Examples with GridDhtPartitionState org.apache.ignite.internal.processors.cache.distributed.dht.topology.GridDhtPartitionState used on opensource projects

Example 16 with GridDhtPartitionState

use of org.apache.ignite.internal.processors.cache.distributed.dht.topology.GridDhtPartitionState in project ignite by apache.

the class GridCacheDatabaseSharedManager method dumpPartitionsInfo.

/**
 * Retrieves from page memory meta information about given {@code grp} group partitions
 * and dumps this information to log INFO level.
 *
 * @param grp Cache group.
 * @param log Logger.
 * @throws IgniteCheckedException If failed.
 */
private static void dumpPartitionsInfo(CacheGroupContext grp, IgniteLogger log) throws IgniteCheckedException {
    PageMemoryEx pageMem = (PageMemoryEx) grp.dataRegion().pageMemory();
    IgnitePageStoreManager pageStore = grp.shared().pageStore();
    assert pageStore != null : "Persistent cache should have initialize page store manager.";
    for (int p = 0; p < grp.affinity().partitions(); p++) {
        GridDhtLocalPartition part = grp.topology().localPartition(p);
        if (part != null) {
            if (log.isInfoEnabled())
                log.info("Partition [grp=" + grp.cacheOrGroupName() + ", id=" + p + ", state=" + part.state() + ", counter=" + part.dataStore().partUpdateCounter() + ", size=" + part.fullSize() + "]");
            continue;
        }
        if (!pageStore.exists(grp.groupId(), p))
            continue;
        pageStore.ensure(grp.groupId(), p);
        if (pageStore.pages(grp.groupId(), p) <= 1) {
            if (log.isInfoEnabled())
                log.info("Partition [grp=" + grp.cacheOrGroupName() + ", id=" + p + ", state=N/A (only file header) ]");
            continue;
        }
        long partMetaId = pageMem.partitionMetaPageId(grp.groupId(), p);
        long partMetaPage = pageMem.acquirePage(grp.groupId(), partMetaId);
        try {
            long pageAddr = pageMem.readLock(grp.groupId(), partMetaId, partMetaPage);
            try {
                PagePartitionMetaIO io = PagePartitionMetaIO.VERSIONS.forPage(pageAddr);
                GridDhtPartitionState partState = fromOrdinal(io.getPartitionState(pageAddr));
                String state = partState != null ? partState.toString() : "N/A";
                long updateCntr = io.getUpdateCounter(pageAddr);
                long size = io.getSize(pageAddr);
                if (log.isInfoEnabled())
                    log.info("Partition [grp=" + grp.cacheOrGroupName() + ", id=" + p + ", state=" + state + ", counter=" + updateCntr + ", size=" + size + "]");
            } finally {
                pageMem.readUnlock(grp.groupId(), partMetaId, partMetaPage);
            }
        } finally {
            pageMem.releasePage(grp.groupId(), partMetaId, partMetaPage);
        }
    }
}

Example 17 with GridDhtPartitionState

use of org.apache.ignite.internal.processors.cache.distributed.dht.topology.GridDhtPartitionState 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 18 with GridDhtPartitionState

use of org.apache.ignite.internal.processors.cache.distributed.dht.topology.GridDhtPartitionState in project ignite by apache.

the class GridDhtPartitionsExchangeFuture method assignPartitionSizes.

/**
 * @param top Topology.
 */
private void assignPartitionSizes(GridDhtPartitionTopology top) {
    Map<Integer, Long> partSizes = new HashMap<>();
    for (Map.Entry<UUID, GridDhtPartitionsSingleMessage> e : msgs.entrySet()) {
        GridDhtPartitionsSingleMessage singleMsg = e.getValue();
        GridDhtPartitionMap partMap = singleMsg.partitions().get(top.groupId());
        if (partMap == null)
            continue;
        for (Map.Entry<Integer, GridDhtPartitionState> e0 : partMap.entrySet()) {
            int p = e0.getKey();
            GridDhtPartitionState state = e0.getValue();
            if (state == GridDhtPartitionState.OWNING)
                partSizes.put(p, singleMsg.partitionSizes(top.groupId()).get(p));
        }
    }
    for (GridDhtLocalPartition locPart : top.currentLocalPartitions()) {
        if (locPart.state() == GridDhtPartitionState.OWNING)
            partSizes.put(locPart.id(), locPart.fullSize());
    }
    top.globalPartSizes(partSizes);
}

Example 19 with GridDhtPartitionState

use of org.apache.ignite.internal.processors.cache.distributed.dht.topology.GridDhtPartitionState in project ignite by apache.

the class SchemaIndexCachePartitionWorker method processPartition.

