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

Example 6 with GridDhtPartitionState

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

the class CheckpointWorkflow method fillCacheGroupState.

/**
 * Fill cache group state in checkpoint record.
 *
 * @param cpRec Checkpoint record for filling.
 * @throws IgniteCheckedException if fail.
 */
private void fillCacheGroupState(CheckpointRecord cpRec) throws IgniteCheckedException {
    GridCompoundFuture grpHandleFut = checkpointCollectPagesInfoPool == null ? null : new GridCompoundFuture();
    for (CacheGroupContext grp : cacheGroupsContexts.get()) {
        if (grp.isLocal() || !grp.walEnabled())
            continue;
        Runnable r = () -> {
            ArrayList<GridDhtLocalPartition> parts = new ArrayList<>(grp.topology().localPartitions().size());
            for (GridDhtLocalPartition part : grp.topology().currentLocalPartitions()) parts.add(part);
            CacheState state = new CacheState(parts.size());
            for (GridDhtLocalPartition part : parts) {
                GridDhtPartitionState partState = part.state();
                if (partState == LOST)
                    partState = OWNING;
                state.addPartitionState(part.id(), part.dataStore().fullSize(), part.updateCounter(), (byte) partState.ordinal());
            }
            synchronized (cpRec) {
                cpRec.addCacheGroupState(grp.groupId(), state);
            }
        };
        if (checkpointCollectPagesInfoPool == null)
            r.run();
        else
            try {
                GridFutureAdapter<?> res = new GridFutureAdapter<>();
                checkpointCollectPagesInfoPool.execute(U.wrapIgniteFuture(r, res));
                grpHandleFut.add(res);
            } catch (RejectedExecutionException e) {
                assert false : "Task should never be rejected by async runner";
                // to protect from disabled asserts and call to failure handler
                throw new IgniteException(e);
            }
    }
    if (grpHandleFut != null) {
        grpHandleFut.markInitialized();
        grpHandleFut.get();
    }
}

Example 7 with GridDhtPartitionState

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

the class RecordDataV1Serializer method readPlainRecord.

