Race condition when doing various ALTERs in a system under a heavy load #24134
Description
Describe the bug
New stress 01900_alter_race_condition_stress that can be found in #24132 fails when server is run with thread fuzzer enabled. This test was created based on the fails seen in RBAC TestFlows module (https://github.com/ClickHouse/ClickHouse/blob/master/tests/testflows/rbac/tests/privileges/alter/alter_column.py).
Does it reproduce on recent release?
Yes.
How to reproduce
Enable thread fuzzer in current environment
export THREAD_FUZZER_CPU_TIME_PERIOD_US=1000
export THREAD_FUZZER_SLEEP_PROBABILITY=0.1
export THREAD_FUZZER_SLEEP_TIME_US=100000
export THREAD_FUZZER_pthread_mutex_lock_BEFORE_MIGRATE_PROBABILITY=1
export THREAD_FUZZER_pthread_mutex_lock_AFTER_MIGRATE_PROBABILITY=1
export THREAD_FUZZER_pthread_mutex_unlock_BEFORE_MIGRATE_PROBABILITY=1
export THREAD_FUZZER_pthread_mutex_unlock_AFTER_MIGRATE_PROBABILITY=1
export THREAD_FUZZER_pthread_mutex_lock_BEFORE_SLEEP_PROBABILITY=0.001
export THREAD_FUZZER_pthread_mutex_lock_AFTER_SLEEP_PROBABILITY=0.001
export THREAD_FUZZER_pthread_mutex_unlock_BEFORE_SLEEP_PROBABILITY=0.001
export THREAD_FUZZER_pthread_mutex_unlock_AFTER_SLEEP_PROBABILITY=0.001
export THREAD_FUZZER_pthread_mutex_lock_BEFORE_SLEEP_TIME_US=10000
export THREAD_FUZZER_pthread_mutex_lock_AFTER_SLEEP_TIME_US=10000
export THREAD_FUZZER_pthread_mutex_unlock_BEFORE_SLEEP_TIME_US=10000
export THREAD_FUZZER_pthread_mutex_unlock_AFTER_SLEEP_TIME_US=10000
start server so that it gets all the environment variables
sudo -E -u clickhouse ./build/programs/clickhouse-server --config-file=/etc/clickhouse-server/config.xml
repeatedly run new tests until fail
tests/clickhouse-test -b build/programs/clickhouse -c build/programs/clickhouse-client --queries tests/queries/ 01900_alter_race_condition
Expected behavior
Test should not fail with thread fuzzer enabled.