[jboss-jira] [JBoss JIRA] (JGRP-2172) Non-blocking flow control

[Bela Ban (JIRA)]

    [ https://issues.jboss.org/browse/JGRP-2172?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13422301#comment-13422301 ] 

Bela Ban commented on JGRP-2172:
--------------------------------

A minor change in semantics wrt non-blocking flow control versus blocking flow control: if sender P sends messages 1 (700K) and 2 (300K), then blocking flow control guarantees that P1 is sent first down the stack into UNICAST3 land where it gets a seqno, then P2. Thus, P1 will be delivered before P2.

However, these semantics are not guaranteed with non-blocking flow control: if we for example have 500K left, then P1 (700K) will be queued and the call returns so that P can send message 2. P2 (300K) will be sent successfully, but P1 will only get sent once we get new credits from the destination member. This means P2 may get delivered before P1.

Typically this is not a problem as P1 and P2 are probably unrelated, or else P would wait for a response to P1 before sending P2 (e.g. a GET request, then a GET response and - based on the response - a PUT request).

Just wanted to point this out.

> Non-blocking flow control
> -------------------------
>
>                 Key: JGRP-2172
>                 URL: https://issues.jboss.org/browse/JGRP-2172
>             Project: JGroups
>          Issue Type: Feature Request
>            Reporter: Bela Ban
>            Assignee: Bela Ban
>             Fix For: 4.0.4
>
>
> Sending a message through FlowControl (UFC, MFC) should not block if {{Message.Flag.NB_FC}} (non-blocking flow control) is set.
> Instead, the message should be added to a queue (bounded if {{max_size}} > 0, else unbounded). The max queue size is given in bytes, so we can estimate what the memory penalty for reaching that size would be (if bounded).
> The queued messages are sent when credits arrive. TBD: when credits arrive, should blocked threads or queued messages be released first?
> Non-blocking flow control can be used by both external and internal threads.
> If the queue is unbounded, then it is the responsibility of the application (e.g. Infinispan) to make sure the queue doesn't grow to an untenable size.



--
This message was sent by Atlassian JIRA
(v7.2.3#72005)