/**
 * Reads {@code WalRecord} of {@code type} from input.
 * Input should be plain(not encrypted).
 *
 * @param type Record type.
 * @param in Input
 * @param encrypted Record was encrypted.
 * @param recordSize Record size.
 * @return Deserialized record.
 * @throws IOException If failed.
 * @throws IgniteCheckedException If failed.
 */
WALRecord readPlainRecord(RecordType type, ByteBufferBackedDataInput in, boolean encrypted, int recordSize) throws IOException, IgniteCheckedException {
    WALRecord res;
    switch(type) {
        case PAGE_RECORD:
            byte[] arr = new byte[pageSize];
            int cacheId = in.readInt();
            long pageId = in.readLong();
            in.readFully(arr);
            res = new PageSnapshot(new FullPageId(pageId, cacheId), arr, encrypted ? realPageSize : pageSize);
            break;
        case CHECKPOINT_RECORD:
            long msb = in.readLong();
            long lsb = in.readLong();
            boolean hasPtr = in.readByte() != 0;
            long idx = hasPtr ? in.readLong() : 0;
            int off = hasPtr ? in.readInt() : 0;
            int len = hasPtr ? in.readInt() : 0;
            Map<Integer, CacheState> states = readPartitionStates(in);
            boolean end = in.readByte() != 0;
            WALPointer walPtr = hasPtr ? new WALPointer(idx, off, len) : null;
            CheckpointRecord cpRec = new CheckpointRecord(new UUID(msb, lsb), walPtr, end);
            cpRec.cacheGroupStates(states);
            res = cpRec;
            break;
        case META_PAGE_INIT:
            cacheId = in.readInt();
            pageId = in.readLong();
            int ioType = in.readUnsignedShort();
            int ioVer = in.readUnsignedShort();
            long treeRoot = in.readLong();
            long reuseListRoot = in.readLong();
            res = new MetaPageInitRecord(cacheId, pageId, ioType, ioVer, treeRoot, reuseListRoot, log);
            break;
        case INDEX_META_PAGE_DELTA_RECORD:
            res = new MetaPageUpdateIndexDataRecord(in);
            break;
        case PARTITION_META_PAGE_UPDATE_COUNTERS:
            res = new MetaPageUpdatePartitionDataRecord(in);
            break;
        case PARTITION_META_PAGE_UPDATE_COUNTERS_V2:
            res = new MetaPageUpdatePartitionDataRecordV2(in);
            break;
        case PARTITION_META_PAGE_DELTA_RECORD_V3:
            res = new MetaPageUpdatePartitionDataRecordV3(in);
            break;
        case MEMORY_RECOVERY:
            long ts = in.readLong();
            res = new MemoryRecoveryRecord(ts);
            break;
        case PARTITION_DESTROY:
            cacheId = in.readInt();
            int partId = in.readInt();
            res = new PartitionDestroyRecord(cacheId, partId);
            break;
        case DATA_RECORD:
        case DATA_RECORD_V2:
            int entryCnt = in.readInt();
            if (entryCnt == 1)
                res = new DataRecord(readPlainDataEntry(in, type), 0L);
            else {
                List<DataEntry> entries = new ArrayList<>(entryCnt);
                for (int i = 0; i < entryCnt; i++) entries.add(readPlainDataEntry(in, type));
                res = new DataRecord(entries, 0L);
            }
            break;
        case ENCRYPTED_DATA_RECORD:
        case ENCRYPTED_DATA_RECORD_V2:
        case ENCRYPTED_DATA_RECORD_V3:
            entryCnt = in.readInt();
            if (entryCnt == 1)
                res = new DataRecord(readEncryptedDataEntry(in, type), 0L);
            else {
                List<DataEntry> entries = new ArrayList<>(entryCnt);
                for (int i = 0; i < entryCnt; i++) entries.add(readEncryptedDataEntry(in, type));
                res = new DataRecord(entries, 0L);
            }
            break;
        case METASTORE_DATA_RECORD:
            int strLen = in.readInt();
            byte[] strBytes = new byte[strLen];
            in.readFully(strBytes);
            String key = new String(strBytes);
            int valLen = in.readInt();
            assert valLen >= 0;
            byte[] val;
            if (valLen > 0) {
                val = new byte[valLen];
                in.readFully(val);
            } else
                val = null;
            return new MetastoreDataRecord(key, val);
        case HEADER_RECORD:
            long magic = in.readLong();
            if (magic != HeaderRecord.REGULAR_MAGIC && magic != HeaderRecord.COMPACTED_MAGIC)
                throw new EOFException("Magic is corrupted [actual=" + U.hexLong(magic) + ']');
            int ver = in.readInt();
            res = new HeaderRecord(ver);
            break;
        case DATA_PAGE_INSERT_RECORD:
            {
                cacheId = in.readInt();
                pageId = in.readLong();
                int size = in.readUnsignedShort();
                in.ensure(size);
                byte[] payload = new byte[size];
                in.readFully(payload);
                res = new DataPageInsertRecord(cacheId, pageId, payload);
