Examples with CheckpointRecord org.apache.ignite.internal.pagemem.wal.record.CheckpointRecord used on opensource projects

Example 26 with CheckpointRecord

use of org.apache.ignite.internal.pagemem.wal.record.CheckpointRecord in project ignite by apache.

the class GridCacheDatabaseSharedManager method performBinaryMemoryRestore.

/**
 * @param status Checkpoint status.
 * @param cacheGroupsPredicate Cache groups to restore.
 * @throws IgniteCheckedException If failed.
 * @throws StorageException In case I/O error occurred during operations with storage.
 */
private RestoreBinaryState performBinaryMemoryRestore(CheckpointStatus status, IgnitePredicate<Integer> cacheGroupsPredicate, IgniteBiPredicate<WALRecord.RecordType, WALPointer> recordTypePredicate, boolean finalizeState) throws IgniteCheckedException {
    if (log.isInfoEnabled())
        log.info("Checking memory state [lastValidPos=" + status.endPtr + ", lastMarked=" + status.startPtr + ", lastCheckpointId=" + status.cpStartId + ']');
    WALPointer recPtr = status.endPtr;
    boolean apply = status.needRestoreMemory();
    try {
        WALRecord startRec = !CheckpointStatus.NULL_PTR.equals(status.startPtr) || apply ? cctx.wal().read(status.startPtr) : null;
        if (apply) {
            if (finalizeState)
                U.quietAndWarn(log, "Ignite node stopped in the middle of checkpoint. Will restore memory state and " + "finish checkpoint on node start.");
            cctx.cache().cacheGroupDescriptors().forEach((grpId, desc) -> {
                if (!cacheGroupsPredicate.apply(grpId))
                    return;
                try {
                    DataRegion region = cctx.database().dataRegion(desc.config().getDataRegionName());
                    if (region == null || !cctx.isLazyMemoryAllocation(region))
                        return;
                    region.pageMemory().start();
                } catch (IgniteCheckedException e) {
                    throw new IgniteException(e);
                }
            });
            cctx.pageStore().beginRecover();
            if (!(startRec instanceof CheckpointRecord))
                throw new StorageException("Checkpoint marker doesn't point to checkpoint record " + "[ptr=" + status.startPtr + ", rec=" + startRec + "]");
            WALPointer cpMark = ((CheckpointRecord) startRec).checkpointMark();
            if (cpMark != null) {
                if (log.isInfoEnabled())
                    log.info("Restoring checkpoint after logical recovery, will start physical recovery from " + "back pointer: " + cpMark);
                recPtr = cpMark;
            }
        } else
            cctx.wal().notchLastCheckpointPtr(status.startPtr);
    } catch (NoSuchElementException e) {
        throw new StorageException("Failed to read checkpoint record from WAL, persistence consistency " + "cannot be guaranteed. Make sure configuration points to correct WAL folders and WAL folder is " + "properly mounted [ptr=" + status.startPtr + ", walPath=" + persistenceCfg.getWalPath() + ", walArchive=" + persistenceCfg.getWalArchivePath() + "]");
    }
    AtomicReference<Throwable> applyError = new AtomicReference<>();
    StripedExecutor exec = cctx.kernalContext().pools().getStripedExecutorService();
    Semaphore semaphore = new Semaphore(semaphorePertmits(exec));
    long start = U.currentTimeMillis();
    long lastArchivedSegment = cctx.wal().lastArchivedSegment();
    WALIterator it = cctx.wal().replay(recPtr, recordTypePredicate);
    RestoreBinaryState restoreBinaryState = new RestoreBinaryState(status, it, lastArchivedSegment, cacheGroupsPredicate);
    AtomicLong applied = new AtomicLong();
    try {
        while (restoreBinaryState.hasNext()) {
            if (applyError.get() != null)
                break;
            WALRecord rec = restoreBinaryState.next();
            if (rec == null)
                break;
            switch(rec.type()) {
                case PAGE_RECORD:
                    if (restoreBinaryState.needApplyBinaryUpdate()) {
                        PageSnapshot pageSnapshot = (PageSnapshot) rec;
                        // Here we do not require tag check because we may be applying memory changes after
                        // several repetitive restarts and the same pages may have changed several times.
                        int groupId = pageSnapshot.fullPageId().groupId();
                        int partId = partId(pageSnapshot.fullPageId().pageId());
                        if (skipRemovedIndexUpdates(groupId, partId))
                            break;
