Some stress tests fail with 'Cannot start clickhouse-server', could this be related to the change ?

@tavplubix is this PR valid way to enable ZK fault injections for inserts in tests?

Yep, it looks valid

Test failures:

• test_consistent_parts_after_clone_replica/test.py::test_inconsistent_parts_if_drop_while_replica_not_active - cluster failed to start for a reason
• test_read_only_table/test.py::test_restart_zookeeper(link) - can be non-deterministic, - we hardly can guarantee that server'll reconnect in 5 secs timeout to another keeper node (despite doing our best), technically, I guess we only can measure the time after which we can successfully execute INSERT:

        # ClickHouse should +- immediately reconnect to another zookeeper node
        cluster.stop_zookeeper_nodes([node1_zk])
        time.sleep(5)
    
        for table_id in range(NUM_TABLES):
>           node1.query(
                f"INSERT INTO test_table_{table_id} VALUES (6), (7), (8), (9), (10);"
            )

The simplest we can do - is to increase timeout. A bit more sophisticated - retries for inserts on test side. We could use keeper retries on insert as well, but then we need to get from query_log time to execute the query - it'll make the test more complicated. @tavplubix WDYT?

time.sleep(5) never works, we should replace it with something like node.query_with_retry

time.sleep(5) never works, we should replace it with something like node.query_with_retry

Fixed in #43326