Examples with PartitionReservation org.apache.ignite.internal.processors.query.h2.twostep.PartitionReservation used on opensource projects

Example 1 with PartitionReservation

use of org.apache.ignite.internal.processors.query.h2.twostep.PartitionReservation in project ignite by apache.

the class GridMapQueryExecutor method onQueryRequest0.

/**
 * @param node Node authored request.
 * @param qryId Query ID.
 * @param reqId Request ID.
 * @param segmentId index segment ID.
 * @param schemaName Schema name.
 * @param qrys Queries to execute.
 * @param cacheIds Caches which will be affected by these queries.
 * @param topVer Topology version.
 * @param partsMap Partitions map for unstable topology.
 * @param parts Explicit partitions for current node.
 * @param pageSize Page size.
 * @param distributedJoins Query distributed join mode.
 * @param enforceJoinOrder Enforce join order H2 flag.
 * @param replicated Replicated only flag.
 * @param timeout Query timeout.
 * @param params Query parameters.
 * @param lazy Streaming flag.
 * @param mvccSnapshot MVCC snapshot.
 * @param dataPageScanEnabled If data page scan is enabled.
 */
private void onQueryRequest0(final ClusterNode node, final long qryId, final long reqId, final int segmentId, final String schemaName, final Collection<GridCacheSqlQuery> qrys, final List<Integer> cacheIds, final AffinityTopologyVersion topVer, final Map<UUID, int[]> partsMap, final int[] parts, final int pageSize, final boolean distributedJoins, final boolean enforceJoinOrder, final boolean replicated, final int timeout, final Object[] params, boolean lazy, @Nullable final MvccSnapshot mvccSnapshot, Boolean dataPageScanEnabled, boolean treatReplicatedAsPartitioned) {
    boolean performanceStatsEnabled = ctx.performanceStatistics().enabled();
    if (performanceStatsEnabled)
        IoStatisticsQueryHelper.startGatheringQueryStatistics();
    // Prepare to run queries.
    GridCacheContext<?, ?> mainCctx = mainCacheContext(cacheIds);
    MapNodeResults nodeRess = resultsForNode(node.id());
    MapQueryResults qryResults = null;
    PartitionReservation reserved = null;
    QueryContext qctx = null;
    // We don't use try with resources on purpose - the catch block must also be executed in the context of this span.
    TraceSurroundings trace = MTC.support(ctx.tracing().create(SQL_QRY_EXEC_REQ, MTC.span()).addTag(SQL_QRY_TEXT, () -> qrys.stream().map(GridCacheSqlQuery::query).collect(Collectors.joining("; "))));
    try {
        if (topVer != null) {
            // Reserve primary for topology version or explicit partitions.
            reserved = h2.partitionReservationManager().reservePartitions(cacheIds, topVer, parts, node.id(), reqId);
            if (reserved.failed()) {
                sendRetry(node, reqId, segmentId, reserved.error());
                return;
            }
        }
        // Prepare query context.
        DistributedJoinContext distributedJoinCtx = null;
        if (distributedJoins && !replicated) {
            distributedJoinCtx = new DistributedJoinContext(topVer, partsMap, node.id(), reqId, segmentId, pageSize);
        }
        qctx = new QueryContext(segmentId, h2.backupFilter(topVer, parts, treatReplicatedAsPartitioned), distributedJoinCtx, mvccSnapshot, reserved, true);
        qryResults = new MapQueryResults(h2, reqId, qrys.size(), mainCctx, lazy, qctx);
        // qctx is set, we have to release reservations inside of it.
        reserved = null;
        if (distributedJoinCtx != null)
            qryCtxRegistry.setShared(node.id(), reqId, qctx);
        if (nodeRess.put(reqId, segmentId, qryResults) != null)
            throw new IllegalStateException();
        if (nodeRess.cancelled(reqId)) {
            qryCtxRegistry.clearShared(node.id(), reqId);
            nodeRess.cancelRequest(reqId);
            throw new QueryCancelledException();
        }
        // Run queries.
        int qryIdx = 0;
        boolean evt = mainCctx != null && mainCctx.events().isRecordable(EVT_CACHE_QUERY_EXECUTED);
        for (GridCacheSqlQuery qry : qrys) {
            H2PooledConnection conn = h2.connections().connection(schemaName);
            H2Utils.setupConnection(conn, qctx, distributedJoins, enforceJoinOrder, lazy);
            MapQueryResult res = new MapQueryResult(h2, mainCctx, node.id(), qry, params, conn, log);
            qryResults.addResult(qryIdx, res);