                break;
            }
        case DATA_PAGE_UPDATE_RECORD:
            {
                cacheId = in.readInt();
                pageId = in.readLong();
                int itemId = in.readInt();
                int size = in.readUnsignedShort();
                in.ensure(size);
                byte[] payload = new byte[size];
                in.readFully(payload);
                res = new DataPageUpdateRecord(cacheId, pageId, itemId, payload);
                break;
            }
        case DATA_PAGE_INSERT_FRAGMENT_RECORD:
            {
                cacheId = in.readInt();
                pageId = in.readLong();
                final long lastLink = in.readLong();
                final int payloadSize = in.readInt();
                final byte[] payload = new byte[payloadSize];
                in.readFully(payload);
                res = new DataPageInsertFragmentRecord(cacheId, pageId, payload, lastLink);
                break;
            }
        case DATA_PAGE_REMOVE_RECORD:
            cacheId = in.readInt();
            pageId = in.readLong();
            int itemId = in.readUnsignedByte();
            res = new DataPageRemoveRecord(cacheId, pageId, itemId);
            break;
        case DATA_PAGE_SET_FREE_LIST_PAGE:
            cacheId = in.readInt();
            pageId = in.readLong();
            long freeListPage = in.readLong();
            res = new DataPageSetFreeListPageRecord(cacheId, pageId, freeListPage);
            break;
        case MVCC_DATA_PAGE_MARK_UPDATED_RECORD:
            cacheId = in.readInt();
            pageId = in.readLong();
            itemId = in.readInt();
            long newMvccCrd = in.readLong();
            long newMvccCntr = in.readLong();
            int newMvccOpCntr = in.readInt();
            res = new DataPageMvccMarkUpdatedRecord(cacheId, pageId, itemId, newMvccCrd, newMvccCntr, newMvccOpCntr);
            break;
        case MVCC_DATA_PAGE_TX_STATE_HINT_UPDATED_RECORD:
            cacheId = in.readInt();
            pageId = in.readLong();
            itemId = in.readInt();
            byte txState = in.readByte();
            res = new DataPageMvccUpdateTxStateHintRecord(cacheId, pageId, itemId, txState);
            break;
        case MVCC_DATA_PAGE_NEW_TX_STATE_HINT_UPDATED_RECORD:
            cacheId = in.readInt();
            pageId = in.readLong();
            itemId = in.readInt();
            byte newTxState = in.readByte();
            res = new DataPageMvccUpdateNewTxStateHintRecord(cacheId, pageId, itemId, newTxState);
            break;
        case INIT_NEW_PAGE_RECORD:
            cacheId = in.readInt();
            pageId = in.readLong();
            ioType = in.readUnsignedShort();
            ioVer = in.readUnsignedShort();
            long virtualPageId = in.readLong();
            res = new InitNewPageRecord(cacheId, pageId, ioType, ioVer, virtualPageId, log);
            break;
        case BTREE_META_PAGE_INIT_ROOT:
            cacheId = in.readInt();
            pageId = in.readLong();
            long rootId = in.readLong();
            res = new MetaPageInitRootRecord(cacheId, pageId, rootId);
            break;
        case BTREE_META_PAGE_INIT_ROOT2:
            cacheId = in.readInt();
            pageId = in.readLong();
            long rootId2 = in.readLong();
            int inlineSize = in.readShort();
            res = new MetaPageInitRootInlineRecord(cacheId, pageId, rootId2, inlineSize);
            break;
        case BTREE_META_PAGE_INIT_ROOT_V3:
            cacheId = in.readInt();
            pageId = in.readLong();
            long rootId3 = in.readLong();
            int inlineSize3 = in.readShort();
            long flags = in.readLong();
            byte[] revHash = new byte[IgniteProductVersion.REV_HASH_SIZE];
            byte maj = in.readByte();
            byte min = in.readByte();
            byte maint = in.readByte();
            long verTs = in.readLong();
            in.readFully(revHash);
            IgniteProductVersion createdVer = new IgniteProductVersion(maj, min, maint, verTs, revHash);
            res = new MetaPageInitRootInlineFlagsCreatedVersionRecord(cacheId, pageId, rootId3, inlineSize3, flags, createdVer);
            break;
        case BTREE_META_PAGE_ADD_ROOT:
            cacheId = in.readInt();
            pageId = in.readLong();
            rootId = in.readLong();
            res = new MetaPageAddRootRecord(cacheId, pageId, rootId);
            break;
        case BTREE_META_PAGE_CUT_ROOT:
            cacheId = in.readInt();
            pageId = in.readLong();
            res = new MetaPageCutRootRecord(cacheId, pageId);
            break;
        case BTREE_INIT_NEW_ROOT:
            cacheId = in.readInt();
            pageId = in.readLong();
            rootId = in.readLong();
            ioType = in.readUnsignedShort();
            ioVer = in.readUnsignedShort();
            long leftId = in.readLong();
            long rightId = in.readLong();
            BPlusIO<?> io = BPlusIO.getBPlusIO(ioType, ioVer);
