Gleb Natapov comments

Results 119 comments of


                                            Gleb Natapov

Write handler blocks topology barriers for the duration of the timeout if live host is detected by gossip as having restarted

I do not think there is anything wrong with the SP logic. Tests either should not depend on read/write timeout or test.py should not set it to 5 minutes.

Write handler blocks topology barriers for the duration of the timeout if live host is detected by gossip as having restarted

> If #15924 is caused by this -- then it's a problem. We don't drop below number of required replicas by CL in that test so no writes should timeout...

Write handler blocks topology barriers for the duration of the timeout if live host is detected by gossip as having restarted

> I don't understand. The issue is that user writes are timing out when they should not. What issue? This issue? I do not see any indication that here something...

Write handler blocks topology barriers for the duration of the timeout if live host is detected by gossip as having restarted

> The problem is that legit passing tests take 5min for no good reason. patient: Doctor, if I do this it hurts doctor: well, then do not do that! If...

Write handler blocks topology barriers for the duration of the timeout if live host is detected by gossip as having restarted

> > But I still do not understand what exactly happened in this particular case. > > The test is just waiting for tablet rebuild to complete at this point,...

test.py: teardown may race with ongoing test operation, causing IPs to be released prematurely

> @gleb-cloudius, @kbr-scylla, Do you have a reproducer for this, some test example or an existing test? Sorry I missed it. I hit the problem when something inside `test_coordinator_queue_management.py` failed...

dtest: topology_test.TestTopology.test_crash_during_decommission got unexpected errors: raft::request_aborted (Request is aborted by a caller)

It looks like the test starts a decommission and restarts some other node in a loop until the decommission fails. Why not re-write the test to stop the node, wait...

Decommission failed with error: remote replied with status code 500 Internal Server Error and coredumps due to possible reason that new node marked UN earler during bootstrap and replace with MV/SI

> And according to node log Scylla initialization completed at: > >``` >2024-03-09T18:41:33.002+00:00 longevity-mv-si-4d-master-db-node-6eb71640-6 !INFO | scylla[6406]: [shard 0:main] >init - Scylla version 5.5.0~dev-0.20240307.19bcea6216f4 initialization completed. >``` @aleksbykov where do...

Decommission failed with error: remote replied with status code 500 Internal Server Error and coredumps due to possible reason that new node marked UN earler during bootstrap and replace with MV/SI

Reproducer: ```diff diff --git a/test/topology_custom/test_mv_decommission.py b/test/topology_custom/test_mv_decommission.py new file mode 100644 index 0000000000..2b7ce1664d --- /dev/null +++ b/test/topology_custom/test_mv_decommission.py @@ -0,0 +1,74 @@ +import asyncio +import pytest +import time +import logging + +from...

Decommission failed with error: remote replied with status code 500 Internal Server Error and coredumps due to possible reason that new node marked UN earler during bootstrap and replace with MV/SI

Suspected culprit is f2ff701489995fdc84ea7ca0addbec2918c35be4. The problem is that MV code assumes that base and view erm come from the same topology, but after the patch above erms update for tables...