wip-subset: yaml assembly is putting tasks out-of-order
Here's one example; I am seeing many: http://qa-proxy.ceph.com/teuthology/gregf-2015-04-12_16:25:14-fs-greg-fs-testing-1-testing-basic-multi/844780/config.yaml
tasks: - internal.check_packages: null - internal.lock_machines: - 3 - plana,mira,burnupi - internal.save_config: null - internal.check_lock: null - internal.connect: null - internal.push_inventory: null - internal.serialize_remote_roles: null - internal.check_conflict: null - internal.check_ceph_data: null - internal.vm_setup: null - kernel: *id001 - internal.base: null - internal.archive: null - internal.coredump: null - internal.sudo: null - internal.syslog: null - internal.timer: null - chef: null - clock.check: null - mds_thrash: null - install: null - ceph: null - ceph-fuse: null - workunit: clients: all: - suites/fsstress.sh
Notice that the mds_thrash task is placed ahead of "install". Obviously things go terribly wrong.
As far as I can tell, nothing in the yaml fragments which this job is generated from have changed in more than a year, which leads me to believe that the teuthology code for assembling these must have changed. I'm not sure if it's consistently certain jobs being mis-ordered or if something else is going on.
#1 Updated by Andrew Schoen over 4 years ago
It doesn't look like this bit of teuthology code has changed anytime recently. It seems to construct the yaml fragments based on the alpha-order of the directory name. Maybe renaming ceph-thrash to just 'thrash' will get you the yaml you need?
Here are the fragments it's using for the linked run (in alpha-order).
['ceph-thrash/default.yaml', 'ceph/base.yaml', 'clusters/mds-1active-1standby.yaml', 'debug/mds_client.yaml', 'fs/btrfs.yaml', 'msgr-failures/none.yaml', 'overrides/whitelist_wrongly_marked_down.yaml', 'tasks/cfuse_workunit_suites_fsstress.yaml']
#2 Updated by Greg Farnum over 4 years ago
- Subject changed from yaml assembly is putting tasks out-of-order to sip-subset: yaml assembly is putting tasks out-of-order
- Assignee changed from Greg Farnum to Samuel Just
This turns out to be only a problem in wip-subset; you can see it by scheduling a dry run of the fs/thrash suite. It orders "ceph-thrash/default.yaml" ahead of "ceph/base.yaml".
That's kind of a blocker for merge...