osd: clean up process_peering_events

[363921219]

Signed-off-by: Jie Wang jie.wang@kylin-cloud.com

[ghost]

jenkins test this please (eio/scrub failure)

[liewegas]

1. Are you sure the assert is valid?
2. Can we drop the curmap assignement inside the loop instead? Or, move curmap scope into the loop?

[ghost]

jenkins test this please (flake8 related error, now fixed)

[363921219]

@liewegas

Are you sure the assert is valid?

sure, "process_peering_events" will not be called if pgs is empty.

Can we drop the curmap assignement inside the loop instead?

osdmap may be updated when 'process_peering_events' run, so i am not sure it is valid.

Or, move curmap scope into the loop?

curmap is need outside the loop

[tchaikov]

i concur with @363921219, ThreadPool won't pass an empty list to BatchWorkQueue if _void_dequeue() returns NULL.

and dispatch_context(rctx, 0, curmap, &handle) out side of the loop uses the last curmap passed to the last call advance_pg(). so we can compare the curmap.epoch with the nextmap.epoch before sending the messages out.

[tchaikov]

we can move this assert up into PeeringWQ::_process()`.

[tchaikov]

and also put more details into the commit message: clean up is too vague a word, IMHO.

[363921219]

@tchaikov thanks for your comments

[tchaikov]

could you remove "Reviewed-by: Kefu Chai kchai@redhat.com" from your commit messages? it will be added to the merge commit by the one who merges your changes.

[tchaikov]

and please squash these two commit into a single one.

[ghost]

test this please (asok error now fixed in master)

[tchaikov]

lgtm.

[ghost]

jenkins test this please (eio)

[tchaikov]

http://pulpito.ceph.com/kchai-2016-11-22_09:44:32-rados-wip-kefu-testing---basic-mira/