[
https://issues.apache.org/jira/browse/HADOOP-15602?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16542640#comment-16542640
]
Vinayakumar B commented on HADOOP-15602:
----------------------------------------
Thank you [~daryn].
{quote}I'm not sure there's benefit to using a custom callq and scheduler etc
since those are predicated on weighting users. The user isn't known until the
auth completes so everything is prio 0 and a fair call queue is a no-op.
{quote}
Okay. I just wanted to re-use existing abstraction. May be in future, some
other special call queue impl could help. {{FairCallQueue}} will be used only
on configuration, even for RPC Queue. By default, {{LinkedBlockingQueue}} will
be used for both.
{quote}I'll try to look at the more throughly this week. Did you make sure that
it works correctly when requests are split over multiple sasl packets? Or that
clients are correctly disconnected after errors, etc?
{quote}
Do you mean multiple SASL messages from same connection? If yes, - This should
not happen as client waits for previous SASL message's response before sending
next sasl message.
If you are talking about {{RPC_CONTINUATION_PACKET}}, I could not find any
implementation yet about this, for normal RPC request as well. All RPCs and
SASL are sent as {{RPC_FINAL_PACKET}} in the header, right now.
{code:java}
/**
* <code>RPC_CONTINUATION_PACKET = 1;</code>
*
* <pre>
* not implemented yet
* </pre>
*/
RPC_CONTINUATION_PACKET(1, 1),
{code}
> Support SASL Rpc request handling in separate Handlers
> -------------------------------------------------------
>
> Key: HADOOP-15602
> URL: https://issues.apache.org/jira/browse/HADOOP-15602
> Project: Hadoop Common
> Issue Type: Improvement
> Components: ipc
> Reporter: Vinayakumar B
> Assignee: Vinayakumar B
> Priority: Major
> Attachments: HADOOP-15602.01.patch
>
>
> Right now, during RPC Connection establishment, all SASL requests are
> considered as OutOfBand requests and handled within the same Reader thread.
> SASL handling involves authentication with Kerberos and SecretManagers(for
> Token validation). During this time, Reader thread would be blocked, hence
> blocking all the incoming RPC requests on other established connections. Some
> secretManager impls require to communicate to external systems (ex: ZK) for
> verification.
> SASL RPC handling in separate dedicated handlers, would enable Reader threads
> to read RPC requests from established connections without blocking.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
