mirror of
https://github.com/ClickHouse/ClickHouse.git
synced 2024-11-28 18:42:26 +00:00
05a8c73eb9
Found with stress tests for 00626_replace_partition_from_table_zookeeper [1]:
2021.03.15 00:59:48.200106 [ 27417 ] {0f47dbeb-938a-4560-8408-a7cc7b3bafb8} <Trace> ContextAccess (default): Access granted: CREATE TABLE ON test_31.dst_r1
...
2021.03.15 00:59:48.403227 [ 27417 ] {0f47dbeb-938a-4560-8408-a7cc7b3bafb8} <Debug> test_31.dst_r1: This table /clickhouse/test_00626/dst_1 is already created, will add new replica
2021.03.15 00:59:48.736450 [ 83006 ] {b2db1355-3ec3-4e3a-9c79-f93f27c6e658} <Trace> ContextAccess (default): Access granted: CREATE TABLE ON test_31.dst_r2
...
2021.03.15 00:59:48.851768 [ 83006 ] {b2db1355-3ec3-4e3a-9c79-f93f27c6e658} <Debug> test_31.dst_r2: This table /clickhouse/test_00626/dst_1 is already created, will add new replica
...
2021.03.15 00:59:48.919059 [ 366 ] {} <Debug> test_31.dst_r2 (ReplicatedMergeTreeQueue): Loading queue from /clickhouse/test_00626/dst_1/replicas/2/queue
2021.03.15 00:59:48.919948 [ 366 ] {} <Debug> test_31.dst_r2 (ReplicatedMergeTreeQueue): Having 3 queue entries to load, 0 entries already loaded.
2021.03.15 00:59:48.921833 [ 366 ] {} <Trace> test_31.dst_r2 (ReplicatedMergeTreeQueue): Loaded queue
...
2021.03.15 00:59:51.904230 [ 246952 ] {59753eea-3896-45ca-8625-fdaa094ee9ef} <Trace> ContextAccess (default): Access granted: SYSTEM SYNC REPLICA ON test_31.dst_r2
...
2021.03.15 01:04:51.913683 [ 246952 ] {59753eea-3896-45ca-8625-fdaa094ee9ef} <Error> InterpreterSystemQuery: SYNC REPLICA test_31.dst_r2: Timed out!
[1]: https://clickhouse-test-reports.s3.yandex.net/21716/402bf77783cbda48a9ee1b748bfce3c52ef8fe11/stress_test_(memory)/test_run.txt.out.log
But the problem is more generic, so fix all tests.
71 lines
2.7 KiB
Bash
Executable File
71 lines
2.7 KiB
Bash
Executable File
#!/usr/bin/env bash
|
|
|
|
CURDIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
|
|
# shellcheck source=../shell_config.sh
|
|
. "$CURDIR"/../shell_config.sh
|
|
|
|
$CLICKHOUSE_CLIENT --query "DROP TABLE IF EXISTS concurrent_mutate_kill"
|
|
|
|
$CLICKHOUSE_CLIENT --query "CREATE TABLE concurrent_mutate_kill (key UInt64, value String) ENGINE = ReplicatedMergeTree('/clickhouse/tables/$CLICKHOUSE_TEST_ZOOKEEPER_PREFIX/concurrent_mutate_kill', '1') ORDER BY key PARTITION BY key % 100 SETTINGS max_replicated_mutations_in_queue=1000, number_of_free_entries_in_pool_to_execute_mutation=0,max_replicated_merges_in_queue=1000"
|
|
|
|
$CLICKHOUSE_CLIENT --query "INSERT INTO concurrent_mutate_kill SELECT number, toString(number) FROM numbers(1000000)"
|
|
|
|
function alter_thread
|
|
{
|
|
while true; do
|
|
TYPE=$($CLICKHOUSE_CLIENT --query "SELECT type FROM system.columns WHERE table='concurrent_mutate_kill' and database='${CLICKHOUSE_DATABASE}' and name='value'")
|
|
if [ "$TYPE" == "String" ]; then
|
|
$CLICKHOUSE_CLIENT --query "ALTER TABLE concurrent_mutate_kill MODIFY COLUMN value UInt64 SETTINGS replication_alter_partitions_sync=2"
|
|
else
|
|
$CLICKHOUSE_CLIENT --query "ALTER TABLE concurrent_mutate_kill MODIFY COLUMN value String SETTINGS replication_alter_partitions_sync=2"
|
|
fi
|
|
done
|
|
}
|
|
|
|
function kill_mutation_thread
|
|
{
|
|
while true; do
|
|
# find any mutation and kill it
|
|
mutation_id=$($CLICKHOUSE_CLIENT --query "SELECT mutation_id FROM system.mutations WHERE is_done=0 and database='${CLICKHOUSE_DATABASE}' and table='concurrent_mutate_kill' LIMIT 1")
|
|
if [ ! -z "$mutation_id" ]; then
|
|
$CLICKHOUSE_CLIENT --query "KILL MUTATION WHERE mutation_id='$mutation_id' and table='concurrent_mutate_kill' and database='${CLICKHOUSE_DATABASE}'" 1> /dev/null
|
|
sleep 1
|
|
fi
|
|
done
|
|
}
|
|
|
|
|
|
export -f alter_thread;
|
|
export -f kill_mutation_thread;
|
|
|
|
TIMEOUT=30
|
|
|
|
timeout $TIMEOUT bash -c alter_thread 2> /dev/null &
|
|
timeout $TIMEOUT bash -c kill_mutation_thread 2> /dev/null &
|
|
|
|
wait
|
|
|
|
$CLICKHOUSE_CLIENT --query "SYSTEM SYNC REPLICA concurrent_mutate_kill"
|
|
|
|
# with timeout alter query can be not finished yet, so to execute new alter
|
|
# we use retries
|
|
counter=0
|
|
while true; do
|
|
if $CLICKHOUSE_CLIENT --query "ALTER TABLE concurrent_mutate_kill MODIFY COLUMN value Int64 SETTINGS replication_alter_partitions_sync=2" 2> /dev/null ; then
|
|
break
|
|
fi
|
|
|
|
if [ "$counter" -gt 120 ]
|
|
then
|
|
break
|
|
fi
|
|
sleep 0.5
|
|
counter=$(($counter + 1))
|
|
done
|
|
|
|
$CLICKHOUSE_CLIENT --query "SHOW CREATE TABLE concurrent_mutate_kill"
|
|
$CLICKHOUSE_CLIENT --query "OPTIMIZE TABLE concurrent_mutate_kill FINAL"
|
|
$CLICKHOUSE_CLIENT --query "SELECT sum(value) FROM concurrent_mutate_kill"
|
|
|
|
$CLICKHOUSE_CLIENT --query "DROP TABLE IF EXISTS concurrent_mutate_kill"
|