frostfs-testcases/pytest_tests/testsuites/metrics/test_shard_metrics.py

178 lines
7.4 KiB
Python
Raw Normal View History

2024-05-31 13:58:31 +00:00
import random
import re
import allure
import pytest
from frostfs_testlib import reporter
2024-06-25 13:23:25 +00:00
from frostfs_testlib.resources.error_patterns import OBJECT_NOT_FOUND
2024-05-31 13:58:31 +00:00
from frostfs_testlib.resources.wellknown_acl import EACL_PUBLIC_READ_WRITE
2024-06-25 13:23:25 +00:00
from frostfs_testlib.steps.cli.container import create_container
from frostfs_testlib.steps.cli.object import get_object, put_object
from frostfs_testlib.steps.metrics import check_metrics_counter
2024-05-31 13:58:31 +00:00
from frostfs_testlib.steps.node_management import node_shard_list, node_shard_set_mode
2024-06-25 13:23:25 +00:00
from frostfs_testlib.steps.storage_policy import get_nodes_with_object
2024-05-31 13:58:31 +00:00
from frostfs_testlib.storage.cluster import Cluster, ClusterNode
from frostfs_testlib.storage.controllers import ShardsWatcher
from frostfs_testlib.storage.dataclasses.wallet import WalletInfo
from frostfs_testlib.testing import parallel, wait_for_success
from frostfs_testlib.testing.cluster_test_base import ClusterTestBase
from frostfs_testlib.utils.file_utils import generate_file
class TestShardMetrics(ClusterTestBase):
@pytest.fixture()
@allure.title("Get two shards for set mode")
def two_shards_and_node(self, cluster: Cluster) -> tuple[str, str, ClusterNode]:
node = random.choice(cluster.cluster_nodes)
shards = node_shard_list(node.storage_node)
two_shards = random.sample(shards, k=2)
yield two_shards[0], two_shards[1], node
for shard in two_shards:
node_shard_set_mode(node.storage_node, shard, "read-write")
node_shard_list(node.storage_node)
@pytest.fixture()
@allure.title("Revert all shards mode")
def revert_all_shards_mode(self):
yield
parallel(self.set_shard_rw_mode, self.cluster.cluster_nodes)
def set_shard_rw_mode(self, node: ClusterNode):
watcher = ShardsWatcher(node)
shards = watcher.get_shards()
for shard in shards:
watcher.set_shard_mode(shard["shard_id"], mode="read-write")
watcher.await_for_all_shards_status(status="read-write")
@staticmethod
2024-06-25 13:23:25 +00:00
def get_error_count_from_logs(cluster_node: ClusterNode, object_path: str, object_name: str):
2024-05-31 13:58:31 +00:00
error_count = 0
try:
2024-06-25 13:23:25 +00:00
logs = cluster_node.host.get_filtered_logs("error count", unit="frostfs-storage")
2024-05-31 13:58:31 +00:00
# search error logs for current object
2024-06-25 13:23:25 +00:00
for error_line in logs.split("\n"):
2024-05-31 13:58:31 +00:00
if object_path in error_line and object_name in error_line:
result = re.findall(r'"error\scount":\s(\d+)', error_line)
error_count += sum(map(int, result))
except RuntimeError as e:
...
