Files
impala/tests/custom_cluster/test_process_failures.py
Riza Suminto 55feffb41b IMPALA-13850 (part 1): Wait until CatalogD active before resetting
In HA mode, CatalogD initialization can fail to complete within
reasonable time. Log messages showed that CatalogD is blocked trying to
acquire "CatalogServer.catalog_lock_" when calling
CatalogServer::UpdateActiveCatalogd() during statestore subscriber
registration. catalog_lock_ was held by GatherCatalogUpdatesThread which
is calling GetCatalogDelta(), which waits for the java lock versionLock_
which is held by the thread doing CatalogServiceCatalog.reset().

This patch remove catalog reset in JniCatalog constructor. In turn,
catalogd-server.cc is now responsible to trigger the metadata
reset (Invaidate Metadata) only if:

1. It is the active CatalogD, and
2. Gathering thread has collect the first topic update or CatalogD is
   set with catalog_topic_mode other than "minimal".

The later prerequisite is to ensure that all coordinators are not
blocked waiting for full topic update in on-demand metadata mode. This
is all managed by a new thread method TriggerResetMetadata that monitor
and trigger the initial reset metadata.

Note that this is a behavior change in on-demand catalog
mode (catalog_topic_mode=minimal). Previously, on-demand catalog mode
will send full database list in its first catalog topic update. This
behavior change is OK since coordinator can request metadata on-demand.

After this patch, catalog-server.active-status and /healthz page can
turn into true and OK respectively even if the very first metadata reset
is still ongoing. Observer that cares about having fully populated
metadata should check other metrics such as catalog.num-db,
catalog.num-tables, or /catalog page content.

Updated start-impala-cluster.py readiness check to wait for at least 1
table to be seen by coordinators, except during create-load-data.sh
execution (there is no table yet) and when use_local_catalog=true (local
catalog cache does not start with any table). Modified startup flag
checking from reading the actual command line args to reading the
'/varz?json' page of the daemon. Cleanup impala_service.py to fix some
flake8 issues.

Slightly update TestLocalCatalogCompactUpdates::test_restart_catalogd so
that unique_database cleanup is successful.

Testing:
- Refactor test_catalogd_ha.py to reduce repeated code, use
  unique_database fixture, and additionally validate /healthz page of
  both active and standby catalogd. Changed it to test using hs2
  protocol by default.
- Run and pass test_catalogd_ha.py and test_concurrent_ddls.py.
- Pass core tests.

Change-Id: I58cc66dcccedb306ff11893f2916ee5ee6a3efc1
Reviewed-on: http://gerrit.cloudera.org:8080/22634
Reviewed-by: Riza Suminto <riza.suminto@cloudera.com>
Tested-by: Riza Suminto <riza.suminto@cloudera.com>
2025-04-17 01:59:54 +00:00

