[hotfix][metrics] Fix NullArgumentException when calculating checkpoint statistics with empty data

[qiref]

What is the purpose of the change

When a Flink job starts without any checkpoint data, the DescriptiveStatisticsHistogramStatistics
may be initialized with an empty array. The maybeInitPercentile() method only checked for null but not for empty arrays, causing Apache Commons Math's Percentile.evaluate() to throw a NullArgumentException when accessing the checkpoint statistics page in the Web UI.

org.apache.commons.math3.exception.NullArgumentException: input array
	at org.apache.commons.math3.util.MathArrays.verifyValues(MathArrays.java:1753) ~[commons-math3-3.6.1.jar:3.6.1]
	at org.apache.commons.math3.stat.descriptive.AbstractUnivariateStatistic.test(AbstractUnivariateStatistic.java:158) ~[commons-math3-3.6.1.jar:3.6.1]
	at org.apache.commons.math3.stat.descriptive.rank.Percentile.evaluate(Percentile.java:272) ~[commons-math3-3.6.1.jar:3.6.1]
	at org.apache.commons.math3.stat.descriptive.rank.Percentile.evaluate(Percentile.java:241) ~[commons-math3-3.6.1.jar:3.6.1]
	at org.apache.flink.runtime.metrics.DescriptiveStatisticsHistogramStatistics$CommonMetricsSnapshot.getPercentile(DescriptiveStatisticsHistogramStatistics.java:163) ~[classes/:?]
	at org.apache.flink.runtime.metrics.DescriptiveStatisticsHistogramStatistics.getQuantile(DescriptiveStatisticsHistogramStatistics.java:57) ~[classes/:?]
	at org.apache.flink.runtime.checkpoint.StatsSummarySnapshot.getQuantile(StatsSummarySnapshot.java:108) ~[classes/:?]
	at org.apache.flink.runtime.rest.messages.checkpoints.StatsSummaryDto.valueOf(StatsSummaryDto.java:81) ~[classes/:?]
	at org.apache.flink.runtime.rest.handler.job.checkpoints.CheckpointingStatisticsHandler.createCheckpointingStatistics(CheckpointingStatisticsHandler.java:128) ~[classes/:?]
	at org.apache.flink.runtime.rest.handler.job.checkpoints.CheckpointingStatisticsHandler.handleCheckpointStatsRequest(CheckpointingStatisticsHandler.java:85) ~[classes/:?]
	at org.apache.flink.runtime.rest.handler.job.checkpoints.CheckpointingStatisticsHandler.handleCheckpointStatsRequest(CheckpointingStatisticsHandler.java:59) ~[classes/:?]
	at org.apache.flink.runtime.rest.handler.job.checkpoints.AbstractCheckpointStatsHandler.lambda$handleRequest$1(AbstractCheckpointStatsHandler.java:89) ~[classes/:?]
	at java.util.concurrent.CompletableFuture.uniApply(CompletableFuture.java:616) [?:1.8.0_432]
	at java.util.concurrent.CompletableFuture$UniApply.tryFire(CompletableFuture.java:591) [?:1.8.0_432]
	at java.util.concurrent.CompletableFuture$Completion.run(CompletableFuture.java:456) [?:1.8.0_432]
	at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511) [?:1.8.0_432]
	at java.util.concurrent.FutureTask.run(FutureTask.java:266) [?:1.8.0_432]
	at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$201(ScheduledThreadPoolExecutor.java:180) [?:1.8.0_432]
	at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:293) [?:1.8.0_432]
	at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149) [?:1.8.0_432]
	at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624) [?:1.8.0_432]
	at java.lang.Thread.run(Thread.java:750) [?:1.8.0_432]

Brief change log

This commit adds an additional check for empty arrays in the maybeInitPercentile() method. When the data array is null or empty, it now uses a default array {0.0} to avoid the exception while maintaining reasonable default values for all percentile calculations.

Verifying this change

This change is already covered by existing tests, such as:

• DescriptiveStatisticsHistogramTest - Tests the histogram statistics calculation with various data scenarios
• The existing tests continue to pass with this change, and the fix prevents the NullArgumentException in production scenarios

Manual Verification:

• Started a Flink cluster with a streaming job that has no checkpoint configured
• Accessed the checkpoint statistics page in Web UI (http://localhost:8081/#/job/<job-id>/checkpoints)
• Verified that the page loads successfully without throwing NullArgumentException
• After enabling checkpoints and completing some checkpoints, verified that real statistics are calculated correctly

Does this pull request potentially affect one of the following parts:

• Dependencies (does it add or upgrade a dependency): (no)
• The public API, i.e., is any changed class annotated with @Public(Evolving): (no)
• The serializers: (no)
• The runtime per-record code paths (performance sensitive): (no)
• Anything that affects deployment or recovery: JobManager (and its components), Checkpointing, Kubernetes/Yarn, ZooKeeper: (no)
• The S3 file system connector: (no)

Documentation

• Does this pull request introduce a new feature? (no)

[flinkbot]

CI report:

• da263a7 Azure: SUCCESS

Bot commands

The @flinkbot bot supports the following commands:

• @flinkbot run azure re-run the last Azure build

[davidradl]

Looks good. Could we have a unit test for this please? Maybe at the level that hit the NullArgumentException.

[qiref]

Done! Added tests that specifically target the NullArgumentException scenario. The bug is now immortalized in our test suite. Tests can be verified with:

mvn test -pl flink-runtime -Dtest=DescriptiveStatisticsHistogramTest