[
https://issues.apache.org/jira/browse/HDFS-13873?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16717580#comment-16717580
]
Erik Krogen commented on HDFS-13873:
------------------------------------
I like your logic, [~shv], but I think it could still result in the live-lock
issue [~daryn] was originally concerned with if you have a very read-heavy
cluster with very few writes, if you have either a malicious client or a lag in
edit log tailing. Let's assume there is only 100 write/s and
{{client.LastSeenId - observer.LastWrittenId == 10000}} (because the client is
malicious or because the observer fell far behind), the server will not reject
the request, but the observer will take a long time before it serves the
request, and if there are multiple such request it could fill all of the
handlers. I think ideally we also need to add some sort of threshold based on
time the request has waited or number of times the request has been pushed back
into the queue.
> ObserverNode should reject read requests when it is too far behind.
> -------------------------------------------------------------------
>
> Key: HDFS-13873
> URL: https://issues.apache.org/jira/browse/HDFS-13873
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: hdfs-client, namenode
> Affects Versions: HDFS-12943
> Reporter: Konstantin Shvachko
> Assignee: Chao Sun
> Priority: Major
>
> Add a server-side threshold for ObserverNode to reject read requests when it
> is too far behind.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