return error_count
@staticmethod
@wait_for_success(180, 30)
def get_object_path_and_name_file(oid: str, cid: str, node: ClusterNode) -> tuple[str, str]:
oid_path = f"{oid[0]}/{oid[1]}/{oid[2]}/{oid[3]}"
object_path = None
with reporter.step("Search object file"):
node_shell = node.storage_node.host.get_shell()
data_path = node.storage_node.get_data_directory()
all_datas = node_shell.exec(f"ls -la {data_path}/data | awk '{{ print $9 }}'").stdout.strip()
for data_dir in all_datas.replace(".", "").strip().split("\n"):
check_dir = node_shell.exec(
f" [ -d {data_path}/data/{data_dir}/data/{oid_path} ] && echo 1 || echo 0"
).stdout
if "1" in check_dir:
object_path = f"{data_path}/data/{data_dir}/data/{oid_path}"
object_name = f"{oid[4:]}.{cid}"
break
assert object_path is not None, f"{oid} object not found in directory - {data_path}/data"
return object_path, object_name
@allure.title("Metric for shard mode")
def test_shard_metrics_set_mode(self, two_shards_and_node: tuple[str, str, ClusterNode]):
metrics_counter = 1
shard1, shard2, node = two_shards_and_node
with reporter.step("Shard1 set to mode 'read-only'"):
node_shard_set_mode(node.storage_node, shard1, "read-only")
with reporter.step(f"Check shard metrics, 'the mode will change to 'READ_ONLY'"):
2024-06-25 13:23:25 +00:00
check_metrics_counter(
[node],
counter_exp=metrics_counter,
command="frostfs_node_engine_mode_info",
mode="READ_ONLY",
shard_id=shard1,
2024-05-31 13:58:31 +00:00
)
with reporter.step("Shard2 set to mode 'degraded-read-only'"):
node_shard_set_mode(node.storage_node, shard2, "degraded-read-only")
with reporter.step(f"Check shard metrics, 'the mode will change to 'DEGRADED_READ_ONLY'"):
2024-06-25 13:23:25 +00:00
check_metrics_counter(
[node],
counter_exp=metrics_counter,
2024-05-31 13:58:31 +00:00
command="frostfs_node_engine_mode_info",
mode="DEGRADED_READ_ONLY",
shard_id=shard2,
)
with reporter.step("Both shards set to mode 'read-write'"):
for shard in [shard1, shard2]:
node_shard_set_mode(node.storage_node, shard, "read-write")
2024-06-25 13:23:25 +00:00
with reporter.step(f"Check shard metrics, 'the mode will change to 'READ_WRITE'"):
for shard in [shard1, shard2]:
check_metrics_counter(
[node],
counter_exp=metrics_counter,
command="frostfs_node_engine_mode_info",
mode="READ_WRITE",
shard_id=shard,
2024-05-31 13:58:31 +00:00
)
@allure.title("Metric for error count on shard")
def test_shard_metrics_error_count(
self, max_object_size: int, default_wallet: WalletInfo, cluster: Cluster, revert_all_shards_mode
):
file_path = generate_file(round(max_object_size * 0.8))
with reporter.step(f"Create container"):
cid = create_container(
wallet=default_wallet,
shell=self.shell,
2024-06-25 13:23:25 +00:00
endpoint=cluster.default_rpc_endpoint,
2024-05-31 13:58:31 +00:00
rule="REP 1 CBF 1",
basic_acl=EACL_PUBLIC_READ_WRITE,
)
with reporter.step("Put object"):
2024-06-25 13:23:25 +00:00
oid = put_object(default_wallet, file_path, cid, self.shell, cluster.default_rpc_endpoint)
2024-05-31 13:58:31 +00:00
with reporter.step("Get object nodes"):
2024-06-25 13:23:25 +00:00
object_storage_nodes = get_nodes_with_object(cid, oid, self.shell, cluster.storage_nodes)
object_nodes = [
cluster_node
for cluster_node in cluster.cluster_nodes
if cluster_node.storage_node in object_storage_nodes
]
node = random.choice(object_nodes)
2024-05-31 13:58:31 +00:00
with reporter.step("Search object in system."):
object_path, object_name = self.get_object_path_and_name_file(oid, cid, node)
with reporter.step("Block read file"):
node.host.get_shell().exec(f"chmod a-r {object_path}/{object_name}")
with reporter.step("Get object, expect error"):
2024-06-25 13:23:25 +00:00
with pytest.raises(RuntimeError, match=OBJECT_NOT_FOUND):
2024-05-31 13:58:31 +00:00
get_object(
wallet=default_wallet,
cid=cid,
oid=oid,
shell=self.shell,
endpoint=node.storage_node.get_rpc_endpoint(),
)
with reporter.step(f"Get shard error count from logs"):
2024-06-25 13:23:25 +00:00
counter = self.get_error_count_from_logs(node, object_path, object_name)
2024-05-31 13:58:31 +00:00
with reporter.step(f"Check shard error metrics"):
2024-06-25 13:23:25 +00:00
check_metrics_counter([node], counter_exp=counter, command="frostfs_node_engine_errors_total")