Examples with StripedExecutor org.apache.ignite.internal.util.StripedExecutor used on opensource projects

Example 1 with StripedExecutor

use of org.apache.ignite.internal.util.StripedExecutor in project ignite by apache.

the class IgniteKernal method createExecutorDescription.

/**
 * Create description of an executor service for logging.
 *
 * @param execSvcName Name of the service.
 * @param execSvc Service to create a description for.
 */
private String createExecutorDescription(String execSvcName, ExecutorService execSvc) {
    int poolSize = 0;
    int poolActiveThreads = 0;
    int poolQSize = 0;
    if (execSvc instanceof ThreadPoolExecutor) {
        ThreadPoolExecutor exec = (ThreadPoolExecutor) execSvc;
        poolSize = exec.getPoolSize();
        poolActiveThreads = Math.min(poolSize, exec.getActiveCount());
        poolQSize = exec.getQueue().size();
    } else if (execSvc instanceof StripedExecutor) {
        StripedExecutor exec = (StripedExecutor) execSvc;
        poolSize = exec.stripesCount();
        poolActiveThreads = exec.activeStripesCount();
        poolQSize = exec.queueSize();
    }
    int poolIdleThreads = poolSize - poolActiveThreads;
    return execSvcName + " [active=" + poolActiveThreads + ", idle=" + poolIdleThreads + ", qSize=" + poolQSize + "]";
}

Example 2 with StripedExecutor

use of org.apache.ignite.internal.util.StripedExecutor in project ignite by apache.

the class GridCacheDatabaseSharedManager method applyLogicalUpdates.

