty_map_offline_teardown,
|
|
):
|
|
"""
|
|
The test makes network map empty (set offline status on all storage nodes) then returns all nodes to map and checks that object can read through s3.
|
|
|
|
Steps:
|
|
1. Check that bucket is empty
|
|
2: PUT object into bucket
|
|
3: Check that object exists in bucket
|
|
4: Exclude all storage nodes from network map (set status OFFLINE)
|
|
5: Return all storage nodes to network map
|
|
6: Check that we can read object from #2
|
|
Args:
|
|
bucket: bucket which contains tested object
|
|
simple_object_size: size of object
|
|
"""
|
|
file_path = generate_file(simple_object_size.value)
|
|
file_name = s3_helper.object_key_from_file_path(file_path)
|
|
bucket_objects = [file_name]
|
|
|
|
objects_list = s3_client.list_objects(bucket)
|
|
assert not objects_list, f"Expected empty bucket, got {objects_list}"
|
|
|
|
with reporter.step("Put object into bucket"):
|
|
s3_client.put_object(bucket, file_path)
|
|
|
|
with reporter.step("Check that object exists in bucket"):
|
|
s3_helper.check_objects_in_bucket(s3_client, bucket, bucket_objects)
|
|
|
|
storage_nodes = self.cluster.storage_nodes
|
|
with reporter.step("Exclude all storage nodes from network map"):
|
|
for node in storage_nodes:
|
|
stopped_nodes.append(node)
|
|
exclude_node_from_network_map(node, node, shell=self.shell, cluster=self.cluster)
|
|
|
|
with reporter.step("Return all storage nodes to network map"):
|
|
for node in storage_nodes:
|
|
include_node_to_network_map(node, node, shell=self.shell, cluster=self.cluster)
|
|
stopped_nodes.remove(node)
|
|
|
|
with reporter.step("Check that we can read object"):
|
|
s3_helper.check_objects_in_bucket(s3_client, bucket, bucket_objects)
|
|
|
|
@reporter.step("Teardown after EmptyMap stop service test")
|
|
@pytest.fixture()
|
|
def empty_map_stop_service_teardown(self, cluster_state_controller: ClusterStateController):
|
|
yield
|
|
with reporter.step("Return all storage nodes to network map"):
|
|
cluster_state_controller.start_all_stopped_services()
|
|
for node in stopped_nodes:
|
|
check_node_in_map(node, shell=self.shell, alive_node=node)
|
|
|
|
@pytest.mark.failover_empty_map_stop_service
|
|
@allure.title("Empty network map via stop all storage services (s3_client={s3_client})")
|
|
def test_stop_all_storage_nodes(
|
|
self,
|
|
s3_client: S3ClientWrapper,
|
|
bucket: str,
|
|
simple_object_size: ObjectSize,
|
|
empty_map_stop_service_teardown,
|
|
cluster_state_controller: ClusterStateController,
|
|
):
|
|
"""
|
|
The test makes network map empty (stop storage service on all nodes
|
|
then use 'frostfs-adm morph delete-nodes' to delete nodes from map)
|
|
then start all services and checks that object can read through s3.
