test: the mariadb pair's client write-resume bound of 3 s is at the edge on the lossy link #148

New issue

Closed

opened 2026-10-14 12:57:48 -03:00 by marcos-mendez · 0 comments

marcos-mendez commented

2026-10-14 12:57:48 -03:00

(Migrated from github.com)

tests/test_mariadb_netns.py:223 asserts first_ok_after_s - carried_s < 3.0. On the design link (250 ms, 2 % loss) the result is at the edge: 3.01 s on main (run 38489547963, ffa494b) and 3.09 s on keel#147 (run 38591193702). keel's own part stays below its bound (writable_after_carried_s 2.11 s < 3.0 s). The rest is the test client: its sample interval and one TCP retransmission (about 1 s) after a lost packet.

Decision needed: keep 3 s as the client-side target and make keel faster, or measure the client side with the loss retransmission in the bound. Until then the job fails about one run in three.

tests/test_mariadb_netns.py:223 asserts first_ok_after_s - carried_s < 3.0. On the design link (250 ms, 2 % loss) the result is at the edge: 3.01 s on main (run 38489547963, ffa494b) and 3.09 s on keel#147 (run 38591193702). keel's own part stays below its bound (writable_after_carried_s 2.11 s < 3.0 s). The rest is the test client: its sample interval and one TCP retransmission (about 1 s) after a lost packet. Decision needed: keep 3 s as the client-side target and make keel faster, or measure the client side with the loss retransmission in the bound. Until then the job fails about one run in three.