/**
 * @param status Last registered checkpoint status.
 * @param restoreMeta Metastore restore phase if {@code true}.
 * @throws IgniteCheckedException If failed to apply updates.
 * @throws StorageException If IO exception occurred while reading write-ahead log.
 */
private RestoreLogicalState applyLogicalUpdates(CheckpointStatus status, IgnitePredicate<Integer> cacheGroupsPredicate, IgniteBiPredicate<WALRecord.RecordType, WALPointer> recordTypePredicate, boolean restoreMeta) throws IgniteCheckedException {
    if (log.isInfoEnabled())
        log.info("Applying lost " + (restoreMeta ? "metastore" : "cache") + " updates since last checkpoint record [lastMarked=" + status.startPtr + ", lastCheckpointId=" + status.cpStartId + ']');
    if (!restoreMeta)
        cctx.kernalContext().query().skipFieldLookup(true);
    long start = U.currentTimeMillis();
    AtomicReference<Throwable> applyError = new AtomicReference<>();
    AtomicLong applied = new AtomicLong();
    long lastArchivedSegment = cctx.wal().lastArchivedSegment();
    StripedExecutor exec = cctx.kernalContext().pools().getStripedExecutorService();
    Semaphore semaphore = new Semaphore(semaphorePertmits(exec));
    Map<GroupPartitionId, Integer> partitionRecoveryStates = new HashMap<>();
    WALIterator it = cctx.wal().replay(status.startPtr, recordTypePredicate);
    RestoreLogicalState restoreLogicalState = new RestoreLogicalState(status, it, lastArchivedSegment, cacheGroupsPredicate, partitionRecoveryStates);
    final IgniteTxManager txManager = cctx.tm();
    try {
        while (restoreLogicalState.hasNext()) {
            WALRecord rec = restoreLogicalState.next();
            if (rec == null)
                break;
            switch(rec.type()) {
                case TX_RECORD:
                    if (restoreMeta) {
                        // Also restore tx states.
                        TxRecord txRec = (TxRecord) rec;
                        txManager.collectTxStates(txRec);
                    }
                    break;
                case // Calculate initial partition states
                CHECKPOINT_RECORD:
                    CheckpointRecord cpRec = (CheckpointRecord) rec;
                    for (Map.Entry<Integer, CacheState> entry : cpRec.cacheGroupStates().entrySet()) {
                        CacheState cacheState = entry.getValue();
                        for (int i = 0; i < cacheState.size(); i++) {
                            int partId = cacheState.partitionByIndex(i);
                            byte state = cacheState.stateByIndex(i);
                            // Ignore undefined state.
                            if (state != -1) {
                                partitionRecoveryStates.put(new GroupPartitionId(entry.getKey(), partId), (int) state);
                            }
                        }
                    }
                    break;
                case ROLLBACK_TX_RECORD:
                    RollbackRecord rbRec = (RollbackRecord) rec;
                    CacheGroupContext ctx = cctx.cache().cacheGroup(rbRec.groupId());
                    if (ctx != null && !ctx.isLocal()) {
                        GridDhtLocalPartition part = ctx.topology().forceCreatePartition(rbRec.partitionId());
                        ctx.offheap().dataStore(part).updateInitialCounter(rbRec.start(), rbRec.range());
                    }
                    break;
                case MVCC_DATA_RECORD:
                case DATA_RECORD:
                case DATA_RECORD_V2:
                case ENCRYPTED_DATA_RECORD:
                case ENCRYPTED_DATA_RECORD_V2:
                case ENCRYPTED_DATA_RECORD_V3:
                    DataRecord dataRec = (DataRecord) rec;
                    int entryCnt = dataRec.entryCount();
                    for (int i = 0; i < entryCnt; i++) {
                        DataEntry dataEntry = dataRec.get(i);
                        if (!restoreMeta && txManager.uncommitedTx(dataEntry))
                            continue;
                        int cacheId = dataEntry.cacheId();
                        DynamicCacheDescriptor cacheDesc = cctx.cache().cacheDescriptor(cacheId);
                        // Can empty in case recovery node on blt changed.
                        if (cacheDesc == null)
                            continue;
                        stripedApply(() -> {
                            GridCacheContext cacheCtx = cctx.cacheContext(cacheId);
                            if (skipRemovedIndexUpdates(cacheCtx.groupId(), PageIdAllocator.INDEX_PARTITION))
                                cctx.kernalContext().query().markAsRebuildNeeded(cacheCtx, true);
                            try {
                                applyUpdate(cacheCtx, dataEntry);
                            } catch (IgniteCheckedException e) {
                                U.error(log, "Failed to apply data entry, dataEntry=" + dataEntry + ", ptr=" + dataRec.position());
                                applyError.compareAndSet(null, e);
                            }
                            applied.incrementAndGet();
                        }, cacheDesc.groupId(), dataEntry.partitionId(), exec, semaphore);
                    }
                    break;
                case MVCC_TX_RECORD:
                    MvccTxRecord txRecord = (MvccTxRecord) rec;
                    byte txState = convertToTxState(txRecord.state());
                    cctx.coordinators().updateState(txRecord.mvccVersion(), txState, true);
                    break;
                case PART_META_UPDATE_STATE:
                    PartitionMetaStateRecord metaStateRecord = (PartitionMetaStateRecord) rec;
                    GroupPartitionId groupPartitionId = new GroupPartitionId(metaStateRecord.groupId(), metaStateRecord.partitionId());
                    restoreLogicalState.partitionRecoveryStates.put(groupPartitionId, (int) metaStateRecord.state());
                    break;
                case METASTORE_DATA_RECORD:
                    MetastoreDataRecord metastoreDataRecord = (MetastoreDataRecord) rec;
                    metaStorage.applyUpdate(metastoreDataRecord.key(), metastoreDataRecord.value());
                    break;
                case META_PAGE_UPDATE_NEXT_SNAPSHOT_ID:
                case META_PAGE_UPDATE_LAST_SUCCESSFUL_SNAPSHOT_ID:
                case META_PAGE_UPDATE_LAST_SUCCESSFUL_FULL_SNAPSHOT_ID:
                case META_PAGE_UPDATE_LAST_ALLOCATED_INDEX:
                    PageDeltaRecord pageDelta = (PageDeltaRecord) rec;
                    stripedApplyPage((pageMem) -> {
                        try {
                            applyPageDelta(pageMem, pageDelta, false);
                        } catch (IgniteCheckedException e) {
                            U.error(log, "Failed to apply page delta, " + pageDelta);
                            applyError.compareAndSet(null, e);
                        }
                    }, pageDelta.groupId(), partId(pageDelta.pageId()), exec, semaphore);
                    break;
                case MASTER_KEY_CHANGE_RECORD_V2:
                    cctx.kernalContext().encryption().applyKeys((MasterKeyChangeRecordV2) rec);
                    break;
                case REENCRYPTION_START_RECORD:
                    cctx.kernalContext().encryption().applyReencryptionStartRecord((ReencryptionStartRecord) rec);
                    break;
                case INDEX_ROOT_PAGE_RENAME_RECORD:
                    IndexRenameRootPageRecord record = (IndexRenameRootPageRecord) rec;
                    int cacheId = record.cacheId();
                    GridCacheContext cacheCtx = cctx.cacheContext(cacheId);
                    if (cacheCtx != null) {
                        IgniteCacheOffheapManager offheap = cacheCtx.offheap();
                        for (int i = 0; i < record.segments(); i++) offheap.renameRootPageForIndex(cacheId, record.oldTreeName(), record.newTreeName(), i);
                    }
                    break;
                case PARTITION_CLEARING_START_RECORD:
                    PartitionClearingStartRecord rec0 = (PartitionClearingStartRecord) rec;
                    CacheGroupContext grp = this.ctx.cache().cacheGroup(rec0.groupId());
                    if (grp != null) {
                        GridDhtLocalPartition part;
                        try {
                            part = grp.topology().forceCreatePartition(rec0.partitionId());
                        } catch (IgniteCheckedException e) {
                            throw new IgniteException("Cannot get or create a partition [groupId=" + rec0.groupId() + ", partitionId=" + rec0.partitionId() + "]", e);
                        }
                        stripedApply(() -> {
                            try {
                                part.updateClearVersion(rec0.clearVersion());
                                IgniteInternalFuture<?> clearFut = grp.shared().evict().evictPartitionAsync(grp, part, new GridFutureAdapter<>());
                                clearFut.get();
                                part.updateClearVersion();
                            } catch (IgniteCheckedException e) {
                                U.error(log, "Failed to apply partition clearing record, " + rec0);
                                applyError.compareAndSet(null, e);
                            }
                        }, rec0.groupId(), rec0.partitionId(), exec, semaphore);
                    }
                    break;
                default:
            }
        }
    } finally {
        it.close();
        if (!restoreMeta)
            cctx.kernalContext().query().skipFieldLookup(false);
    }
    awaitApplyComplete(exec, applyError);
    if (log.isInfoEnabled())
        log.info("Finished applying WAL changes [updatesApplied=" + applied + ", time=" + (U.currentTimeMillis() - start) + " ms]");
    for (DatabaseLifecycleListener lsnr : getDatabaseListeners(cctx.kernalContext())) lsnr.afterLogicalUpdatesApplied(this, restoreLogicalState);
    return restoreLogicalState;
}

Example 3 with StripedExecutor

use of org.apache.ignite.internal.util.StripedExecutor 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;
}