If we have a managed ledger, ml and we write 2 entries to it, if both entries fail, both will end up calling ManagedLedgerImpl#ledgerClosed with the ledger the write failed on as a parameter.
However, depending on timing, the second call to ledgerClosed could end up adding a new ledger to the ledger list, even though the current ledger is _not_ failing (as the failing ledger was replaced by the first call). This was the cause of a flake in ManagedLedgerErrorsTest#recoverLongTimeAfterMultipleWriteErrors as reported in (#240). However, it's not possible to get a deterministic test for this as the timings need to be very precise. The failing addComplete needs to run before first error handling completes, but the runnable with ledgerClosed for the second failure needs to run after the first error handling completes, but before the write resends from the first error handling complete. [ Full content available at: https://github.com/apache/incubator-pulsar/pull/2573 ] This message was relayed via gitbox.apache.org for [email protected]
