Closed kostasrim closed 1 week ago
Impact on a full run: https://github.com/dragonflydb/dragonfly/actions/runs/11796502461
(I will post the results once it completes)
x86 debug 28 minutes now takes 22 arm debug 42 minutes now takes 34 x86 release 28 minutes now takes 19 arm release 46 minutes now takes 28
Savings on a full run 24minutes (because arm tests run sequentially)
https://github.com/dragonflydb/dragonfly/actions/runs/11797723885/job/32862315798 vs https://github.com/dragonflydb/dragonfly/actions/runs/11793906726/job/32850430451
At some point I will take care of:
180.20s call dragonfly/replication_test.py::test_replicaof_reject_on_load[df_seeder_factory0-df_factory0]
It's the next low hanging fruit
We have a few very slow tests on the CI. Specifically:
This PR changes:
It refactors the
test_big_value_serialization
. First, the test was incorrect, in fact for some reason it did not even run with big value serialization flag set. Second, the test usedexecute_command
in a loop which took asubstantial
amount of time. The changes on my machine reduce the total running time of the test from4 * 170 seconds
onaverage
to a staggering1 minute and 10 seconds
for all 4.Also notice, that we don't really need to test big value serialization on a stress test. It's redundant because when we set the value to a small number (4096 etc) it will have the same effect regardless of the stress load. I removed that one test case, saving as roughly (235 * 2) seconds