            byte[] rowBytes = new byte[io.getItemSize()];
            in.readFully(rowBytes);
            res = new NewRootInitRecord<>(cacheId, pageId, rootId, (BPlusInnerIO<?>) io, leftId, rowBytes, rightId);
            break;
        case BTREE_PAGE_RECYCLE:
            cacheId = in.readInt();
            pageId = in.readLong();
            long newPageId = in.readLong();
            res = new RecycleRecord(cacheId, pageId, newPageId);
            break;
        case BTREE_PAGE_INSERT:
            cacheId = in.readInt();
            pageId = in.readLong();
            ioType = in.readUnsignedShort();
            ioVer = in.readUnsignedShort();
            int itemIdx = in.readUnsignedShort();
            rightId = in.readLong();
            io = BPlusIO.getBPlusIO(ioType, ioVer);
            rowBytes = new byte[io.getItemSize()];
            in.readFully(rowBytes);
            res = new InsertRecord<>(cacheId, pageId, io, itemIdx, rowBytes, rightId);
            break;
        case BTREE_FIX_LEFTMOST_CHILD:
            cacheId = in.readInt();
            pageId = in.readLong();
            rightId = in.readLong();
            res = new FixLeftmostChildRecord(cacheId, pageId, rightId);
            break;
        case BTREE_FIX_COUNT:
            cacheId = in.readInt();
            pageId = in.readLong();
            int cnt = in.readUnsignedShort();
            res = new FixCountRecord(cacheId, pageId, cnt);
            break;
        case BTREE_PAGE_REPLACE:
            cacheId = in.readInt();
            pageId = in.readLong();
            ioType = in.readUnsignedShort();
            ioVer = in.readUnsignedShort();
            itemIdx = in.readUnsignedShort();
            io = BPlusIO.getBPlusIO(ioType, ioVer);
            rowBytes = new byte[io.getItemSize()];
            in.readFully(rowBytes);
            res = new ReplaceRecord<>(cacheId, pageId, io, rowBytes, itemIdx);
            break;
        case BTREE_PAGE_REMOVE:
            cacheId = in.readInt();
            pageId = in.readLong();
            itemIdx = in.readUnsignedShort();
            cnt = in.readUnsignedShort();
            res = new RemoveRecord(cacheId, pageId, itemIdx, cnt);
            break;
        case BTREE_PAGE_INNER_REPLACE:
            cacheId = in.readInt();
            pageId = in.readLong();
            int dstIdx = in.readUnsignedShort();
            long srcPageId = in.readLong();
            int srcIdx = in.readUnsignedShort();
            long rmvId = in.readLong();
            res = new InnerReplaceRecord<>(cacheId, pageId, dstIdx, srcPageId, srcIdx, rmvId);
            break;
        case BTREE_FORWARD_PAGE_SPLIT:
            cacheId = in.readInt();
            pageId = in.readLong();
            long fwdId = in.readLong();
            ioType = in.readUnsignedShort();
            ioVer = in.readUnsignedShort();
            srcPageId = in.readLong();
            int mid = in.readUnsignedShort();
            cnt = in.readUnsignedShort();
            res = new SplitForwardPageRecord(cacheId, pageId, fwdId, ioType, ioVer, srcPageId, mid, cnt);
            break;
        case BTREE_EXISTING_PAGE_SPLIT:
            cacheId = in.readInt();
            pageId = in.readLong();
            mid = in.readUnsignedShort();
            fwdId = in.readLong();
            res = new SplitExistingPageRecord(cacheId, pageId, mid, fwdId);
            break;
        case BTREE_PAGE_MERGE:
            cacheId = in.readInt();
            pageId = in.readLong();
            long prntId = in.readLong();
            int prntIdx = in.readUnsignedShort();
            rightId = in.readLong();
            boolean emptyBranch = in.readBoolean();
            res = new MergeRecord<>(cacheId, pageId, prntId, prntIdx, rightId, emptyBranch);
            break;
        case BTREE_FIX_REMOVE_ID:
            cacheId = in.readInt();
            pageId = in.readLong();
            rmvId = in.readLong();
            res = new FixRemoveId(cacheId, pageId, rmvId);
            break;
        case PAGES_LIST_SET_NEXT:
            cacheId = in.readInt();
            pageId = in.readLong();
            long nextPageId = in.readLong();
            res = new PagesListSetNextRecord(cacheId, pageId, nextPageId);
            break;
        case PAGES_LIST_SET_PREVIOUS:
            cacheId = in.readInt();
            pageId = in.readLong();
            long prevPageId = in.readLong();
            res = new PagesListSetPreviousRecord(cacheId, pageId, prevPageId);
            break;
        case PAGES_LIST_INIT_NEW_PAGE:
            cacheId = in.readInt();
            pageId = in.readLong();
            ioType = in.readInt();
            ioVer = in.readInt();
            newPageId = in.readLong();
            prevPageId = in.readLong();
            long addDataPageId = in.readLong();
            res = new PagesListInitNewPageRecord(cacheId, pageId, ioType, ioVer, newPageId, prevPageId, addDataPageId, log);
            break;
        case PAGES_LIST_ADD_PAGE:
