On the MXL25641-HDK-6 CDRouter-UPnP-IPv46 (generic-ipv6-upnp) package the
job is killed by the 52-min CDRouter cutoff at ~test 61/81 (ipv6_upnp_210),
even though Mozart and Freedom complete the same 81-test package on the
same pipeline / same MR in ~45 min. The killed test was healthy and would
have passed in another ~42 s; only OSPv2 is affected.
Per-test execution times on OSPv2 are essentially identical to the last
known-good run on this job (72f7f516..bd9b1d94 bisect window) - the
regression is not in the tests themselves. CDRouter logs show that on
OSPv2, between every IGD2 and IPv6 test, CDRouter prints
NOTICE: The EUT is not connected to the test controller
NOTICE: The EUT did not reconnect within 30 seconds
before starting the next test. ~38 such 30-second waits accumulate
~18.5 min of dead time on top of the otherwise normal ~33-min runtime,
exactly closing the gap to the 52-min cutoff. The regression bisects to
e266163bea ("ci: cdrouter: switch USP testing from STOMP to MQTTv5 as
primary MTP", MR #2579) - the same change that exposed the half-dead
MQTTv5 socket bug tracked in PPW-1837. The CDRouter test controller
reaches the EUT via the configured USP MTP between tests; with MQTTv5 the
socket goes silently half-dead and the reachability probe times out at
30 s before CDRouter gives up and proceeds.
So lets workaround it by switching this one job (and its -Failing sibling
which runs the same package) to WebSocket MTP, mirroring what the
certification jobs already do (see 99912e34). Mozart and Freedom run the
same package without any inter-test waits and are intentionally left
untouched.
Revert this block once the system-level fix for the half-dead MQTTv5
socket (PPW-1837) lands.
References: PPW-1897, PPW-1837, PCF-2565
Assisted-by: claude:opus-4-7:1m-context
Signed-off-by: Petr Štetiar <petr.stetiar@prplfoundation.org>
dd6f757ff6