/**
 * Process partition.
 *
 * @throws IgniteCheckedException If failed.
 */
private void processPartition() throws IgniteCheckedException {
    if (stop())
        return;
    checkCancelled();
    boolean reserved = false;
    GridDhtPartitionState partState = locPart.state();
    if (partState != EVICTED)
        reserved = (partState == OWNING || partState == MOVING || partState == LOST) && locPart.reserve();
    if (!reserved)
        return;
    try {
        GridCursor<? extends CacheDataRow> cursor = locPart.dataStore().cursor(cctx.cacheId(), null, null, KEY_ONLY);
        boolean locked = false;
        try {
            int cntr = 0;
            while (!stop() && cursor.next()) {
                KeyCacheObject key = cursor.get().key();
                if (!locked) {
                    cctx.shared().database().checkpointReadLock();
                    locked = true;
                }
                processKey(key);
                if (++cntr % batchSize == 0) {
                    cctx.shared().database().checkpointReadUnlock();
                    locked = false;
                }
                cctx.cache().metrics0().addIndexRebuildKeyProcessed(1);
                if (locPart.state() == RENTING)
                    break;
            }
            wrappedClo.addNumberProcessedKeys(cntr);
        } finally {
            if (locked)
                cctx.shared().database().checkpointReadUnlock();
        }
    } finally {
        locPart.release();
        if (partsCnt.getAndUpdate(v -> v > 0 ? v - 1 : 0) > 0)
            cctx.group().metrics().decrementIndexBuildCountPartitionsLeft();
    }
}

Example 20 with GridDhtPartitionState

use of org.apache.ignite.internal.processors.cache.distributed.dht.topology.GridDhtPartitionState in project ignite by apache.

the class GridCacheDhtPreloadDelayedSelfTest method testAutomaticPreloadWithEmptyCache.

/**
 * @throws Exception If failed.
 */
@Test
public void testAutomaticPreloadWithEmptyCache() throws Exception {
    preloadMode = SYNC;
    delay = 0;
    Collection<Ignite> ignites = new ArrayList<>();
    try {
        for (int i = 0; i < 5; i++) {
            ignites.add(startGrid(i));
            if (persistenceEnabled()) {
                if (i == 0)
                    grid(0).cluster().active(true);
                else
                    resetBaselineTopology();
            }
            awaitPartitionMapExchange(true, true, null, false);
            for (Ignite g : ignites) {
                info(">>> Checking affinity for grid: " + g.name());
                GridDhtPartitionTopology top = topology(g);
                GridDhtPartitionFullMap fullMap = top.partitionMap(true);
                for (Map.Entry<UUID, GridDhtPartitionMap> fe : fullMap.entrySet()) {
                    UUID nodeId = fe.getKey();
                    GridDhtPartitionMap m = fe.getValue();
                    for (Map.Entry<Integer, GridDhtPartitionState> e : m.entrySet()) {
                        int p = e.getKey();
                        GridDhtPartitionState state = e.getValue();
                        Collection<ClusterNode> nodes = affinityNodes(g, p);
                        Collection<UUID> nodeIds = U.nodeIds(nodes);
                        assert nodeIds.contains(nodeId) : "Invalid affinity mapping [nodeId=" + nodeId + ", part=" + p + ", state=" + state + ", igniteInstanceName=" + G.ignite(nodeId).name() + ", affNames=" + U.nodes2names(nodes) + ", affIds=" + nodeIds + ']';
                    }
                }
            }
        }
    } finally {
        stopAllGrids();
    }
}