                        stripedApplyPage((pageMem) -> {
                            try {
                                applyPageSnapshot(pageMem, pageSnapshot);
                                applied.incrementAndGet();
                            } catch (Throwable t) {
                                U.error(log, "Failed to apply page snapshot. rec=[" + pageSnapshot + ']');
                                applyError.compareAndSet(null, (t instanceof IgniteCheckedException) ? (IgniteCheckedException) t : new IgniteCheckedException("Failed to apply page snapshot", t));
                            }
                        }, groupId, partId, exec, semaphore);
                    }
                    break;
                case PART_META_UPDATE_STATE:
                    PartitionMetaStateRecord metaStateRecord = (PartitionMetaStateRecord) rec;
                    {
                        int groupId = metaStateRecord.groupId();
                        int partId = metaStateRecord.partitionId();
                        stripedApplyPage((pageMem) -> {
                            GridDhtPartitionState state = fromOrdinal(metaStateRecord.state());
                            if (state == null || state == GridDhtPartitionState.EVICTED)
                                schedulePartitionDestroy(groupId, partId);
                            else {
                                try {
                                    cancelOrWaitPartitionDestroy(groupId, partId);
                                } catch (Throwable t) {
                                    U.error(log, "Failed to cancel or wait partition destroy. rec=[" + metaStateRecord + ']');
                                    applyError.compareAndSet(null, (t instanceof IgniteCheckedException) ? (IgniteCheckedException) t : new IgniteCheckedException("Failed to cancel or wait partition destroy", t));
                                }
                            }
                        }, groupId, partId, exec, semaphore);
                    }
                    break;
                case PARTITION_DESTROY:
                    PartitionDestroyRecord destroyRecord = (PartitionDestroyRecord) rec;
                    {
                        int groupId = destroyRecord.groupId();
                        int partId = destroyRecord.partitionId();
                        stripedApplyPage((pageMem) -> {
                            pageMem.invalidate(groupId, partId);
                            schedulePartitionDestroy(groupId, partId);
                        }, groupId, partId, exec, semaphore);
                    }
                    break;
                default:
                    if (restoreBinaryState.needApplyBinaryUpdate() && rec instanceof PageDeltaRecord) {
                        PageDeltaRecord pageDelta = (PageDeltaRecord) rec;
                        int groupId = pageDelta.groupId();
                        int partId = partId(pageDelta.pageId());
                        if (skipRemovedIndexUpdates(groupId, partId))
                            break;
                        stripedApplyPage((pageMem) -> {
                            try {
                                applyPageDelta(pageMem, pageDelta, true);
                                applied.incrementAndGet();
                            } catch (Throwable t) {
                                U.error(log, "Failed to apply page delta. rec=[" + pageDelta + ']');
                                applyError.compareAndSet(null, (t instanceof IgniteCheckedException) ? (IgniteCheckedException) t : new IgniteCheckedException("Failed to apply page delta", t));
                            }
                        }, groupId, partId, exec, semaphore);
                    }
            }
        }
    } finally {
        it.close();
        awaitApplyComplete(exec, applyError);
    }
    if (!finalizeState)
        return null;
    WALPointer lastReadPtr = restoreBinaryState.lastReadRecordPointer();
    if (status.needRestoreMemory()) {
        if (restoreBinaryState.needApplyBinaryUpdate())
            throw new StorageException("Failed to restore memory state (checkpoint marker is present " + "on disk, but checkpoint record is missed in WAL) " + "[cpStatus=" + status + ", lastRead=" + lastReadPtr + "]");
        if (log.isInfoEnabled())
            log.info("Finished applying memory changes [changesApplied=" + applied + ", time=" + (U.currentTimeMillis() - start) + " ms]");
        finalizeCheckpointOnRecovery(status.cpStartTs, status.cpStartId, status.startPtr, exec);
    }
    return restoreBinaryState;
}

