clickhouse-test: replace clickhouse-driver with http interface (via http.client)#30065
Conversation
Before ClickHouse#29856 `CREATE DATABASE` overwrites it. Reported-by: @amosbird
Since it will not configure testcase args and fail eventually, and later we have a try/catch anyway, this should be enough.
|
@Mergifyio update |
|
Command
|
|
@azat Need to merge this PR ASAP, because currently It is unpleasant and makes me to debug with log messages with Fatal level |
Interesting, 30 seconds wasn't enough to create database, and trace_log is empty (since it is ASAN build and it has only Memory related profiling, this is OK). Server logs: Maybe disabling
Same problem, and here we should have traces: But trace_log is also empty (for this query_id, or thread_id+timestamp), hm. Let's try disabling
|
Agree, actually all of the failures was classified and the reason for them is not clickhouse-test.
Yes, this has been fixed in a separate patch here, if you want, you can cherry-pick it. |
|
JSONEachRow cannot be parsed with a simple json.loads(), instead it should be passed to json.loads() line by line. Fixes: ClickHouse#30065
Set --server-check-retries to 90 (and this is ~45 seconds), since right now sometimes it is not enough [1]. [1]: https://clickhouse-test-reports.s3.yandex.net/30191/0e34a9d550cfe6924fe575871f36c44dd44acdaa/functional_stateless_tests_(memory).html#fail1 And the reason I guess is clickhouse-test had been rewritten to http.client in ClickHouse#30065, and since now it does not need to execute clickhouse-client binary, which in debug/sanitizers builds can take also sometime. That said that with clickhouse-client for hung check it was not 15 seconds, but more (each clickhouse-client requires 0.6sec with sanitizers for simple SELECT 1, while w/o 0.1second, also too much should be optimized)
Changelog category (leave one):
Cons of clickhouse-driver:
this is not the same, so we need to set timeout by ourselves)
Cc: @alesapin
Cc: @tavplubix