            try {
                res.lock();
                // Ensure we are on the target node for this replicated query.
                if (qry.node() == null || (segmentId == 0 && qry.node().equals(ctx.localNodeId()))) {
                    String sql = qry.query();
                    Collection<Object> params0 = F.asList(qry.parameters(params));
                    PreparedStatement stmt = conn.prepareStatement(sql, H2StatementCache.queryFlags(distributedJoins, enforceJoinOrder));
                    H2Utils.bindParameters(stmt, params0);
                    MapH2QueryInfo qryInfo = new MapH2QueryInfo(stmt, qry.query(), node.id(), qryId, reqId, segmentId);
                    ResultSet rs = h2.executeSqlQueryWithTimer(stmt, conn, sql, timeout, qryResults.queryCancel(qryIdx), dataPageScanEnabled, qryInfo);
                    if (evt) {
                        ctx.event().record(new CacheQueryExecutedEvent<>(node, "SQL query executed.", EVT_CACHE_QUERY_EXECUTED, CacheQueryType.SQL.name(), mainCctx.name(), null, qry.query(), null, null, params, node.id(), null));
                    }
                    assert rs instanceof JdbcResultSet : rs.getClass();
                    if (qryResults.cancelled()) {
                        rs.close();
                        throw new QueryCancelledException();
                    }
                    res.openResult(rs, qryInfo);
                    final GridQueryNextPageResponse msg = prepareNextPage(nodeRess, node, qryResults, qryIdx, segmentId, pageSize, dataPageScanEnabled);
                    if (msg != null)
                        sendNextPage(node, msg);
                } else {
                    assert !qry.isPartitioned();
                    qryResults.closeResult(qryIdx);
                }
                qryIdx++;
            } finally {
                try {
                    res.unlockTables();
                } finally {
                    res.unlock();
                }
            }
        }
        if (!lazy)
            qryResults.releaseQueryContext();
    } catch (Throwable e) {
        if (qryResults != null) {
            nodeRess.remove(reqId, segmentId, qryResults);
            qryResults.close();
            // If a query is cancelled before execution is started partitions have to be released.
            if (!lazy || !qryResults.isAllClosed())
                qryResults.releaseQueryContext();
        } else
            releaseReservations(qctx);
        if (e instanceof QueryCancelledException)
            sendError(node, reqId, e);
        else {
            SQLException sqlEx = X.cause(e, SQLException.class);
            if (sqlEx != null && sqlEx.getErrorCode() == ErrorCode.STATEMENT_WAS_CANCELED)
                sendQueryCancel(node, reqId);
            else {
                GridH2RetryException retryErr = X.cause(e, GridH2RetryException.class);
                if (retryErr != null) {
                    final String retryCause = String.format("Failed to execute non-collocated query (will retry) [localNodeId=%s, rmtNodeId=%s, reqId=%s, " + "errMsg=%s]", ctx.localNodeId(), node.id(), reqId, retryErr.getMessage());
                    sendRetry(node, reqId, segmentId, retryCause);
                } else {
                    QueryRetryException qryRetryErr = X.cause(e, QueryRetryException.class);
                    if (qryRetryErr != null)
                        sendError(node, reqId, qryRetryErr);
                    else {
                        if (e instanceof Error) {
                            U.error(log, "Failed to execute local query.", e);
                            throw (Error) e;
                        }
                        U.warn(log, "Failed to execute local query.", e);
                        sendError(node, reqId, e);
                    }
                }
            }
        }
    } finally {
        if (reserved != null)
            reserved.release();
        if (trace != null)
            trace.close();
        if (performanceStatsEnabled) {
            IoStatisticsHolder stat = IoStatisticsQueryHelper.finishGatheringQueryStatistics();
            if (stat.logicalReads() > 0 || stat.physicalReads() > 0) {
                ctx.performanceStatistics().queryReads(GridCacheQueryType.SQL_FIELDS, node.id(), reqId, stat.logicalReads(), stat.physicalReads());
            }
        }
    }
}