|
|
|
|
Steps:
|
|
1. Check that bucket is empty
|
|
2: PUT object into bucket
|
|
3: Check that object exists in bucket
|
|
4: Exclude all storage nodes from network map (stop storage service
|
|
and manual exclude from map)
|
|
5: Return all storage nodes to network map
|
|
6: Check that we can read object from #2
|
|
Args:
|
|
bucket: bucket which contains tested object
|
|
simple_object_size: size of object
|
|
"""
|
|
file_path = generate_file(simple_object_size.value)
|
|
file_name = s3_helper.object_key_from_file_path(file_path)
|
|
bucket_objects = [file_name]
|
|
|
|
objects_list = s3_client.list_objects(bucket)
|
|
assert not objects_list, f"Expected empty bucket, got {objects_list}"
|
|
|
|
with reporter.step("Put object into bucket"):
|
|
s3_client.put_object(bucket, file_path)
|
|
|
|
with reporter.step("Check that object exists in bucket"):
|
|
s3_helper.check_objects_in_bucket(s3_client, bucket, bucket_objects)
|
|
|
|
with reporter.step("Stop all storage nodes"):
|
|
cluster_state_controller.stop_services_of_type(StorageNode)
|
|
|
|
with reporter.step("Remove all nodes from network map"):
|
|
remove_nodes_from_map_morph(
|
|
shell=self.shell, cluster=self.cluster, remove_nodes=self.cluster.services(StorageNode)
|
|
)
|
|
|
|
with reporter.step("Return all storage nodes to network map"):
|
|
self.return_nodes_after_stop_with_check_empty_map(cluster_state_controller)
|
|
|
|
with reporter.step("Check that object exists in bucket"):
|
|
s3_helper.check_objects_in_bucket(s3_client, bucket, bucket_objects)
|
|
|
|
@reporter.step("Return all nodes to cluster with check empty map first")
|
|
def return_nodes_after_stop_with_check_empty_map(self, cluster_state_controller: ClusterStateController) -> None:
|
|
first_node = self.cluster.cluster_nodes[0].service(StorageNode)
|
|
|
|
with reporter.step("Start first node and check network map"):
|
|
cluster_state_controller.start_service_of_type(self.cluster.cluster_nodes[0], StorageNode)
|
|
wait_for_node_to_be_ready(first_node)
|
|
|
|
for check_node in self.cluster.storage_nodes:
|
|
check_node_not_in_map(check_node, shell=self.shell, alive_node=first_node)
|
|
|
|
for node in self.cluster.cluster_nodes[1:]:
|
|
storage_node = node.service(StorageNode)
|
|
|
|
cluster_state_controller.start_service_of_type(node, StorageNode)
|
|
wait_for_node_to_be_ready(storage_node)
|
|
|
|
sleep(datetime_utils.parse_time(MORPH_BLOCK_TIME))
|
|
self.tick_epochs(1)
|
|
|
|
check_node_in_map(storage_node, shell=self.shell, alive_node=first_node)
|
|
|
|
@allure.title("Object loss from fstree/blobovnicza (versioning=enabled, s3_client={s3_client})")
|
|
def test_s3_fstree_blobovnicza_loss_versioning_on(
|
|
self,
|
|
s3_client: S3ClientWrapper,
|
|
simple_object_size: ObjectSize,
|
|
cluster_state_controller: ClusterStateController,
|
|
):
|
|
bucket = s3_client.create_bucket()
|
|
s3_helper.set_bucket_versioning(s3_client, bucket, VersioningStatus.ENABLED)
|
|
|
|
file_path = generate_file(simple_object_size.value)
|
|
file_name = s3_helper.object_key_from_file_path(file_path)
|
|
|
|
object_versions = []
|
|
with reporter.step("Put object into one bucket"):
|
|
put_object = s3_client.put_object(bucket, file_path)
|
|
s3_helper.check_objects_in_bucket(s3_client, bucket, expected_objects=[file_name])
|
|
object_versions.append(put_object)
|
|
|
|
with reporter.step("Stop all storage nodes"):
|
|
cluster_state_controller.stop_services_of_type(StorageNode)
|
|
|
|
with reporter.step("Delete blobovnicza and fstree from all nodes"):
|
|
for node in self.cluster.storage_nodes:
|
|
node.delete_blobovnicza()
|
|
node.delete_fstree()
|
|
|
|
with reporter.step("Start all storage nodes"):