Example 27 with CheckpointRecord

use of org.apache.ignite.internal.pagemem.wal.record.CheckpointRecord in project ignite by apache.

the class CheckpointWorkflow method markCheckpointBegin.

/**
 * First stage of checkpoint which collects demanded information(dirty pages mostly).
 *
 * @param cpTs Checkpoint start timestamp.
 * @param curr Current checkpoint event info.
 * @param tracker Checkpoint metrics tracker.
 * @param workProgressDispatcher Work progress dispatcher.
 * @return Checkpoint collected info.
 * @throws IgniteCheckedException if fail.
 */
public Checkpoint markCheckpointBegin(long cpTs, CheckpointProgressImpl curr, CheckpointMetricsTracker tracker, WorkProgressDispatcher workProgressDispatcher) throws IgniteCheckedException {
    Collection<DataRegion> checkpointedRegions = dataRegions.get();
    List<CheckpointListener> dbLsnrs = getRelevantCheckpointListeners(checkpointedRegions);
    CheckpointRecord cpRec = new CheckpointRecord(memoryRecoveryRecordPtr);
    memoryRecoveryRecordPtr = null;
    IgniteFuture snapFut = null;
    CheckpointPagesInfoHolder cpPagesHolder;
    int dirtyPagesCount;
    boolean hasPartitionsToDestroy;
    WALPointer cpPtr = null;
    CheckpointContextImpl ctx0 = new CheckpointContextImpl(curr, new PartitionAllocationMap(), checkpointCollectPagesInfoPool, workProgressDispatcher);
    checkpointReadWriteLock.readLock();
    try {
        for (CheckpointListener lsnr : dbLsnrs) lsnr.beforeCheckpointBegin(ctx0);
        ctx0.awaitPendingTasksFinished();
    } finally {
        checkpointReadWriteLock.readUnlock();
    }
    tracker.onLockWaitStart();
    checkpointReadWriteLock.writeLock();
    try {
        curr.transitTo(LOCK_TAKEN);
        tracker.onMarkStart();
        // Listeners must be invoked before we write checkpoint record to WAL.
        for (CheckpointListener lsnr : dbLsnrs) lsnr.onMarkCheckpointBegin(ctx0);
        ctx0.awaitPendingTasksFinished();
        tracker.onListenersExecuteEnd();
        if (curr.nextSnapshot())
            snapFut = snapshotMgr.onMarkCheckPointBegin(curr.snapshotOperation(), ctx0.partitionStatMap());
        fillCacheGroupState(cpRec);
        // There are allowable to replace pages only after checkpoint entry was stored to disk.
        cpPagesHolder = beginAllCheckpoints(checkpointedRegions, curr.futureFor(MARKER_STORED_TO_DISK));
        curr.currentCheckpointPagesCount(cpPagesHolder.pagesNum());
        dirtyPagesCount = cpPagesHolder.pagesNum();
        hasPartitionsToDestroy = !curr.getDestroyQueue().pendingReqs().isEmpty();
        if (dirtyPagesCount > 0 || curr.nextSnapshot() || hasPartitionsToDestroy) {
            // No page updates for this checkpoint are allowed from now on.
            if (wal != null)
                cpPtr = wal.log(cpRec);
            if (cpPtr == null)
                cpPtr = CheckpointStatus.NULL_PTR;
        }
        curr.transitTo(PAGE_SNAPSHOT_TAKEN);
    } finally {
        checkpointReadWriteLock.writeUnlock();
        tracker.onLockRelease();
    }
    curr.transitTo(LOCK_RELEASED);
    for (CheckpointListener lsnr : dbLsnrs) lsnr.onCheckpointBegin(ctx0);
    if (snapFut != null) {
        try {
            snapFut.get();
        } catch (IgniteException e) {
            U.error(log, "Failed to wait for snapshot operation initialization: " + curr.snapshotOperation(), e);
        }
    }
    if (dirtyPagesCount > 0 || hasPartitionsToDestroy) {
        tracker.onWalCpRecordFsyncStart();
        // Sync log outside the checkpoint write lock.
        if (wal != null)
            wal.flush(cpPtr, true);
        tracker.onWalCpRecordFsyncEnd();
        CheckpointEntry checkpointEntry = null;
        if (checkpointMarkersStorage != null)
            checkpointEntry = checkpointMarkersStorage.writeCheckpointEntry(cpTs, cpRec.checkpointId(), cpPtr, cpRec, CheckpointEntryType.START, skipSync);
        curr.transitTo(MARKER_STORED_TO_DISK);
        tracker.onSplitAndSortCpPagesStart();
        GridConcurrentMultiPairQueue<PageMemoryEx, FullPageId> cpPages = splitAndSortCpPagesIfNeeded(cpPagesHolder);
        tracker.onSplitAndSortCpPagesEnd();
        return new Checkpoint(checkpointEntry, cpPages, curr);
    } else {
        if (curr.nextSnapshot() && wal != null)
            wal.flush(null, true);
        return new Checkpoint(null, GridConcurrentMultiPairQueue.EMPTY, curr);
    }
}

Example 28 with CheckpointRecord

use of org.apache.ignite.internal.pagemem.wal.record.CheckpointRecord in project ignite by apache.

the class IgnitePdsRecoveryAfterFileCorruptionTest method generateWal.