            cacheId = in.readInt();
            pageId = in.readLong();
            long dataPageId = in.readLong();
            res = new PagesListAddPageRecord(cacheId, pageId, dataPageId);
            break;
        case PAGES_LIST_REMOVE_PAGE:
            cacheId = in.readInt();
            pageId = in.readLong();
            long rmvdPageId = in.readLong();
            res = new PagesListRemovePageRecord(cacheId, pageId, rmvdPageId);
            break;
        case TRACKING_PAGE_DELTA:
            cacheId = in.readInt();
            pageId = in.readLong();
            long pageIdToMark = in.readLong();
            long nextSnapshotId0 = in.readLong();
            long lastSuccessfulSnapshotId0 = in.readLong();
            res = new TrackingPageDeltaRecord(cacheId, pageId, pageIdToMark, nextSnapshotId0, lastSuccessfulSnapshotId0);
            break;
        case META_PAGE_UPDATE_NEXT_SNAPSHOT_ID:
            cacheId = in.readInt();
            pageId = in.readLong();
            long nextSnapshotId = in.readLong();
            res = new MetaPageUpdateNextSnapshotId(cacheId, pageId, nextSnapshotId);
            break;
        case META_PAGE_UPDATE_LAST_SUCCESSFUL_FULL_SNAPSHOT_ID:
            cacheId = in.readInt();
            pageId = in.readLong();
            long lastSuccessfulFullSnapshotId = in.readLong();
            res = new MetaPageUpdateLastSuccessfulFullSnapshotId(cacheId, pageId, lastSuccessfulFullSnapshotId);
            break;
        case META_PAGE_UPDATE_LAST_SUCCESSFUL_SNAPSHOT_ID:
            cacheId = in.readInt();
            pageId = in.readLong();
            long lastSuccessfulSnapshotId = in.readLong();
            long lastSuccessfulSnapshotTag = in.readLong();
            res = new MetaPageUpdateLastSuccessfulSnapshotId(cacheId, pageId, lastSuccessfulSnapshotId, lastSuccessfulSnapshotTag);
            break;
        case META_PAGE_UPDATE_LAST_ALLOCATED_INDEX:
            cacheId = in.readInt();
            pageId = in.readLong();
            int lastAllocatedIdx = in.readInt();
            res = new MetaPageUpdateLastAllocatedIndex(cacheId, pageId, lastAllocatedIdx);
            break;
        case PART_META_UPDATE_STATE:
            cacheId = in.readInt();
            partId = in.readInt();
            byte state = in.readByte();
            long updateCntr = in.readLong();
            GridDhtPartitionState partState = GridDhtPartitionState.fromOrdinal(state);
            res = new PartitionMetaStateRecord(cacheId, partId, partState, updateCntr);
            break;
        case PAGE_LIST_META_RESET_COUNT_RECORD:
            cacheId = in.readInt();
            pageId = in.readLong();
            res = new PageListMetaResetCountRecord(cacheId, pageId);
            break;
        case ROTATED_ID_PART_RECORD:
            cacheId = in.readInt();
            pageId = in.readLong();
            byte rotatedIdPart = in.readByte();
            res = new RotatedIdPartRecord(cacheId, pageId, rotatedIdPart);
            break;
        case SWITCH_SEGMENT_RECORD:
            throw new EOFException("END OF SEGMENT");
        case TX_RECORD:
            res = txRecordSerializer.readTx(in);
            break;
        case MASTER_KEY_CHANGE_RECORD:
        case MASTER_KEY_CHANGE_RECORD_V2:
            int keyNameLen = in.readInt();
            byte[] keyNameBytes = new byte[keyNameLen];
            in.readFully(keyNameBytes);
            String masterKeyName = new String(keyNameBytes);
            int keysCnt = in.readInt();
            List<T2<Integer, GroupKeyEncrypted>> grpKeys = new ArrayList<>(keysCnt);
            boolean readKeyId = type == MASTER_KEY_CHANGE_RECORD_V2;
            for (int i = 0; i < keysCnt; i++) {
                int grpId = in.readInt();
                int keyId = readKeyId ? in.readByte() & 0xff : 0;
                int grpKeySize = in.readInt();
                byte[] grpKey = new byte[grpKeySize];
                in.readFully(grpKey);
                grpKeys.add(new T2<>(grpId, new GroupKeyEncrypted(keyId, grpKey)));
            }
            res = new MasterKeyChangeRecordV2(masterKeyName, grpKeys);
            break;
        case REENCRYPTION_START_RECORD:
            int grpsCnt = in.readInt();
            Map<Integer, Byte> map = U.newHashMap(grpsCnt);
            for (int i = 0; i < grpsCnt; i++) {
                int grpId = in.readInt();
                byte keyId = in.readByte();
                map.put(grpId, keyId);
            }
            res = new ReencryptionStartRecord(map);
            break;
        case INDEX_ROOT_PAGE_RENAME_RECORD:
            res = new IndexRenameRootPageRecord(in);
            break;
        case PARTITION_CLEARING_START_RECORD:
            int partId0 = in.readInt();
            int grpId = in.readInt();
            long clearVer = in.readLong();
            res = new PartitionClearingStartRecord(partId0, grpId, clearVer);
            break;
        default:
            throw new UnsupportedOperationException("Type: " + type);
    }
    return res;
}