237 lines
10 KiB
Python

# Licensed to the Apache Software Foundation (ASF) under one
# or more contributor license agreements. See the NOTICE file
# distributed with this work for additional information
# regarding copyright ownership. The ASF licenses this file
# to you under the Apache License, Version 2.0 (the
# "License"); you may not use this file except in compliance
# with the License. You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an
# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
# KIND, either express or implied. See the License for the
# specific language governing permissions and limitations
# under the License.
from __future__ import absolute_import, division, print_function
from builtins import range
import pytest
from tests.common.custom_cluster_test_suite import (
DEFAULT_CLUSTER_SIZE,
CustomClusterTestSuite)
from tests.common.impala_connection import RUNNING
from tests.common.test_result_verifier import error_msg_startswith
# The exact query doesn't matter much for these tests, just want a query that touches
# data on all nodes.
QUERY = "select count(l_comment) from tpch.lineitem"
# The default number of statestore subscribers should be the default cluster size plus
# one for the catalogd.
DEFAULT_NUM_SUBSCRIBERS = DEFAULT_CLUSTER_SIZE + 1
class TestProcessFailures(CustomClusterTestSuite):
"""Validates killing and restarting impalad processes between query executions."""
@classmethod
def setup_class(cls):
if cls.exploration_strategy() != 'exhaustive':
pytest.skip('These tests only run in exhaustive')
super(TestProcessFailures, cls).setup_class()
@pytest.mark.execute_serially
def test_restart_coordinator(self):
"""Restarts the coordinator between queries."""
impalad = self.cluster.get_any_impalad()
client = impalad.service.create_hs2_client()
self.execute_query_expect_success(client, QUERY)
statestored = self.cluster.statestored
impalad.restart()
statestored.service.wait_for_live_subscribers(DEFAULT_NUM_SUBSCRIBERS, timeout=60)
# Reconnect
client = impalad.service.create_hs2_client()
impalad.service.wait_for_metric_value('catalog.ready', 1, timeout=60)
self.execute_query_expect_success(client, QUERY)
@pytest.mark.execute_serially
@CustomClusterTestSuite.with_args(num_exclusive_coordinators=1,
impalad_args="--status_report_max_retry_s=600 --status_report_interval_ms=1000")
def test_kill_coordinator(self):
""""Tests that when a coordinator running multiple queries is killed, all
running fragments on executors are cancelled."""
impalad = self.cluster.impalads[0]
client = impalad.service.create_hs2_client()
assert client is not None
# A query which is cancelable and takes long time to execute
query = "select * from tpch.lineitem t1, tpch.lineitem t2, tpch.lineitem t3 " \
"where t1.l_orderkey = t2.l_orderkey and t1.l_orderkey = t3.l_orderkey and " \
"t3.l_orderkey = t2.l_orderkey order by t1.l_orderkey, t2.l_orderkey, " \
"t3.l_orderkey limit 300"
num_concurrent_queries = 3
handles = []
# Run num_concurrent_queries asynchronously
for _ in range(num_concurrent_queries):
handles.append(client.execute_async(query))
# Wait for the queries to start running
for handle in handles:
client.wait_for_impala_state(handle, RUNNING, 1000)
# Kill the coordinator
impalad.kill()
# Assert that all executors have 0 in-flight fragments
for i in range(1, len(self.cluster.impalads)):
self.cluster.impalads[i].service.wait_for_metric_value(
"impala-server.num-fragments-in-flight", 0, timeout=30)
@pytest.mark.execute_serially
@CustomClusterTestSuite.with_args(
impalad_args='--use_local_catalog',
catalogd_args='--catalog_topic_mode=minimal')
def test_restart_statestore(self):
"""Tests the cluster still functions when the statestore dies."""
impalad = self.cluster.get_any_impalad()
client = impalad.service.create_hs2_client()
statestored = self.cluster.statestored
statestored.kill()
impalad.service.wait_for_metric_value(
'statestore-subscriber.connected', False, timeout=60)
# impalad should still see the same number of live backends
assert impalad.service.get_num_known_live_backends() == DEFAULT_CLUSTER_SIZE
self.execute_query_expect_success(client, QUERY)
# Reconnect
statestored.start()
impalad.service.wait_for_metric_value(
'statestore-subscriber.connected', True, timeout=60)
statestored.service.wait_for_live_subscribers(DEFAULT_NUM_SUBSCRIBERS, timeout=60)
# Wait for the number of live backends to reach the cluster size. Even though
# all backends have subscribed to the statestore, this impalad may not have
# received the update yet.
impalad.service.wait_for_num_known_live_backends(DEFAULT_CLUSTER_SIZE, timeout=60)
self.execute_query_expect_success(client, QUERY)
@pytest.mark.execute_serially