/**
 * @param mem Mem.
 * @param storeMgr Store manager.
 * @param wal Wal.
 * @param cacheId Cache id.
 * @param pages Pages.
 */
private void generateWal(final PageMemoryImpl mem, final IgnitePageStoreManager storeMgr, final IgniteWriteAheadLogManager wal, final int cacheId, FullPageId[] pages) throws Exception {
    // Mark the start position.
    CheckpointRecord cpRec = new CheckpointRecord(null);
    WALPointer start = wal.log(cpRec);
    wal.flush(start, false);
    for (FullPageId fullId : pages) {
        long page = mem.acquirePage(fullId.groupId(), fullId.pageId());
        try {
            long pageAddr = mem.writeLock(fullId.groupId(), fullId.pageId(), page);
            PageIO.setPageId(pageAddr, fullId.pageId());
            try {
                for (int j = PageIO.COMMON_HEADER_END; j < mem.realPageSize(fullId.groupId()); j += 4) PageUtils.putInt(pageAddr, j, j + (int) fullId.pageId());
            } finally {
                mem.writeUnlock(fullId.groupId(), fullId.pageId(), page, null, true);
            }
        } finally {
            mem.releasePage(fullId.groupId(), fullId.pageId(), page);
        }
    }
    Collection<FullPageId> pageIds = mem.beginCheckpoint(new GridFinishedFuture());
    info("Acquired pages for checkpoint: " + pageIds.size());
    try {
        long begin = System.currentTimeMillis();
        long cp = 0;
        AtomicLong write = new AtomicLong();
        PageStoreWriter pageStoreWriter = (fullPageId, buf, tag) -> {
            int groupId = fullPageId.groupId();
            long pageId = fullPageId.pageId();
            for (int j = PageIO.COMMON_HEADER_END; j < mem.realPageSize(groupId); j += 4) assertEquals(j + (int) pageId, buf.getInt(j));
            buf.rewind();
            long writeStart = System.nanoTime();
            storeMgr.write(cacheId, pageId, buf, tag, true);
            long writeEnd = System.nanoTime();
            write.getAndAdd(writeEnd - writeStart);
        };
        ByteBuffer tmpBuf = ByteBuffer.allocate(mem.pageSize());
        tmpBuf.order(ByteOrder.nativeOrder());
        for (FullPageId fullId : pages) {
            if (pageIds.contains(fullId)) {
                long cpStart = System.nanoTime();
                mem.checkpointWritePage(fullId, tmpBuf, pageStoreWriter, null);
                long cpEnd = System.nanoTime();
                cp += cpEnd - cpStart;
            }
        }
        long syncStart = System.currentTimeMillis();
        storeMgr.sync(cacheId, 0);
        long end = System.currentTimeMillis();
        info("Written pages in " + (end - begin) + "ms, copy took " + (cp / 1_000_000) + "ms, " + "write took " + (write.get() / 1_000_000) + "ms, sync took " + (end - syncStart) + "ms");
    } finally {
        info("Finishing checkpoint...");
        mem.finishCheckpoint();
        info("Finished checkpoint");
    }
    wal.flush(wal.log(new CheckpointRecord(null)), false);
    for (FullPageId fullId : pages) {
        long page = mem.acquirePage(fullId.groupId(), fullId.pageId());
        try {
            assertFalse("Page has a temp heap copy after the last checkpoint: [cacheId=" + fullId.groupId() + ", pageId=" + fullId.pageId() + "]", mem.hasTempCopy(page));
            assertFalse("Page is dirty after the last checkpoint: [cacheId=" + fullId.groupId() + ", pageId=" + fullId.pageId() + "]", mem.isDirty(fullId.groupId(), fullId.pageId(), page));
        } finally {
            mem.releasePage(fullId.groupId(), fullId.pageId(), page);
        }
    }
}

Example 29 with CheckpointRecord

use of org.apache.ignite.internal.pagemem.wal.record.CheckpointRecord in project ignite by apache.

the class IgnitePdsCorruptedStoreTest method testWalFsyncWriteHeaderFailure.