Example 8 with GridDhtPartitionState

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

the class GridPartitionStateMap method entrySet.

/**
 * {@inheritDoc}
 */
@Override
public Set<Entry<Integer, GridDhtPartitionState>> entrySet() {
    return new AbstractSet<Entry<Integer, GridDhtPartitionState>>() {

        @Override
        public Iterator<Entry<Integer, GridDhtPartitionState>> iterator() {
            return new Iterator<Entry<Integer, GridDhtPartitionState>>() {

                /**
                 * Current {@link GridPartitionStateMap#states} index.
                 */
                private int idx;

                /**
                 * Current key value.
                 */
                private int cur;

                @Override
                public boolean hasNext() {
                    idx = states.nextSetBit(idx);
                    return idx != -1;
                }

                @Override
                public Entry<Integer, GridDhtPartitionState> next() {
                    if (!hasNext())
                        throw new NoSuchElementException();
                    cur = idx / BITS;
                    int bitN = idx % BITS;
                    // Get state value from BitSet like in GridPartitionStateMap#state, but don't process known zero bits.
                    int st = 1 << bitN;
                    // Accumulating values of remaining bits
                    for (int i = 1; i < BITS - bitN; i++) st |= (states.get(idx + i) ? 1 : 0) << i + bitN;
                    final int ordinal = st - 1;
                    idx += (BITS - bitN);
                    return new Entry<Integer, GridDhtPartitionState>() {

                        int p = cur;

                        @Override
                        public Integer getKey() {
                            return p;
                        }

                        @Override
                        public GridDhtPartitionState getValue() {
                            return GridDhtPartitionState.fromOrdinal(ordinal);
                        }

                        @Override
                        public GridDhtPartitionState setValue(GridDhtPartitionState val) {
                            return setState(p, val);
                        }
                    };
                }

                @Override
                public void remove() {
                    setState(cur, null);
                }
            };
        }

        @Override
        public int size() {
            return GridPartitionStateMap.this.size();
        }
    };
}

Example 9 with GridDhtPartitionState

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

the class SnapshotPartitionsVerifyHandler method invoke.

/**
 * {@inheritDoc}
 */
@Override
public Map<PartitionKeyV2, PartitionHashRecordV2> invoke(SnapshotHandlerContext opCtx) throws IgniteCheckedException {
    SnapshotMetadata meta = opCtx.metadata();
    Set<Integer> grps = F.isEmpty(opCtx.groups()) ? new HashSet<>(meta.partitions().keySet()) : opCtx.groups().stream().map(CU::cacheId).collect(Collectors.toSet());
    Set<File> partFiles = new HashSet<>();
    IgniteSnapshotManager snpMgr = cctx.snapshotMgr();
    for (File dir : snpMgr.snapshotCacheDirectories(meta.snapshotName(), meta.folderName())) {
        int grpId = CU.cacheId(cacheGroupName(dir));
        if (!grps.remove(grpId))
            continue;
        Set<Integer> parts = meta.partitions().get(grpId) == null ? Collections.emptySet() : new HashSet<>(meta.partitions().get(grpId));
        for (File part : cachePartitionFiles(dir)) {
            int partId = partId(part.getName());
            if (!parts.remove(partId))
                continue;
            partFiles.add(part);
        }
        if (!parts.isEmpty()) {
            throw new IgniteException("Snapshot data doesn't contain required cache group partition " + "[grpId=" + grpId + ", snpName=" + meta.snapshotName() + ", consId=" + meta.consistentId() + ", missed=" + parts + ", meta=" + meta + ']');
        }
    }
    if (!grps.isEmpty()) {
        throw new IgniteException("Snapshot data doesn't contain required cache groups " + "[grps=" + grps + ", snpName=" + meta.snapshotName() + ", consId=" + meta.consistentId() + ", meta=" + meta + ']');
    }
    Map<PartitionKeyV2, PartitionHashRecordV2> res = new ConcurrentHashMap<>();
    ThreadLocal<ByteBuffer> buff = ThreadLocal.withInitial(() -> ByteBuffer.allocateDirect(meta.pageSize()).order(ByteOrder.nativeOrder()));
    GridKernalContext snpCtx = snpMgr.createStandaloneKernalContext(meta.snapshotName(), meta.folderName());
    for (GridComponent comp : snpCtx) comp.start();
    try {
        U.doInParallel(snpMgr.snapshotExecutorService(), partFiles, part -> {
            String grpName = cacheGroupName(part.getParentFile());
            int grpId = CU.cacheId(grpName);
            int partId = partId(part.getName());
            FilePageStoreManager storeMgr = (FilePageStoreManager) cctx.pageStore();
            try (FilePageStore pageStore = (FilePageStore) storeMgr.getPageStoreFactory(grpId, false).createPageStore(getTypeByPartId(partId), part::toPath, val -> {
            })) {
                if (partId == INDEX_PARTITION) {
                    checkPartitionsPageCrcSum(() -> pageStore, INDEX_PARTITION, FLAG_IDX);
                    return null;
                }
                if (grpId == MetaStorage.METASTORAGE_CACHE_ID) {
                    checkPartitionsPageCrcSum(() -> pageStore, partId, FLAG_DATA);
                    return null;
                }
                ByteBuffer pageBuff = buff.get();
                pageBuff.clear();
                pageStore.read(0, pageBuff, true);
                long pageAddr = GridUnsafe.bufferAddress(pageBuff);
                PagePartitionMetaIO io = PageIO.getPageIO(pageBuff);
                GridDhtPartitionState partState = fromOrdinal(io.getPartitionState(pageAddr));
                if (partState != OWNING) {
                    throw new IgniteCheckedException("Snapshot partitions must be in the OWNING " + "state only: " + partState);
                }
                long updateCntr = io.getUpdateCounter(pageAddr);
                long size = io.getSize(pageAddr);
                if (log.isDebugEnabled()) {
                    log.debug("Partition [grpId=" + grpId + ", id=" + partId + ", counter=" + updateCntr + ", size=" + size + "]");
                }
                // Snapshot partitions must always be in OWNING state.
                // There is no `primary` partitions for snapshot.
                PartitionKeyV2 key = new PartitionKeyV2(grpId, partId, grpName);
                PartitionHashRecordV2 hash = calculatePartitionHash(key, updateCntr, meta.consistentId(), GridDhtPartitionState.OWNING, false, size, snpMgr.partitionRowIterator(snpCtx, grpName, partId, pageStore));
                assert hash != null : "OWNING must have hash: " + key;
                res.put(key, hash);
            } catch (IOException e) {
                throw new IgniteCheckedException(e);
            }
            return null;
        });
    } catch (Throwable t) {
        log.error("Error executing handler: ", t);
        throw t;
    } finally {
        for (GridComponent comp : snpCtx) comp.stop(true);
    }
    return res;
}

