[
https://issues.apache.org/jira/browse/HBASE-19924?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16401374#comment-16401374
]
Chia-Ping Tsai commented on HBASE-19924:
----------------------------------------
*Q1*
Given that we need to check both of size and count, it should be ok to get rid
of following methods
{code:java}
/**
* Checks if it is possible to execute the specified operation.
*
* @param estimateWriteSize the write size that will be checked against the
available quota
* @param estimateReadSize the read size that will be checked against the
available quota
* @throws ThrottlingException thrown if not enough avialable resources to
perform operation.
*/
void checkQuota(long estimateWriteSize, long estimateReadSize)
throws ThrottlingException;
/**
* Removes the specified write and read amount from the quota.
* At this point the write and read amount will be an estimate,
* that will be later adjusted with a consumeWrite()/consumeRead() call.
*
* @param writeSize the write size that will be removed from the current quota
* @param readSize the read size that will be removed from the current quota
*/
void grabQuota(long writeSize, long readSize);
{code}
*Q2*
avialable _> available
*Q3 (unrelated to this jira)*
Why we consume the write/read req by heap size? Should we make both methods to
accept both size and count.
{code:TimeBasedLimiter.java}
@Override
public void consumeWrite(final long size) {
reqSizeLimiter.consume(size);
writeSizeLimiter.consume(size);
}
@Override
public void consumeRead(final long size) {
reqSizeLimiter.consume(size);
readSizeLimiter.consume(size);
}
{code}
*Q4 (unrelated to this jira)*
Is it necessary to add the condition check (xxxSize > 0)? Is it impossible to
have a zero size op? For example, the Result generated by a check-exist Get may
have zero heap size since the heap size of Result is the sum of heap size of
cells.
{code}
if (estimateWriteSize > 0) {
if (!writeReqsLimiter.canExecute(writeReqs)) {
ThrottlingException.throwNumWriteRequestsExceeded(writeReqsLimiter.waitInterval());
}
if (!writeSizeLimiter.canExecute(estimateWriteSize)) {
ThrottlingException.throwWriteSizeExceeded(writeSizeLimiter.waitInterval(estimateWriteSize));
}
}
if (estimateReadSize > 0) {
{code}
> hbase rpc throttling does not work for multi() with request count rater.
> ------------------------------------------------------------------------
>
> Key: HBASE-19924
> URL: https://issues.apache.org/jira/browse/HBASE-19924
> Project: HBase
> Issue Type: Bug
> Components: rpc
> Affects Versions: 0.16.0, 1.2.6
> Reporter: huaxiang sun
> Assignee: huaxiang sun
> Priority: Major
> Attachments: HBASE-19924-master-v001.patch
>
>
> Basically, rpc throttling does not work for request count based rater for
> multi. for the following code, when it calls limiter's checkQuota(),
> numWrites/numReads is lost.
> {code:java}
> @Override
> public void checkQuota(int numWrites, int numReads, int numScans) throws
> ThrottlingException {
> writeConsumed = estimateConsume(OperationType.MUTATE, numWrites, 100);
> readConsumed = estimateConsume(OperationType.GET, numReads, 100);
> readConsumed += estimateConsume(OperationType.SCAN, numScans, 1000);
> writeAvailable = Long.MAX_VALUE;
> readAvailable = Long.MAX_VALUE;
> for (final QuotaLimiter limiter : limiters) {
> if (limiter.isBypass()) continue;
> limiter.checkQuota(writeConsumed, readConsumed);
> readAvailable = Math.min(readAvailable, limiter.getReadAvailable());
> writeAvailable = Math.min(writeAvailable, limiter.getWriteAvailable());
> }
> for (final QuotaLimiter limiter : limiters) {
> limiter.grabQuota(writeConsumed, readConsumed);
> }
> }{code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