|
|
cluster_state_controller.start_all_stopped_services()
|
|
|
|
# need to get Delete Marker first
|
|
with reporter.step("Delete the object from the bucket"):
|
|
delete_object = s3_client.delete_object(bucket, file_name)
|
|
object_versions.append(delete_object["VersionId"])
|
|
|
|
# and now delete all versions of object (including Delete Markers)
|
|
with reporter.step("Delete all versions of the object from the bucket"):
|
|
for version in object_versions:
|
|
delete_object = s3_client.delete_object(bucket, file_name, version_id=version)
|
|
|
|
with reporter.step("Delete bucket"):
|
|
s3_client.delete_bucket(bucket)
|
|
|
|
@allure.title("Object loss from fstree/blobovnicza (versioning=disabled, s3_client={s3_client})")
|
|
def test_s3_fstree_blobovnicza_loss_versioning_off(
|
|
self,
|
|
s3_client: S3ClientWrapper,
|
|
simple_object_size: ObjectSize,
|
|
cluster_state_controller: ClusterStateController,
|
|
):
|
|
bucket = s3_client.create_bucket()
|
|
|
|
file_path = generate_file(simple_object_size.value)
|
|
file_name = s3_helper.object_key_from_file_path(file_path)
|
|
|
|
with reporter.step("Put object into one bucket"):
|
|
s3_client.put_object(bucket, file_path)
|
|
s3_helper.check_objects_in_bucket(s3_client, bucket, expected_objects=[file_name])
|
|
|
|
with reporter.step("Stop all storage nodes"):
|
|
cluster_state_controller.stop_services_of_type(StorageNode)
|
|
|
|
with reporter.step("Delete blobovnicza and fstree from all nodes"):
|
|
for node in self.cluster.storage_nodes:
|
|
node.delete_blobovnicza()
|
|
node.delete_fstree()
|
|
|
|
with reporter.step("Start all storage nodes"):
|
|
cluster_state_controller.start_all_stopped_services()
|
|
|
|
with reporter.step("Delete the object from the bucket"):
|
|
s3_client.delete_object(bucket, file_name)
|
|
|
|
with reporter.step("Delete bucket"):
|
|
s3_client.delete_bucket(bucket)
|
|
|
|
@pytest.mark.skip(reason="Need to increase cache lifetime")
|
|
@pytest.mark.parametrize(
|
|
# versioning should NOT be VersioningStatus.SUSPENDED, it needs to be undefined
|
|
"versioning_status",
|
|
[VersioningStatus.ENABLED, VersioningStatus.UNDEFINED],
|
|
)
|
|
@allure.title(
|
|
"After Pilorama.db loss on all nodes list objects should return nothing in second listing (versioning_status={versioning_status}, s3_client={s3_client})"
|
|
)
|
|
def test_s3_pilorama_loss(
|
|
self,
|
|
s3_client: S3ClientWrapper,
|
|
simple_object_size: ObjectSize,
|
|
versioning_status: VersioningStatus,
|
|
cluster_state_controller: ClusterStateController,
|
|
):
|
|
bucket = s3_client.create_bucket()
|
|
s3_helper.set_bucket_versioning(s3_client, bucket, versioning_status)
|
|
|
|
file_path = generate_file(simple_object_size.value)
|
|
file_name = s3_helper.object_key_from_file_path(file_path)
|
|
|
|
with reporter.step("Put object into one bucket"):
|
|
s3_client.put_object(bucket, file_path)
|
|
s3_helper.check_objects_in_bucket(s3_client, bucket, expected_objects=[file_name])
|
|
|
|
with reporter.step("Stop all storage nodes"):
|
|
cluster_state_controller.stop_services_of_type(StorageNode)
|
|
|
|
with reporter.step("Delete pilorama.db from all nodes"):
|
|
for node in self.cluster.storage_nodes:
|
|
for shard in node.get_shards():
|
|
node.delete_file(shard.pilorama)
|
|
|
|
with reporter.step("Start all storage nodes"):
|
|
cluster_state_controller.start_all_stopped_services()
|
|
|
|
with reporter.step("Check list objects first time"):
|
|
objects_list = s3_client.list_objects(bucket)
|
|
assert objects_list, f"Expected not empty bucket"
|
|
|
|
with reporter.step("Check list objects second time"):
|
|
objects_list = s3_client.list_objects(bucket)
|
|
