[ https://issues.apache.org/jira/browse/IGNITE-8661?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16515582#comment-16515582 ]
Alexey Goncharuk commented on IGNITE-8661: ------------------------------------------ Dmitriy, I have comments regarding the changes in DatabaseSharedManager. I think it makes sense to create an object which will contain the recovery state - lastRead, needMemoryRecovery, lastArchivedSegment and will implement the Supplier interface. This will be more readable, because currently you use AtomicReference as a container rather than as a synchronization primitive. > WALItreater is not stopped if can not deserialize record > --------------------------------------------------------- > > Key: IGNITE-8661 > URL: https://issues.apache.org/jira/browse/IGNITE-8661 > Project: Ignite > Issue Type: Bug > Affects Versions: 2.5 > Reporter: Dmitriy Govorukhin > Assignee: Dmitriy Govorukhin > Priority: Major > Fix For: 2.6 > > > Currently, we have the following code in RecordV1Serializer.readWithCrc: > {code:java} > static WALRecord readWithCrc(.....) throws EOFException, > IgniteCheckedException { > > try (FileInput.Crc32CheckingFileInput in = in0.startRead(skipCrc)) { > ..... > } > catch (EOFException | SegmentEofException | > WalSegmentTailReachedException e) { > throw e; > } > catch (Exception e) { > throw new IgniteCheckedException("Failed to read WAL record at > position: " + startPos, e); > } > } > {code} > So, any runtime error will be remapped to IgniteCheckedException, which will > lead to iterator stop due to the following code: > AbstractWalRecordsIterator.advanceRecord: > {code} > try { > ...... > } > catch (IOException | IgniteCheckedException e) { > if (e instanceof WalSegmentTailReachedException) > throw (WalSegmentTailReachedException)e; > if (!(e instanceof SegmentEofException)) > handleRecordException(e, actualFilePtr); > return null; > } > {code} > Any IgniteCheckedException will be ignored and iterator goes ahead to the > next segment. > I suggest to make the following changes: > 1) It is unexpected behavior, need to fix it. We should only stop iteration > on known exceptions > 2) Also, need to provide ability skip records by type or some pointer for the > StandaloneWalRecordsIterator -- This message was sent by Atlassian JIRA (v7.6.3#76005)