fix getPreviousPosition npe

[gaozhangmin]

Fixes #11619

Motivation

There's a chance that headMap might be empty even if a empty check added. because remove would be executed in ledgers. after the empty check between line 3214 and line 3219 in following code snippet:

headMap is a view of ledgers

pulsar/managed-ledger/src/main/java/org/apache/bookkeeper/mledger/impl/ManagedLedgerImpl.java

Lines 3206 to 3222 in 31231d6

	NavigableMap<Long, LedgerInfo> headMap = ledgers.headMap(position.getLedgerId(), false);
	if (headMap.isEmpty()) {
	// There is no previous ledger, return an invalid position in the current ledger
	return PositionImpl.get(position.getLedgerId(), -1);
	}
	// We need to find the most recent non-empty ledger
	for (long ledgerId : headMap.descendingKeySet()) {
	LedgerInfo li = headMap.get(ledgerId);
	if (li.getEntries() > 0) {
	return PositionImpl.get(li.getLedgerId(), li.getEntries() - 1);
	}
	}
	// in case there are only empty ledgers, we return a position in the first one
	return PositionImpl.get(headMap.firstEntry().getKey(), -1);

Modifications

Add a null checker of headMap.firstEntry

Verifying this change
Make sure that the change passes the CI checks.
This change is a trivial rework / code cleanup without any test coverage.

[BewareMyPower]

This fix may fix the NPE of the original issue. But I thinks there're still possibilities for NPE because when we iterate over the headMap, there's a chance that headMap.get(ledgerId) returns null even if ledgerId is in its key set.
IMO, a better solution is make a copy of ledgers so the copied ledgers is constant during getPreviousPosition.

[eolivelli]

LGTM good catch

unfortunately there is no good way to add a test

[hangc0276]

move to 2.8.2

[hangc0276]

move back to 2.8.1