Jobs with worker class qemu_x86_64 are taken by machines without this class, causing incomplete jobs
Jobs with worker class qemu_x86_64 (settings and var.json) are taken by machines without this class, causing incomplete jobs.
For the jobs taken by overdrive2, the worker class on settings and var.json is different (qemu_x86_64 in settings and qemu_aarch64_maintenance in var.json)
H1 When a big number of jobs are created by cloning a job (e.g. 100), 5% of these jobs are taken by a worker without a matching worker class.
H2 workers.ini is not configured properly. REJECTED BY E2-1 and E2-2
H3 The different worker classes on settings and var.json for the jobs taken by overdrive2 happened because the workers support multi-webui.
E1-1 Execute the workers on overdrive2 and QA-Power8-5-kvm with verbose mode and clone a job 100 times.
R1-1 Not done yet
E2-1 Check that the worker classes are properly configured in workers.ini on overdrive2.
R2-1 overdrive2 uses: qemu_aarch64_maintenance
E2-2 Check that the worker classes are properly configured on workers.ini on QA-Power8-5-kvm.
R2-2 QA-Power8-5-kvm uses: qemu_ppc64le,qemu_ppc64le_no_tmpfs
E3-2 Check that the worker.ini configuration for QA-Power8-5-kvm is set for one webui.
R3-2 One webui configured: http://openqa.suse.de
#1 Updated by coolo over 4 years ago
What exactly did you do? WORKER_CLASS is different in vars.json than in settings: https://openqa.suse.de/tests/889178/file/vars.json
#2 Updated by coolo over 4 years ago
Somehow multiple webuis are related:
Apr 20 14:14:31 overdrive2 worker: [ERROR] 502 response: Proxy Error (remaining tries: 2) Apr 20 14:32:03 overdrive2 worker: [INFO] registering worker with openQA http://lord.arch.suse.de... Apr 20 14:32:03 overdrive2 worker: [INFO] got job 889178: 00889178-sle-12-SP3-Server-DVD-x86_64-Build0340-textmode_statistics_poo_18634@64bit Apr 20 14:32:03 overdrive2 worker: [INFO] 2015: WORKING 889178 Apr 20 14:32:11 overdrive2 worker: child 2015 died with exit status 256
#9 Updated by szarate about 4 years ago
- Assignee set to szarate
This this pr should take care of this. Problem is related to how the jobs are being added to the database, moving everything into a transaction as coolo suggested, solves the problem (apparently).
Time to hunt for poo#20002 as i belive that the condition there is a bit different.