Work cancellation and timeout handling #113
Conversation
|
@kennethlakin #105 was merged, but this one still needs rebasing |
kennethlakin
force-pushed
the
work-cancellation
branch
from
5605424 to
bfaf9df
Compare
|
Rebased. I don't know why Elvis is complaining here. |
|
Actually, hang on. There are some whitespace changes that I accidentally un-did. Gimmy a few minutes to fix that. |
kennethlakin
force-pushed
the
work-cancellation
branch
from
bfaf9df to
97e7d66
Compare
|
Okay. Whitespace issues are fixed. Feel free to merge. |
| @@ -393,12 +420,14 @@ at_rest(timeout, State) -> | |||
| ok = gen_fsm:send_event(self(), Work), | |||
| {next_state, at_rest, NewState} | |||
| end; | |||
| at_rest({get_async, {HandleEvent, AsyncMode}, Args, From}, | |||
| at_rest({get_async, {HandleEvent, AsyncMode}, | |||
| {ReqRef, _, _, _} = Args, From}, | |||
There was a problem hiding this comment.
Please check this line's indentation.
There was a problem hiding this comment.
Okay. That line's indentation now matches the gun_response clauses of wait_response/2.
kennethlakin
force-pushed
the
work-cancellation
branch
from
97e7d66 to
a1546d2
Compare
|
Okay. The line in question's indentation now matches the gun_response clauses of wait_response/2. |
kennethlakin
force-pushed
the
work-cancellation
branch
from
a1546d2 to
349e639
Compare
This addresses Issue inaka#109, and then some. When a request is made, shotgun:request makes a fresh reference and passes that along with the work to the shotgun FSM. When a request times out, shotgun:request now sends an async request to cancel the now-useless unit of work, passing along the reference for identification, before notifying the client of the request timeout. Things to know: * Gun has its own request queue. So, if a request has made it to gun, cancellation of the request will only squelch future messages that would have come from that request. It is -however- safe to bounce back to at_rest (as we now do), as messages from the cancelled request will not be sent back to the FSM. * Removal of queued requests makes use of queue:filter. For large queues, this is kinda not fast. Minor changes: * state typedef moved up with the other typedefs. * enqueue_work_or_stop slightly simplified.
kennethlakin
force-pushed
the
work-cancellation
branch
from
349e639 to
57d60e8
Compare
|
It occurred to me that I had inadvertently changed the return value when |
|
Okay, I played around with this some more this weekend and discovered a serious bug. I'll close the PR for now and reissue it when I've pushed the fix. |
This addresses Issue #109 . See the commit comments for more details.
Do note that it's based off of my check_uri-handle-iodata branch, mentioned in PR #105 . (This is probably the source of the branch conflict notification below.) I can rebase it if you like, but the iodata handling change is small and nice to have. :)