assert not objects_list, f"Expected empty bucket, got {objects_list}"
|
|
|
|
with reporter.step("Delete bucket"):
|
|
s3_client.delete_bucket(bucket)
|
|
|
|
|
|
@pytest.mark.failover
|
|
@pytest.mark.failover_data_loss
|
|
class TestStorageDataLoss(ClusterTestBase):
|
|
@allure.title(
|
|
"After metabase loss on all nodes operations on objects and buckets should be still available via S3 (s3_client={s3_client})"
|
|
)
|
|
@pytest.mark.metabase_loss
|
|
def test_metabase_loss(
|
|
self,
|
|
s3_client: S3ClientWrapper,
|
|
simple_object_size: ObjectSize,
|
|
complex_object_size: ObjectSize,
|
|
cluster_state_controller: ClusterStateController,
|
|
after_run_return_all_stopped_services: str,
|
|
file_keeper: FileKeeper,
|
|
):
|
|
allure.dynamic.description(after_run_return_all_stopped_services)
|
|
|
|
with reporter.step("Create bucket"):
|
|
bucket = s3_client.create_bucket()
|
|
|
|
with reporter.step("Put objects into bucket"):
|
|
simple_object_path = generate_file(simple_object_size.value)
|
|
simple_object_key = s3_helper.object_key_from_file_path(simple_object_path)
|
|
|
|
complex_object_path = generate_file(complex_object_size.value)
|
|
complex_object_key = s3_helper.object_key_from_file_path(complex_object_path)
|
|
|
|
s3_client.put_object(bucket, simple_object_path)
|
|
s3_client.put_object(bucket, complex_object_path)
|
|
|
|
with reporter.step("Check objects are in bucket"):
|
|
s3_helper.check_objects_in_bucket(
|
|
s3_client, bucket, expected_objects=[simple_object_key, complex_object_key]
|
|
)
|
|
|
|
with reporter.step("Stop storage services on all nodes"):
|
|
cluster_state_controller.stop_services_of_type(StorageNode)
|
|
|
|
with reporter.step("Delete metabase from all nodes"):
|
|
for node in cluster_state_controller.cluster.storage_nodes:
|
|
node.delete_metabase()
|
|
|
|
with reporter.step("Enable resync_metabase option for storage services"):
|
|
for storage_node in cluster_state_controller.cluster.storage_nodes:
|
|
with reporter.step(f"Enable resync_metabase option for {storage_node}"):
|
|
config_file_path, config = storage_node.get_shards_config()
|
|
if not config["storage"]["shard"]["default"]["resync_metabase"]:
|
|
file_keeper.add(storage_node, config_file_path)
|
|
config["storage"]["shard"]["default"]["resync_metabase"] = True
|
|
storage_node.save_config(config, config_file_path)
|
|
|
|
with reporter.step("Start storage services on all nodes"):
|
|
cluster_state_controller.start_all_stopped_services()
|
|
|
|
with reporter.step("Wait for tree rebalance"):
|
|
# TODO: Use product metric when we have proper ones for this check
|
|
sleep(30)
|
|
|
|
with reporter.step("Delete objects from bucket"):
|
|
with reporter.step("Delete simple object from bucket"):
|
|
with expect_not_raises():
|
|
s3_client.delete_object(bucket, simple_object_key)
|
|
|
|
with reporter.step("Delete complex object from bucket"):
|
|
with expect_not_raises():
|
|
s3_client.delete_object(bucket, complex_object_key)
|
|
|
|
with reporter.step("Delete bucket"):
|
|
with expect_not_raises():
|
|
s3_client.delete_bucket(bucket)
|
|
|
|
@allure.title("Write cache loss on one node should not affect shards and should not produce errors in log")
|
|
@pytest.mark.write_cache_loss
|
|
def test_write_cache_loss_on_one_node(
|
|
self,
|
|
node_under_test: ClusterNode,
|
|
simple_object_size: ObjectSize,
|
|
cluster_state_controller: ClusterStateController,
|
|
shards_watcher: ShardsWatcher,
|
|
default_wallet: str,
|
|
test_start_time: datetime,
|
|
after_run_return_all_stopped_services: str,
|
|
):
|
|
exception_messages = []
|
|
allure.dynamic.description(after_run_return_all_stopped_services)
|
|
|
|
