[
https://issues.apache.org/jira/browse/IGNITE-25422?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Vyacheslav Koptilin updated IGNITE-25422:
-----------------------------------------
Description:
*Motivation*
Currently, there are just these metrics:
{code:java}
public DisruptorMetrics disruptorMetrics(String name) {
return new DisruptorMetrics(
(DistributionMetric) metrics.get(name + ".Batch"),
(DistributionMetric) metrics.get(name + ".Stripes")
);
} {code}
which means batch size and stripe hits.
It looks reasonable to add the following new metrics:
- stripe queue length
- task wait time in the queue (histogram)
was:
*Motivation*
Currently there are just these metrics:
{code:java}
public DisruptorMetrics disruptorMetrics(String name) {
return new DisruptorMetrics(
(DistributionMetric) metrics.get(name + ".Batch"),
(DistributionMetric) metrics.get(name + ".Stripes")
);
} {code}
which means batch size and stripe hits.
We need to add the count of pending tasks (there is a separate queue (ring
buffer) for each stripe):
{code:java}
long cursor = rb.getCursor();
long minConsumer = rb.getMinimumGatingSequence();
long pending = cursor - minConsumer; {code}
This will allow us to make conclusions about the possible delays for task
processing.
Also, the time of processing might be useful.
*Definition of done*
Additional disruptor metrics are created:
- pending tasks (for each stripe, maximum, minimum, median)
- time of processing
> Add metrics for raft disruptors
> -------------------------------
>
> Key: IGNITE-25422
> URL: https://issues.apache.org/jira/browse/IGNITE-25422
> Project: Ignite
> Issue Type: Improvement
> Reporter: Denis Chudov
> Priority: Major
> Labels: ignite-3
>
> *Motivation*
> Currently, there are just these metrics:
> {code:java}
> public DisruptorMetrics disruptorMetrics(String name) {
> return new DisruptorMetrics(
> (DistributionMetric) metrics.get(name + ".Batch"),
> (DistributionMetric) metrics.get(name + ".Stripes")
> );
> } {code}
> which means batch size and stripe hits.
>
> It looks reasonable to add the following new metrics:
> - stripe queue length
> - task wait time in the queue (histogram)
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
