action #12922
closedhard-chained jobs running on same worker one after another
0%
Description
user story¶
For execution of chained jobs on workers without support for publishing hard disk image files we want one job to be executed after another one on the same machine without any other job interfering e.g. to upgrade scenarios
acceptance criteria¶
- AC1: Scheduler ensures that two hard-chained jobs run on the same worker directly one after another without releasing the worker in between
tasks¶
- maybe this is already solved and someone can share insight and tell how it's done?
- Probably what should work is to set a worker class so specific that the same worker is ensured to be used for the second part of the job chain but this is not enough as the worker could still take another job in between.
further details¶
We are trying to implement upgrade test suites on s390x zvm where we have to prepare the SUT with an older version, e.g. SLE12-SP1, and then want to upgrade to SLE12-SP2 but as we use DASD devices we can not allow another job to interfere and mangle with the disk before the upgrade scenario starts on the same disk. For now we are trying to just run multiple test modules again within one scenario, e.g.
load_inst_tests();
UPGRADE=1
load_inst_tests();
but this is very complicated as one has to keep in mind that variables are evaluated during test scheduling time as well as during test execution time so in the end probably variables like "UPGRADE" need to be set in both context. Still, needles matching on these variables would also need to be adjusted or the whole unregister_needles
approach needs to be reworked to keep the needles for the old version around and only dump then later after switching to the upgrade target, e.g. 12-SP2.