def test_kill_restart_worker(self):
"""Verifies a worker is able to be killed."""
impalad = self.cluster.get_any_impalad()
client = impalad.service.create_hs2_client()
self.execute_query_expect_success(client, QUERY)
# select a different impalad and restart it
worker_impalad = self.cluster.get_different_impalad(impalad)
# Start executing a query. It will be cancelled due to a killed worker.
handle = client.execute_async(QUERY)
statestored = self.cluster.statestored
worker_impalad.kill()
# First wait until the the statestore realizes the impalad has gone down.
statestored.service.wait_for_live_subscribers(DEFAULT_NUM_SUBSCRIBERS - 1, timeout=60)
# Wait until the impalad registers another instance went down.
impalad.service.wait_for_num_known_live_backends(DEFAULT_CLUSTER_SIZE - 1, timeout=60)
# Wait until the in-flight query has been cancelled.
# The in-flight query should have been cancelled, reporting a failed worker as the
# cause. The query may have been cancelled because the state store detected a failed
# node, or because a stream sender failed to establish a thrift connection. It is
# non-deterministic which of those paths will initiate cancellation, but in either
# case the query status should include the failed (or unreachable) worker.
query_id = client.handle_id(handle)
error_state = "Failed due to unreachable impalad"
assert impalad.service.wait_for_query_status(query_id, error_state)
assert error_msg_startswith(client.get_log(handle), error_state, query_id)
# Assert that the query status on the query profile web page contains the expected
# failed hostport.
failed_hostport = "%s:%s" % (worker_impalad.service.hostname,
worker_impalad.service.krpc_port)
query_profile_page = impalad.service.read_query_profile_page(query_id)
assert failed_hostport in query_profile_page,\
"Query status did not contain expected hostport %s\n\n%s" % (failed_hostport,
query_profile_page)
# Should work fine even if a worker is down.
self.execute_query_expect_success(client, QUERY)
# Bring the worker back online and validate queries still work.
worker_impalad.start()
statestored.service.wait_for_live_subscribers(DEFAULT_NUM_SUBSCRIBERS, timeout=60)
worker_impalad.service.wait_for_metric_value('catalog.ready', True, timeout=60)
self.execute_query_expect_success(client, QUERY)
@pytest.mark.execute_serially
@CustomClusterTestSuite.with_args(
impalad_args='--use_local_catalog',
catalogd_args='--catalog_topic_mode=minimal')
@pytest.mark.xfail(run=False, reason="IMPALA-9848")
def test_restart_catalogd(self):
# Choose a random impalad verify a query can run against it.
impalad = self.cluster.get_any_impalad()
client = impalad.service.create_hs2_client()
self.execute_query_expect_success(client, QUERY)
# Kill the catalogd.
catalogd = self.cluster.catalogd
catalogd.kill()
# The statestore should detect the catalog service has gone down.
statestored = self.cluster.statestored
statestored.service.wait_for_live_subscribers(DEFAULT_NUM_SUBSCRIBERS - 1, timeout=60)
# We should still be able to execute queries using the impalad.
self.execute_query_expect_success(client, QUERY)
# Start the catalog service back up.
catalogd.start()
statestored.service.wait_for_live_subscribers(DEFAULT_NUM_SUBSCRIBERS, timeout=60)
# Execute a query against the catalog service.
impalad.service.wait_for_metric_value('catalog.ready', True, timeout=60)
self.execute_query_expect_success(client, QUERY)
@pytest.mark.execute_serially
def test_restart_all_impalad(self):
"""Restarts all the impalads and runs a query"""
impalad = self.cluster.get_any_impalad()
client = impalad.service.create_hs2_client()
self.execute_query_expect_success(client, QUERY)
# Kill each impalad and wait for the statestore to register the failures.
for impalad_proc in self.cluster.impalads:
impalad_proc.kill()
statestored = self.cluster.statestored
# There should be 1 remining subscriber, the catalogd
statestored.service.wait_for_live_subscribers(1, timeout=60)
# Start each impalad back up and wait for the statestore to see them.
for impalad_proc in self.cluster.impalads:
impalad_proc.start()
# The impalads should re-register with the statestore on restart at which point they
# can execute queries.
statestored.service.wait_for_live_subscribers(DEFAULT_NUM_SUBSCRIBERS, timeout=60)
for impalad in self.cluster.impalads:
impalad.service.wait_for_num_known_live_backends(DEFAULT_CLUSTER_SIZE, timeout=60)
impalad.service.wait_for_metric_value('catalog.ready', True, timeout=60)
client = impalad.service.create_hs2_client()
self.execute_query_expect_success(client, QUERY)
# Make sure the catalog service is actually back up by executing an operation
# against it.
self.execute_query_expect_success(client, QUERY)