Example 10 with GridDhtPartitionState

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

the class PartitionReservationManager method reservePartitions.

/**
 * @param cacheIds Cache IDs.
 * @param reqTopVer Topology version from request.
 * @param explicitParts Explicit partitions list.
 * @param nodeId Node ID.
 * @param reqId Request ID.
 * @return String which is null in case of success or with causeMessage if failed
 * @throws IgniteCheckedException If failed.
 */
public PartitionReservation reservePartitions(@Nullable List<Integer> cacheIds, AffinityTopologyVersion reqTopVer, final int[] explicitParts, UUID nodeId, long reqId) throws IgniteCheckedException {
    try (TraceSurroundings ignored = MTC.support(ctx.tracing().create(SQL_PARTITIONS_RESERVE, MTC.span()))) {
        assert reqTopVer != null;
        AffinityTopologyVersion topVer = ctx.cache().context().exchange().lastAffinityChangedTopologyVersion(reqTopVer);
        if (F.isEmpty(cacheIds))
            return new PartitionReservation(Collections.emptyList());
        Collection<Integer> partIds;
        if (explicitParts == null)
            partIds = null;
        else if (explicitParts.length == 0)
            partIds = Collections.emptyList();
        else {
            partIds = new ArrayList<>(explicitParts.length);
            for (int explicitPart : explicitParts) partIds.add(explicitPart);
        }
        List<GridReservable> reserved = new ArrayList<>();
        for (int i = 0; i < cacheIds.size(); i++) {
            GridCacheContext<?, ?> cctx = ctx.cache().context().cacheContext(cacheIds.get(i));
            // Cache was not found, probably was not deployed yet.
            if (cctx == null) {
                return new PartitionReservation(reserved, String.format("Failed to reserve partitions for query (cache is not " + "found on local node) [localNodeId=%s, rmtNodeId=%s, reqId=%s, affTopVer=%s, cacheId=%s]", ctx.localNodeId(), nodeId, reqId, topVer, cacheIds.get(i)));
            }
            if (cctx.isLocal() || !cctx.rebalanceEnabled())
                continue;
            // For replicated cache topology version does not make sense.
            final PartitionReservationKey grpKey = new PartitionReservationKey(cctx.name(), cctx.isReplicated() ? null : topVer);
            GridReservable r = reservations.get(grpKey);
            if (explicitParts == null && r != null) {
                // Try to reserve group partition if any and no explicits.
                if (r != REPLICATED_RESERVABLE) {
                    if (!r.reserve())
                        return new PartitionReservation(reserved, String.format("Failed to reserve partitions for query (group " + "reservation failed) [localNodeId=%s, rmtNodeId=%s, reqId=%s, affTopVer=%s, cacheId=%s, " + "cacheName=%s]", ctx.localNodeId(), nodeId, reqId, topVer, cacheIds.get(i), cctx.name()));
                    reserved.add(r);
                    MTC.span().addLog(() -> "Cache partitions were reserved " + r);
                }
            } else {
                // Try to reserve partitions one by one.
                int partsCnt = cctx.affinity().partitions();
                if (cctx.isReplicated()) {
                    // Check all the partitions are in owning state for replicated cache.
                    if (r == null) {
                        // Check only once.
                        for (int p = 0; p < partsCnt; p++) {
                            GridDhtLocalPartition part = partition(cctx, p);
                            // We don't need to reserve partitions because they will not be evicted in replicated caches.
                            GridDhtPartitionState partState = part != null ? part.state() : null;
                            if (partState != OWNING)
                                return new PartitionReservation(reserved, String.format("Failed to reserve partitions for " + "query (partition of REPLICATED cache is not in OWNING state) [" + "localNodeId=%s, rmtNodeId=%s, reqId=%s, affTopVer=%s, cacheId=%s, " + "cacheName=%s, part=%s, partFound=%s, partState=%s]", ctx.localNodeId(), nodeId, reqId, topVer, cacheIds.get(i), cctx.name(), p, (part != null), partState));
                        }
