mirror of
https://github.com/ceph/ceph
synced 2025-01-11 13:41:02 +00:00
Ceph is a distributed object, block, and file storage platform
6295578f16
The peer task does wait_for_clean, and then lost_unfound immediately marks something down. But the PGs become clean before the replica last_epoch_started is moved forward in time, which means they block waiting for the now down OSD. Needlessly. Until we fix this, just do the peer test after. |
||
|---|---|---|
| clusters | ||
| suites | ||
| .gitignore | ||
| README | ||
ceph-qa-suite ------------- clusters/ - some predefined cluster layouts suites/ - sets of collections We have several collections, each a subdirectory within suites/*/. Each collection directory consists of a set facets. The basic idea is that, for each collection, the set of tests to run consists of all combinations of taking one yaml fragment from each facet. For example, given the files suite/collection/ suite/collection/foo/ suite/collection/foo/clusters/ suite/collection/foo/clusters/fixed-3.yaml suite/collection/foo/clusters/fixed-9.yaml suite/collection/foo/tasks/ suite/collection/foo/tasks/a suite/collection/foo/tasks/b suite/collection/foo/tasks/c suite/collection/bar/ suite/collection/bar/clusters/fixed-3.yaml suite/collection/bar/tasks/ suite/collection/bar/tasks/d suite/collection/bar/tasks/e teuthology-suite would run tasks a, b, and c on both fixed-3 and fixed-9 clusters. It would also run tasks d and e on the fixed-3 cluster. Note that the 'clusters' and 'tasks' terminology is actually meaningless here. All teuthology-suite does is stick the yaml fragments together (one from each facet) and run teuthology on the result (optionally combined with any additional fragments passed on the command line). In practice, we can keep common/shared task and cluster definitions in the top-level clusters/ (which are otherwise ignored), and symlink to them from the collections that want to use them.