Recently we noticed that the "check-tempest-dsvm-multinode-full" job, which is non-voting in nova's check jobs, was failing in a new way.
Some notes about how the failure was diagnosed:
- Job is failing
- why? find a signature for the failure, eg
File "tempest/api/compute/admin/test_live_migration.py", line 116, in test_live_block_migration"
AND build_name:"check-tempest-dsvm-multinode-full"
AND project:"openstack/nova"
- Raise a bug: https://bugs.launchpad.net/nova/+bug/1463747
- How often is the job failing?
build_name:"check-tempest-dsvm-multinode-full"
AND message:"Finished: SUCCESS"
vs
build_name:"check-tempest-dsvm-multinode-full"
AND message:"Finished: FAILURE"
(also for this particular job: https://jogo.github.io/gate/multinode.html <-- chrome browser)
- what's the actual cause? check the appropriate logs (in this case n-cpu) and find a signature there.
- logstash.o.o that:
message:"TypeError: string indices must be integers"
AND tags:"screen-n-cpu.txt"
-
If this is a recently-occuring failure only, look at the oldest patchset (build_change) and check gerrit to see if it's related.
-
It does seem to be related: https://review.openstack.org/#/c/177437/
-
Propose a revert, or fix forward. In this case, both:
- In the end the revert was abandoned and the fix merged.
Random notes:
- logstash keeps 10 days' worth of logs.
- Q: what gets indexed? A: this