GridCacheIoManager.processFailedMessage() method

[Valentin Kulichenko]

Igniters,

Can someone please explain the purpose
of GridCacheIoManager.processFailedMessage() method? I've already seen
several reports on user@ about "Unsupported direct type" exception thrown
from this method and each time we just add processing for one more message.

In my opinion this approach is not generic and therefore error-prone. Also
this exception discards original 'cacheMsg.classError()' error and is never
sent to the client, just being logged on the server node.

Is there a more generic solution for this? At the very least I would change
the default behavior to send class error to the client so that user is
properly notified.

Thoughts?

-Val

[dsetrakyan]

On Wed, Dec 2, 2015 at 9:31 PM, Valentin Kulichenko <
[hidden email]> wrote:

> Igniters,
>
> Can someone please explain the purpose
> of GridCacheIoManager.processFailedMessage() method? I've already seen
> several reports on user@ about "Unsupported direct type" exception thrown
> from this method and each time we just add processing for one more message.
>
> In my opinion this approach is not generic and therefore error-prone. Also
> this exception discards original 'cacheMsg.classError()' error and is never
> sent to the client, just being logged on the server node.
>
> Is there a more generic solution for this? At the very least I would change
> the default behavior to send class error to the client so that user is
> properly notified.
>

Agree with this. Simply hanging a cluster is hardly intuitive, even in case
of a user or Ignite bug.


> Thoughts?
>
> -Val
>

[yzhdanov]

This method was introduced exactly to prevent cluster from hanging in case
of unexpected problems during message processing. The thing is that the one
who has added new message has not added corresponding block to mentioned
method.

Agree with the point about original error message should not be suppressed
in any case.

Anton V, can you please file a ticket and fix the comments?

--Yakov

[Anton Vinogradov]

https://issues.apache.org/jira/browse/IGNITE-2079

On Thu, Dec 3, 2015 at 11:26 AM, Yakov Zhdanov <[hidden email]> wrote:

> This method was introduced exactly to prevent cluster from hanging in case
> of unexpected problems during message processing. The thing is that the one
> who has added new message has not added corresponding block to mentioned
> method.
>
> Agree with the point about original error message should not be suppressed
> in any case.
>
> Anton V, can you please file a ticket and fix the comments?
>
> --Yakov
>