with reporter.step(f"Create container on node {node_under_test}"):
|
|
locode = node_under_test.storage_node.get_un_locode()
|
|
placement_rule = f"""REP 1 IN X
|
|
CBF 1
|
|
SELECT 1 FROM C AS X
|
|
FILTER 'UN-LOCODE' EQ '{locode}' AS C"""
|
|
cid = create_container(
|
|
default_wallet,
|
|
self.shell,
|
|
node_under_test.storage_node.get_rpc_endpoint(),
|
|
rule=placement_rule,
|
|
)
|
|
container = StorageContainer(
|
|
StorageContainerInfo(cid, WalletInfo(default_wallet)),
|
|
self.shell,
|
|
cluster_state_controller.cluster,
|
|
)
|
|
|
|
with reporter.step(f"Put couple objects to container on node {node_under_test}"):
|
|
storage_objects: list[StorageObjectInfo] = []
|
|
for _ in range(5):
|
|
storage_object = container.generate_object(
|
|
simple_object_size.value,
|
|
endpoint=node_under_test.storage_node.get_rpc_endpoint(),
|
|
)
|
|
storage_objects.append(storage_object)
|
|
|
|
with reporter.step("Take shards snapshot"):
|
|
shards_watcher.take_shards_snapshot()
|
|
|
|
with reporter.step(f"Stop storage service on node {node_under_test}"):
|
|
cluster_state_controller.stop_service_of_type(node_under_test, StorageNode)
|
|
|
|
with reporter.step(f"Delete write cache from node {node_under_test}"):
|
|
node_under_test.storage_node.delete_write_cache()
|
|
|
|
with reporter.step(f"Start storage service on node {node_under_test}"):
|
|
cluster_state_controller.start_all_stopped_services()
|
|
|
|
with reporter.step("Objects should be available"):
|
|
for storage_object in storage_objects:
|
|
get_object(
|
|
storage_object.wallet_file_path,
|
|
container.get_id(),
|
|
storage_object.oid,
|
|
self.shell,
|
|
node_under_test.storage_node.get_rpc_endpoint(),
|
|
)
|
|
|
|
with reporter.step("No shards should have new errors"):
|
|
shards_watcher.take_shards_snapshot()
|
|
shards_with_errors = shards_watcher.get_shards_with_new_errors()
|
|
if shards_with_errors:
|
|
exception_messages.append(f"Shards have new errors: {shards_with_errors}")
|
|
|
|
with reporter.step("No shards should have degraded status"):
|
|
snapshot = shards_watcher.get_shards_snapshot()
|
|
for shard in snapshot:
|
|
status = snapshot[shard]["mode"]
|
|
if status != "read-write":
|
|
exception_messages.append(f"Shard {shard} changed status to {status}")
|
|
|
|
with reporter.step("No related errors should be in log"):
|
|
if node_under_test.host.is_message_in_logs(
|
|
message_regex=r"\Wno such file or directory\W", since=test_start_time
|
|
):
|
|
exception_messages.append(f"Node {node_under_test} have shard errors in logs")
|
|
|
|
with reporter.step("Pass test if no errors found"):
|
|
assert not exception_messages, "\n".join(exception_messages)
|
|
|
|
@allure.title(
|
|
"Loss of one node should trigger use of tree and storage service in another node (s3_client={s3_client})"
|
|
)
|
|
def test_s3_one_endpoint_loss(
|
|
self,
|
|
bucket,
|
|
s3_client: S3ClientWrapper,
|
|
simple_object_size: ObjectSize,
|
|
after_run_return_all_stopped_services,
|
|
cluster_state_controller: ClusterStateController,
|
|
):
|
|
# TODO: need to check that s3 gate is connected to localhost (such metric will be supported in 1.3)
|
|
|
|
with reporter.step("Stop one node and wait for rebalance connection of s3 gate to storage service"):
|
|
current_node = self.cluster.cluster_nodes[0]
|
|
cluster_state_controller.stop_service_of_type(current_node, StorageNode)
|
|
# waiting for rebalance connection of s3 gate to storage service
|
|
sleep(60)
|
|
|
|
file_path = generate_file(simple_object_size.value)
|
|
file_name = s3_helper.object_key_from_file_path(file_path)
|
|
with reporter.step("Put object into one bucket"):
|
|
put_object = s3_client.put_object(bucket, file_path)
|
|
