optimize_move_to_prewhere = 0|1

prefer_localhost_replica = 0|1

group_by_two_level_threshold
group_by_two_level_threshold_bytes

With 10% probability set to very low value (to activate always) and with 10% set to very high value (to never activate)

distributed_aggregation_memory_efficient = 0|1

allow_experimental_parallel_reading_from_replicas = 0|1

max_insert_threads with probability 1/2 set it uniformly random 1..16

output_format_parallel_formatting = 0|1

PS. We can adapt existing tests then add more randomization in subsequent PR.

@den-crane @SaltTan Maybe you can recall some problematic settings from your experience.

We should print changed settings somewhere in logs to simplify debugging

Love the idea but do you really want to introduce them all at once? The last similar thing done was to run under a random timezone and it took weeks and many PRs to stabilize things.

I'm ok if it will take weeks.

@Mergifyio update

update

✅ Branch has been successfully updated

@den-crane @SaltTan Maybe you can recall some problematic settings from your experience.

I'm sorry it took me so long.

• optimize*
• enable*
• use*
• compile_aggregate_expressions
• compile_expressions
• There was an issue with 'read_backoff' once.

@Mergifyio update

update

✅ Branch has been successfully updated