                        // Mark that we checked this replicated cache.
                        reservations.putIfAbsent(grpKey, REPLICATED_RESERVABLE);
                        MTC.span().addLog(() -> "Cache partitions were reserved [cache=" + cctx.name() + ", partitions=[0.." + partsCnt + ']');
                    }
                } else {
                    // Reserve primary partitions for partitioned cache (if no explicit given).
                    if (explicitParts == null)
                        partIds = cctx.affinity().primaryPartitions(ctx.localNodeId(), topVer);
                    int reservedCnt = 0;
                    for (int partId : partIds) {
                        GridDhtLocalPartition part = partition(cctx, partId);
                        GridDhtPartitionState partState = part != null ? part.state() : null;
                        if (partState != OWNING) {
                            if (partState == LOST)
                                failQueryOnLostData(cctx, part);
                            else {
                                return new PartitionReservation(reserved, String.format("Failed to reserve partitions " + "for query (partition of PARTITIONED cache is not found or not in OWNING " + "state) [localNodeId=%s, rmtNodeId=%s, reqId=%s, affTopVer=%s, cacheId=%s, " + "cacheName=%s, part=%s, partFound=%s, partState=%s]", ctx.localNodeId(), nodeId, reqId, topVer, cacheIds.get(i), cctx.name(), partId, (part != null), partState));
                            }
                        }
                        if (!part.reserve()) {
                            return new PartitionReservation(reserved, String.format("Failed to reserve partitions for query " + "(partition of PARTITIONED cache cannot be reserved) [" + "localNodeId=%s, rmtNodeId=%s, reqId=%s, affTopVer=%s, cacheId=%s, " + "cacheName=%s, part=%s, partFound=%s, partState=%s]", ctx.localNodeId(), nodeId, reqId, topVer, cacheIds.get(i), cctx.name(), partId, true, partState));
                        }
                        reserved.add(part);
                        reservedCnt++;
                        // Double check that we are still in owning state and partition contents are not cleared.
                        partState = part.state();
                        if (partState != OWNING) {
                            if (partState == LOST)
                                failQueryOnLostData(cctx, part);
                            else {
                                return new PartitionReservation(reserved, String.format("Failed to reserve partitions for " + "query (partition of PARTITIONED cache is not in OWNING state after " + "reservation) [localNodeId=%s, rmtNodeId=%s, reqId=%s, affTopVer=%s, " + "cacheId=%s, cacheName=%s, part=%s, partState=%s]", ctx.localNodeId(), nodeId, reqId, topVer, cacheIds.get(i), cctx.name(), partId, partState));
                            }
                        }
                    }
                    final Collection<Integer> finalPartIds = partIds;
                    MTC.span().addLog(() -> "Cache partitions were reserved [cache=" + cctx.name() + ", partitions=" + finalPartIds + ", topology=" + topVer + ']');
                    if (explicitParts == null && reservedCnt > 0) {
                        // We reserved all the primary partitions for cache, attempt to add group reservation.
                        GridDhtPartitionsReservation grp = new GridDhtPartitionsReservation(topVer, cctx, "SQL");
                        if (grp.register(reserved.subList(reserved.size() - reservedCnt, reserved.size()))) {
                            if (reservations.putIfAbsent(grpKey, grp) != null)
                                throw new IllegalStateException("Reservation already exists.");
                            grp.onPublish(new CI1<GridDhtPartitionsReservation>() {

                                @Override
                                public void apply(GridDhtPartitionsReservation r) {
                                    reservations.remove(grpKey, r);
                                }
                            });
                        }
                    }
                }
            }
        }
        return new PartitionReservation(reserved);
    }
}