[
https://issues.apache.org/jira/browse/HBASE-12790?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14388771#comment-14388771
]
ramkrishna.s.vasudevan commented on HBASE-12790:
------------------------------------------------
Thanks Stack for the review.
bq.a new concept – 'grouping' – is added to Scan as public apis with no doc on
what it is about.
I added the comment (not as doc) on the groupingId String variable. Will add a
doc on the setters and getters. bq.Why is a bq.grouping id a 'string'?
What do you think we could add, an Integer? I thought a groupingId could a be a
simple random string formed from a Unique random Id generated and use it with a
clientId - Atleast currently in Phoenix it is a String.
bq.Does FairShareBalancedRPCExecutor add anything? Why in class comment does it
talk about scan when it plainly does nothing with scan. Ditto
FairShareRWQueueRPCExecutor.
Yes. It actually instantiates the PriorityQueue that will be used. In the
BalancedRPCExecutor it will be LinkedBlockingQueue or BoundedPriorityQueue.
Here we instantiate the FairShareBlockingQueue or the
FairSharePriorityBasedBlockingQueue.
What I mean in the comment is that currently as I had tried out only for scan
wanted to highlight that this feature works with SCan only.
bq.and other grouping pollution. Is this because you wanted to avoid putting in
place a fair scheduler implementation?
May be yes. That would be cleaner. I did this way because wanted to use this
new queue only in the read part.
As you suggested if you go with a new scheduler then should we use this for
Replication, write and read also?
bq.Any micro-benchmark compare of this implementation to current q
implementation even with no grouping enabled?
Yes sure.
bq.Is a 'producer' a 'group'? (Odd having method named extractProducer return
group).
Yes here producer is a group. The AbstractPriorityRoundRobinQueue is a generic
impl. So here in this case we have the producer as a simple grouping String.
bq.Do you have any proof this code delivers what is suggested at the top of
this issue, that if all cilent A's scans are queued before all of client B's,
that client B will get some action.
Yes I will add a real world test case. Mainly this patch is to highlight the
idea.
My main question would be should we use this scheduler in all cases like writes
and replication? OR may be as a first cut as in this patch we go only 'scans'
or 'gets'?
Even in the current way we have different queues for writes and reads.
bq.So, users would have to 'enable' this on the cluster? It would not be on by
default? If no degradation in scheduler, why would we not want this always on
Sure we could try this, based on the above question.
Once again thanks for the review.
> Support fairness across parallelized scans
> ------------------------------------------
>
> Key: HBASE-12790
> URL: https://issues.apache.org/jira/browse/HBASE-12790
> Project: HBase
> Issue Type: New Feature
> Reporter: James Taylor
> Assignee: ramkrishna.s.vasudevan
> Labels: Phoenix
> Attachments: AbstractRoundRobinQueue.java, HBASE-12790.patch,
> HBASE-12790_1.patch
>
>
> Some HBase clients parallelize the execution of a scan to reduce latency in
> getting back results. This can lead to starvation with a loaded cluster and
> interleaved scans, since the RPC queue will be ordered and processed on a
> FIFO basis. For example, if there are two clients, A & B that submit largish
> scans at the same time. Say each scan is broken down into 100 scans by the
> client (broken down into equal depth chunks along the row key), and the 100
> scans of client A are queued first, followed immediately by the 100 scans of
> client B. In this case, client B will be starved out of getting any results
> back until the scans for client A complete.
> One solution to this is to use the attached AbstractRoundRobinQueue instead
> of the standard FIFO queue. The queue to be used could be (maybe it already
> is) configurable based on a new config parameter. Using this queue would
> require the client to have the same identifier for all of the 100 parallel
> scans that represent a single logical scan from the clients point of view.
> With this information, the round robin queue would pick off a task from the
> queue in a round robin fashion (instead of a strictly FIFO manner) to prevent
> starvation over interleaved parallelized scans.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
