Performance[BMQ, MQB, MWC]: performance tune-up [1]

[dorjesinpo]

First half of 181

1. Increase Event (batch) size. Keep PUT soft limit
2. Use bsls::BlockGrowth::BSLS_CONSTANT (instead of geometric) for pools
3. Avoid mwcu::MemOutStream while validating in Cluster
4. StorageManager::processReceiptEvent in IO thread (minus one thread switch)
5. FileStore generates Receipt at the end of received batch processing

[678098]

Just an idea:

• get rid of a loop
• get rid of tmp bool doRetry
• add performance hint
• as a bonus, in a debugger you will see that you are currently on the first try or on the second

Suggested change

	bool doRetry = false;
	do {
	buildRc = d_storageEventBuilder.packMessage(type,
	d_config.partitionId(),
	0, // flags
	journalOffsetWords,
	journalRecordBlobBuffer);
	if (buildRc == bmqt::EventBuilderResult::e_EVENT_TOO_BIG && !doRetry) {
	flushIfNeeded(true);
	doRetry = true;
	}
	else {
	doRetry = false;
	}
	} while (doRetry);
	buildRc = d_storageEventBuilder.packMessage(type,
	d_config.partitionId(),
	0, // flags
	journalOffsetWords,
	journalRecordBlobBuffer);
	if (BSLS_PERFORMANCEHINT_PREDICT_UNLIKELY(buildRc == bmqt::EventBuilderResult::e_EVENT_TOO_BIG)) {
	BSLS_PERFORMANCEHINT_UNLIKELY_HINT;
	flushIfNeeded(true);
	buildRc = d_storageEventBuilder.packMessage(type,
	d_config.partitionId(),
	0, // flags
	journalOffsetWords,
	journalRecordBlobBuffer);
	}

[dorjesinpo]

I would prefer a version calling packMessage once, loop is ok, BSLS_PERFORMANCEHINT_PREDICT_UNLIKELY is a good idea

[678098]

Suggested change

	// be primary for one or more partitions, but as per the BMQ
	// be primary for one or more partitions, but as per the BlazingMQ

[678098]

One small comment

[678098]

This declaration doesn't have a definition now, right?

[678098]

Suggested change

static const char* validationResult(const ValidationResult& result);