/**
 * Test node invalidation due to error on WAL write header.
 */
@Test
public void testWalFsyncWriteHeaderFailure() throws Exception {
    IgniteEx ignite = startGrid(0);
    ignite.cluster().active(true);
    ignite.cache(CACHE_NAME1).put(0, 0);
    failingFileIOFactory.createClosure((file, options) -> {
        FileIO delegate = failingFileIOFactory.delegateFactory().create(file, options);
        if (file.getName().endsWith(".wal")) {
            return new FileIODecorator(delegate) {

                @Override
                public int write(ByteBuffer srcBuf) throws IOException {
                    throw new IOException("No space left on device");
                }
            };
        }
        return delegate;
    });
    ignite.context().cache().context().database().checkpointReadLock();
    try {
        ignite.context().cache().context().wal().log(new CheckpointRecord(null), RolloverType.NEXT_SEGMENT);
    } catch (StorageException expected) {
    // No-op.
    } finally {
        ignite.context().cache().context().database().checkpointReadUnlock();
    }
    waitFailure(StorageException.class);
}

Example 30 with CheckpointRecord

use of org.apache.ignite.internal.pagemem.wal.record.CheckpointRecord in project ignite by apache.

the class WalScannerTest method shouldDumpToFileAndLogFoundRecord.

/**
 * @throws Exception If failed.
 */
@Test
public void shouldDumpToFileAndLogFoundRecord() throws Exception {
    // given: File for dumping records and test logger for interception of records.
    File targetFile = Paths.get(U.defaultWorkDirectory(), TEST_DUMP_FILE).toFile();
    long expPageId = 984;
    int grpId = 123;
    IgniteLogger log = mock(IgniteLogger.class);
    when(log.isInfoEnabled()).thenReturn(true);
    ArgumentCaptor<String> valCapture = ArgumentCaptor.forClass(String.class);
    doNothing().when(log).info(valCapture.capture());
    WALIterator mockedIter = mockWalIterator(new IgniteBiTuple<>(NULL_PTR, new PageSnapshot(new FullPageId(expPageId, grpId), dummyPage(1024, expPageId), 1024)), new IgniteBiTuple<>(NULL_PTR, new CheckpointRecord(new WALPointer(5738, 0, 0))), new IgniteBiTuple<>(NULL_PTR, new FixCountRecord(grpId, expPageId, 4)));
    IgniteWalIteratorFactory factory = mock(IgniteWalIteratorFactory.class);
    when(factory.iterator(any(IteratorParametersBuilder.class))).thenReturn(mockedIter);
    Set<T2<Integer, Long>> groupAndPageIds = new HashSet<>();
    groupAndPageIds.add(new T2<>(grpId, expPageId));
    List<String> actualFileRecords = null;
    try {
        // when: Scanning WAL for searching expected page.
        buildWalScanner(withIteratorParameters(), factory).findAllRecordsFor(groupAndPageIds).forEach(printToLog(log).andThen(printToFile(targetFile)));
        actualFileRecords = Files.readAllLines(targetFile.toPath());
    } finally {
        targetFile.delete();
    }
    actualFileRecords = actualFileRecords.stream().filter(it -> it.startsWith("Next WAL record ::")).collect(Collectors.toList());
    // then: Should be find only expected value from file.
    assertEquals(actualFileRecords.size(), 3);
    assertTrue(actualFileRecords.get(0), actualFileRecords.get(0).contains("PageSnapshot ["));
    assertTrue(actualFileRecords.get(1), actualFileRecords.get(1).contains("CheckpointRecord ["));
    assertTrue(actualFileRecords.get(2), actualFileRecords.get(2).contains("FixCountRecord ["));
    // then: Should be find only expected value from log.
    List<String> actualLogRecords = valCapture.getAllValues();
    assertEquals(actualLogRecords.size(), 1);
    assertTrue(actualLogRecords.get(0), actualLogRecords.get(0).contains("PageSnapshot ["));
    assertTrue(actualLogRecords.get(0), actualLogRecords.get(0).contains("CheckpointRecord ["));
    assertTrue(actualLogRecords.get(0), actualLogRecords.get(0).contains("FixCountRecord ["));
}