[
https://issues.apache.org/jira/browse/HBASE-18869?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16197233#comment-16197233
]
Chenxi Tong commented on HBASE-18869:
-------------------------------------
Hi, [[email protected]] thanks for reply!
The ReadRequestsCount not satisfy my requirement, it records scan results in
RSRpcServices, it bigger much than readRequest of requestCount in
RSRpcServices, so i want to use rpcReadRequestsCount to accumulate the
rpcGetRequestCount and rpcScanRequestCount to compute read request per second.
{code:java}
while (i < maxResults) {
// Reset the batch progress to 0 before every call to
RegionScanner#nextRaw. The
// batch limit is a limit on the number of cells per Result. Thus, if
progress is
// being tracked (i.e. scannerContext.keepProgress() is true) then we
need to
// reset the batch progress between nextRaw invocations since we
don't want the
// batch progress from previous calls to affect future calls
scannerContext.setBatchProgress(0);
// Collect values to be returned here
moreRows = scanner.nextRaw(values, scannerContext);
if (!values.isEmpty()) {
Result r = Result.create(values, null, stale,
scannerContext.hasMoreCellsInRow());
lastBlock.setValue(addSize(context, r, lastBlock.getValue()));
results.add(r);
i++;
}
boolean sizeLimitReached =
scannerContext.checkSizeLimit(LimitScope.BETWEEN_ROWS);
boolean timeLimitReached =
scannerContext.checkTimeLimit(LimitScope.BETWEEN_ROWS);
boolean rowLimitReached = i >= maxResults;
limitReached = sizeLimitReached || timeLimitReached ||
rowLimitReached;
if (limitReached || !moreRows) {
if (LOG.isTraceEnabled()) {
LOG.trace("Done scanning. limitReached: " + limitReached + "
moreRows: " + moreRows
+ " scannerContext: " + scannerContext);
}
// We only want to mark a ScanResponse as a heartbeat message in
the event that
// there are more values to be read server side. If there aren't
more values,
// marking it as a heartbeat is wasteful because the client will
need to issue
// another ScanRequest only to realize that they already have all
the values
if (moreRows) {
// Heartbeat messages occur when the time limit has been reached.
builder.setHeartbeatMessage(timeLimitReached);
}
break;
}
values.clear();
}
if (limitReached || moreRows) {
// We stopped prematurely
builder.setMoreResultsInRegion(true);
} else {
// We didn't get a single batch
builder.setMoreResultsInRegion(false);
}
// Check to see if the client requested that we track metrics server
side. If the
// client requested metrics, retrieve the metrics from the scanner
context.
if (trackMetrics) {
Map<String, Long> metrics =
scannerContext.getMetrics().getMetricsMap();
ScanMetrics.Builder metricBuilder = ScanMetrics.newBuilder();
NameInt64Pair.Builder pairBuilder = NameInt64Pair.newBuilder();
for (Entry<String, Long> entry : metrics.entrySet()) {
pairBuilder.setName(entry.getKey());
pairBuilder.setValue(entry.getValue());
metricBuilder.addMetrics(pairBuilder.build());
}
builder.setScanMetrics(metricBuilder.build());
}
}
+region.updateReadRequestsCount(i);+
{code}
this patch is for branch-1.3.0, if you think the idea is feasible, i will
attach the branch for master
thanks~~
> Table rpc metrics
> -----------------
>
> Key: HBASE-18869
> URL: https://issues.apache.org/jira/browse/HBASE-18869
> Project: HBase
> Issue Type: Wish
> Components: metrics
> Affects Versions: 1.3.0
> Reporter: Chenxi Tong
> Priority: Minor
> Labels: metrics
> Attachments: HBASE-18869.patch
>
>
> Hi all
> In the HBASE-15518, uses the MetricsTableWrapperAggregateImpl to aggregate
> each region`s
> totalRequestsCount,readRequestsCount,writeRequestsCount,memstoresSize
> storeFilesSize, tableSize metrics and exports them to jmx, so i want to use
> the totalRequestsCount`rate as table`rpc request, but found it is bigger much
> more than ServerLoad`s requestsPerSecond, i am not sure if it is a correct
> way to do this, how we can collect each table`s rpc request?
> Best wishes!
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
