[fix][broker] Fix loadbalance score calculation problem

[gaozhangmin]

Motivation

There is a problem in my environment, a bundle cannot be loaded to an empty broker(no bundles on it).
I found that this empty broker still has preallocatedBundleData on it which caused the score is higher than others.

pulsar/pulsar-broker/src/main/java/org/apache/pulsar/broker/loadbalance/impl/LeastLongTermMessageRate.java

Lines 53 to 78 in c890a8c

	private static double getScore(final BrokerData brokerData, final ServiceConfiguration conf) {
	final double overloadThreshold = conf.getLoadBalancerBrokerOverloadedThresholdPercentage() / 100.0;
	final double maxUsage = brokerData.getLocalData().getMaxResourceUsage();
	if (maxUsage > overloadThreshold) {
	log.warn("Broker {} is overloaded: max usage={}", brokerData.getLocalData().getWebServiceUrl(), maxUsage);
	return Double.POSITIVE_INFINITY;
	}
	double totalMessageRate = 0;
	for (BundleData bundleData : brokerData.getPreallocatedBundleData().values()) {
	final TimeAverageMessageData longTermData = bundleData.getLongTermData();
	totalMessageRate += longTermData.getMsgRateIn() + longTermData.getMsgRateOut();
	}
	// calculate estimated score
	final TimeAverageBrokerData timeAverageData = brokerData.getTimeAverageData();
	final double timeAverageLongTermMessageRate = timeAverageData.getLongTermMsgRateIn()
	+ timeAverageData.getLongTermMsgRateOut();
	final double totalMessageRateEstimate = totalMessageRate + timeAverageLongTermMessageRate;
	if (log.isDebugEnabled()) {
	log.debug("Broker {} has long term message rate {}",
	brokerData.getLocalData().getWebServiceUrl(), totalMessageRateEstimate);
	}
	return totalMessageRateEstimate;
	}

This issue is caused by the preallocatedBundleData is not cleared from brokerData. LINE 573 and LINE 580 is the condition of clearing preallocatedBundleData.

There is a corner case, which bundle is unload before broker data reportted to Zookeeper. the preallocatedBundleData would never have the chance of clearing from loadData of leader.

pulsar/pulsar-broker/src/main/java/org/apache/pulsar/broker/loadbalance/impl/ModularLoadManagerImpl.java

Lines 570 to 590 in c890a8c

	final Map<String, BundleData> preallocatedBundleData = brokerData.getPreallocatedBundleData();
	synchronized (preallocatedBundleData) {
	for (String preallocatedBundleName : brokerData.getPreallocatedBundleData().keySet()) {
	if (brokerData.getLocalData().getBundles().contains(preallocatedBundleName)) {
	final Iterator<Map.Entry<String, BundleData>> preallocatedIterator =
	preallocatedBundleData.entrySet()
	.iterator();
	while (preallocatedIterator.hasNext()) {
	final String bundle = preallocatedIterator.next().getKey();
	if (bundleData.containsKey(bundle)) {
	preallocatedIterator.remove();
	preallocatedBundleToBroker.remove(bundle);
	}
	}
	}
	// This is needed too in case a broker which was assigned a bundle dies and comes back up.
	preallocatedBundleToBroker.remove(preallocatedBundleName);
	}
	}

Modifications

If current broker doesn't own the bundle, clearing the preallocatedBundleData

Verifying this change

• Make sure that the change passes the CI checks.

(Please pick either of the following options)

This change is a trivial rework / code cleanup without any test coverage.

(or)

This change is already covered by existing tests, such as (please describe tests).

(or)

This change added tests and can be verified as follows:

(example:)

• Added integration tests for end-to-end deployment with large payloads (10MB)
• Extended integration test for recovery after broker failure

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

If the box was checked, please highlight the changes

• Dependencies (add or upgrade a dependency)
• The public API
• The schema
• The default values of configurations
• The threading model
• The binary protocol
• The REST endpoints
• The admin CLI options
• The metrics
• Anything that affects deployment

Documentation

• doc
• doc-required
• doc-not-needed
• doc-complete

Matching PR in forked repository

gaozhangmin#10

[HQebupt]

LGTM

[lhotari]

If there was a bug, why didn't we try to add a unit test to reproduce the issue and prevent further regressions?