[
https://issues.apache.org/jira/browse/HADOOP-2910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12578073#action_12578073
]
Hairong Kuang commented on HADOOP-2910:
---------------------------------------
> Limit the total number of connections.
Most operation systems provide some parameters to limit the total number of
connections on the machine. It seems that it is unnecessary for the IPC server
to concern about it. So I drop this requirement in this jira.
Regarding to limiting the resource consumed by the incoming requests, we could
either
Solution 1. Limit call queue length. The ipc server stops reading from any
connection if the call queue is full.
Solution 2. Limit the amount of memory used in the call queue. This requires
the following changes:
* Call queue contains unmarshaled RPC requests;
* Requests get serialized after being taken out of the call queue and
before being served;
* Keep count of the total size of the partial and complete unmarshaled
requests;
* Stop reading from a connection if the incoming request size + total size
> max buffer size.
Solution 1 is simple. Solution 2 provides more accurate memory management and
parallel request serialization. Potentially we could do a buffer management to
eliminate the frequent buffer allocation & deallocation discussed in
HADOOP-2975. But solution 2 is more complicated and not favorable to large
messages like block reports.
Please comment! Thank you.
> Throttle IPC Client/Server during bursts of requests or server slowdown
> -----------------------------------------------------------------------
>
> Key: HADOOP-2910
> URL: https://issues.apache.org/jira/browse/HADOOP-2910
> Project: Hadoop Core
> Issue Type: Improvement
> Components: ipc
> Affects Versions: 0.16.0
> Reporter: Hairong Kuang
> Fix For: 0.17.0
>
>
> I propose the following to avoid an IPC server being swarmed by too many
> requests and connections
> 1. Limit call queue length or limit the amount of memory used in the call
> queue. This can be done by including the size of a request in the header and
> storing unmarshaled requests in the call queue.
> 2. If the call queue is full or queue buffer is full, stop reading requests
> from sockets. So requests stay at the server's system buffer or at the client
> side and thus eventually throttle the client.
> 3. Limit the total number of connections. Do not accept new connections if
> the connection limit is exceeded. (Note: this solution is unfair to new
> connections.)
> 4. If receive out of memory exception, close the current connection.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