s3_helper.check_objects_in_bucket(s3_client, bucket, expected_objects=[file_name])
|
|
|
|
@allure.title("After Pilorama.db loss on one node object is retrievable (s3_client={s3_client})")
|
|
def test_s3_one_pilorama_loss(
|
|
self,
|
|
s3_client: S3ClientWrapper,
|
|
simple_object_size: ObjectSize,
|
|
cluster_state_controller: ClusterStateController,
|
|
):
|
|
bucket = s3_client.create_bucket(
|
|
location_constraint="load-1-4",
|
|
grant_read="uri=http://acs.amazonaws.com/groups/global/AllUsers",
|
|
)
|
|
s3_helper.set_bucket_versioning(s3_client, bucket, VersioningStatus.ENABLED)
|
|
|
|
with reporter.step("Check bucket versioning"):
|
|
bucket_versioning = s3_client.get_bucket_versioning_status(bucket)
|
|
assert bucket_versioning == "Enabled", "Bucket should have enabled versioning"
|
|
|
|
file_path = generate_file(simple_object_size.value)
|
|
file_name = s3_helper.object_key_from_file_path(file_path)
|
|
|
|
object_versions = []
|
|
with reporter.step("Put object into one bucket"):
|
|
put_object = s3_client.put_object(bucket, file_path)
|
|
s3_helper.check_objects_in_bucket(s3_client, bucket, expected_objects=[file_name])
|
|
object_versions.append(put_object)
|
|
|
|
node_to_check = self.cluster.storage_nodes[0]
|
|
|
|
piloramas_list_before_removing = []
|
|
with reporter.step("Get list of all pilorama.db on shards"):
|
|
for shard in node_to_check.get_shards():
|
|
piloramas_list_before_removing.append(shard.pilorama)
|
|
|
|
with reporter.step("Check that all pilorama.db files exist on node"):
|
|
for pilorama in piloramas_list_before_removing:
|
|
assert node_to_check.is_file_exist(pilorama), f"File {pilorama} does not exist"
|
|
|
|
with reporter.step("Stop all storage nodes"):
|
|
cluster_state_controller.stop_services_of_type(StorageNode)
|
|
|
|
with reporter.step("Delete pilorama.db from one node"):
|
|
for pilorama in piloramas_list_before_removing:
|
|
node_to_check.delete_file(pilorama)
|
|
|
|
with reporter.step("Start all storage nodes"):
|
|
cluster_state_controller.start_all_stopped_services()
|
|
|
|
with reporter.step("Tick epoch to trigger sync and then wait for 1 minute"):
|
|
self.tick_epochs(1)
|
|
sleep(120)
|
|
|
|
with reporter.step("Get list of all pilorama.db after sync"):
|
|
for pilorama in piloramas_list_before_removing:
|
|
assert node_to_check.is_file_exist(pilorama), f"File {pilorama} does not exist"
|
|
|
|
with reporter.step("Check bucket versioning"):
|
|
bucket_versioning = s3_client.get_bucket_versioning_status(bucket)
|
|
assert bucket_versioning == "Enabled", "Bucket should have enabled versioning"
|
|
|
|
with reporter.step("Check list objects"):
|
|
objects_list = s3_client.list_objects(bucket)
|
|
assert objects_list, f"Expected not empty bucket"
|
|
|
|
with reporter.step("Delete the object from the bucket"):
|
|
delete_object = s3_client.delete_object(bucket, file_name)
|
|
assert "DeleteMarker" in delete_object.keys(), "Delete markers not found"
|
|
|
|
with reporter.step("Check list objects"):
|
|
objects_list = s3_client.list_objects_versions(bucket)
|
|
assert objects_list, f"Expected not empty bucket"
|
|
object_versions.append(delete_object["VersionId"])
|
|
|
|
# and now delete all versions of object (including Delete Markers)
|
|
with reporter.step("Delete all versions of the object from the bucket"):
|
|
for version in object_versions:
|
|
delete_object = s3_client.delete_object(bucket, file_name, version_id=version)
|
|
|
|
with reporter.step("Check list objects"):
|
|
objects_list = s3_client.list_objects_versions(bucket)
|
|
assert not objects_list, f"Expected empty bucket"
|
|
|
|
with reporter.step("Delete bucket"):
|
|
s3_client.delete_bucket(bucket)
|