Avoid problem that topic becomes unavailable due to failure of cursor recovery

[massakam]

Motivation

This is a change to avoid #2666.

If a topic is under the following condition and the broker is restarted, that topic becomes unavailable and we can not send or receive messages.

• The managed ledger contains two ledgers and the newer ledger is empty

ledgerInfo {
  ledgerId: 65425
  entries: 2
  size: 238
  timestamp: 1506510325426
}
ledgerInfo {
  ledgerId: 1952728
  timestamp: 0
}

• cursorsLedgerId included in the managed cursor info is -1
• markDeleteLedgerId is between the two ledgers contained in the managed ledger

cursorsLedgerId: -1
markDeleteLedgerId: 1952575
markDeleteEntryId: -1

This is because position gets ahead of the last position of the managed ledger, and IllegalArgumentException occurs in the following line.

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

Line 358 in d5e88c1

messagesConsumedCounter = -getNumberOfEntries(Range.openClosed(position, ledger.getLastPosition()));

Modifications

If position gets ahead of the last position of the managed ledger when recovering the cursor, position is adjusted to the last position.

Result

Even a topic with metadata as described above, the cursor will be recovered successfully.

[massakam]

rerun cpp tests

[massakam]

rerun cpp tests

[merlimat]

@massakam The change LGTM. It would be nice to have a test that reproduces this issue in the Mocked managed ledger tests, to ensure we won't have a regression in the future.

[massakam]

rerun cpp tests

[massakam]

rerun java8 tests

[massakam]

@merlimat I have added the test.

[merlimat]

👍