qpid-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Alan Conway (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (PROTON-1440) pn_connection_wake to return bool status
Date Thu, 16 Mar 2017 20:49:41 GMT

    [ https://issues.apache.org/jira/browse/PROTON-1440?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15928876#comment-15928876
] 

Alan Conway commented on PROTON-1440:
-------------------------------------

Review request: https://github.com/alanconway/qpid-proton/tree/safe-wake https://reviews.apache.org/r/57702/

I have not changed the API - no return from wake. We can't (easily) guaratnee that the return
value is correct, since the user has control of the event stream between _wait() and _done(),
so calls to wake() concurrent with that are somewhat out of our control. So better to say
nothing than claim there will be a WAKE event when there may not be. Wake is fully thread
safe however, it can be called even after the proactor has been freed, but is a no-op.

The real issue is co-ordination with state outside the proactor, which can't be done from
inside. Bindings like C++ can still provide more user-friendly guarantees, e.g. the c++ inject()
 or dispatch qd_deferred_call guarantees  can still be implemented but need sync outside the
proactor (for C++ it's in the event_loop).

> pn_connection_wake to return bool status
> ----------------------------------------
>
>                 Key: PROTON-1440
>                 URL: https://issues.apache.org/jira/browse/PROTON-1440
>             Project: Qpid Proton
>          Issue Type: Improvement
>          Components: proton-c
>    Affects Versions: 0.17.0
>            Reporter: Alan Conway
>            Assignee: Alan Conway
>             Fix For: 0.18.0
>
>
> pn_connection_wake() allows any thread to send a PN_CONNECTION_WAKE event to a proactor-managed
connection. Currently it does not return any value, and the application must ensure it is
not called after the PN_TRANSPORT_CLOSED event is processed for that connection.
> This is tricky for the application to synchronize and inconsistent with other pn_connection_
functions. If refcounts are used to keep a pn_connection_t in memory after the TRANSPORT_CLOSED,
then calling wake() is an error even though the pn_connection_t is valid.
> Change wake to be safe as long as the pn_connection_t is valid. It returns true if the
connection will be woken, false if it is too late.



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@qpid.apache.org
For additional commands, e-mail: dev-help@qpid.apache.org


Mime
View raw message