Example 2 with PartitionReservation

use of org.apache.ignite.internal.processors.query.h2.twostep.PartitionReservation in project ignite by apache.

the class GridMapQueryExecutor method onDmlRequest.

/**
 * @param node Node.
 * @param req DML request.
 */
public void onDmlRequest(final ClusterNode node, final GridH2DmlRequest req) {
    int[] parts = req.queryPartitions();
    List<Integer> cacheIds = req.caches();
    long reqId = req.requestId();
    AffinityTopologyVersion topVer = req.topologyVersion();
    PartitionReservation reserved = null;
    MapNodeResults nodeResults = resultsForNode(node.id());
    // We don't use try with resources on purpose - the catch block must also be executed in the context of this span.
    TraceSurroundings trace = MTC.support(ctx.tracing().create(SpanType.SQL_DML_QRY_EXEC_REQ, MTC.span()).addTag(SQL_QRY_TEXT, req::query));
    try {
        reserved = h2.partitionReservationManager().reservePartitions(cacheIds, topVer, parts, node.id(), reqId);
        if (reserved.failed()) {
            U.error(log, "Failed to reserve partitions for DML request. [localNodeId=" + ctx.localNodeId() + ", nodeId=" + node.id() + ", reqId=" + req.requestId() + ", cacheIds=" + cacheIds + ", topVer=" + topVer + ", parts=" + Arrays.toString(parts) + ']');
            sendUpdateResponse(node, reqId, null, "Failed to reserve partitions for DML request. " + reserved.error());
            return;
        }
        IndexingQueryFilter filter = h2.backupFilter(topVer, parts);
        GridQueryCancel cancel = nodeResults.putUpdate(reqId);
        SqlFieldsQuery fldsQry = new SqlFieldsQuery(req.query());
        if (req.parameters() != null)
            fldsQry.setArgs(req.parameters());
        fldsQry.setEnforceJoinOrder(req.isFlagSet(GridH2QueryRequest.FLAG_ENFORCE_JOIN_ORDER));
        fldsQry.setPageSize(req.pageSize());
        fldsQry.setLocal(true);
        if (req.timeout() > 0 || req.explicitTimeout())
            fldsQry.setTimeout(req.timeout(), TimeUnit.MILLISECONDS);
        boolean local = true;
        final boolean replicated = req.isFlagSet(GridH2QueryRequest.FLAG_REPLICATED);
        if (!replicated && !F.isEmpty(cacheIds) && CU.firstPartitioned(ctx.cache().context(), cacheIds).config().getQueryParallelism() > 1) {
            fldsQry.setDistributedJoins(true);
            local = false;
        }
        UpdateResult updRes = h2.executeUpdateOnDataNode(req.schemaName(), fldsQry, filter, cancel, local);
        GridCacheContext<?, ?> mainCctx = !F.isEmpty(cacheIds) ? ctx.cache().context().cacheContext(cacheIds.get(0)) : null;
        boolean evt = local && mainCctx != null && mainCctx.events().isRecordable(EVT_CACHE_QUERY_EXECUTED);
        if (evt) {
            ctx.event().record(new CacheQueryExecutedEvent<>(node, "SQL query executed.", EVT_CACHE_QUERY_EXECUTED, CacheQueryType.SQL.name(), mainCctx.name(), null, req.query(), null, null, req.parameters(), node.id(), null));
        }
        sendUpdateResponse(node, reqId, updRes, null);
    } catch (Exception e) {
        MTC.span().addTag(ERROR, e::getMessage);
        U.error(log, "Error processing dml request. [localNodeId=" + ctx.localNodeId() + ", nodeId=" + node.id() + ", req=" + req + ']', e);
        sendUpdateResponse(node, reqId, null, e.getMessage());
    } finally {
        if (reserved != null)
            reserved.release();
        nodeResults.removeUpdate(reqId);
        if (trace != null)
            trace.close();
    }
}

Example 3 with PartitionReservation

use of org.apache.ignite.internal.processors.query.h2.twostep.PartitionReservation in project gridgain by gridgain.

the class GridMapQueryExecutor method onQueryRequest0.

/**
 * @param node Node authored request.
 * @param reqId Request ID.
 * @param segmentId index segment ID.
 * @param schemaName Schema name.
 * @param qrys Queries to execute.
 * @param cacheIds Caches which will be affected by these queries.
 * @param topVer Topology version.
 * @param partsMap Partitions map for unstable topology.
 * @param parts Explicit partitions for current node.
 * @param pageSize Page size.
 * @param distributedJoins Query distributed join mode.
 * @param enforceJoinOrder Enforce join order H2 flag.
 * @param replicated Replicated only flag.
 * @param timeout Query timeout.
 * @param params Query parameters.
 * @param lazy Streaming flag.
 * @param mvccSnapshot MVCC snapshot.
 * @param dataPageScanEnabled If data page scan is enabled.
 * @param maxMem Query memory limit.
 * @param runningQryId Running query id.
 */
private void onQueryRequest0(final ClusterNode node, final long reqId, final int segmentId, final String schemaName, final Collection<GridCacheSqlQuery> qrys, final List<Integer> cacheIds, final AffinityTopologyVersion topVer, final Map<UUID, int[]> partsMap, final int[] parts, final int pageSize, final boolean distributedJoins, final boolean enforceJoinOrder, final boolean replicated, final int timeout, final Object[] params, boolean lazy, @Nullable final MvccSnapshot mvccSnapshot, Boolean dataPageScanEnabled, long maxMem, @Nullable Long runningQryId, boolean treatReplicatedAsPartitioned) {
    // Prepare to run queries.
    GridCacheContext<?, ?> mainCctx = mainCacheContext(cacheIds);
    MapNodeResults nodeRess = resultsForNode(node.id());
    MapQueryResults qryResults = null;
    PartitionReservation reserved = null;
    QueryContext qctx = null;
    // We don't use try with resources on purpose - the catch block must also be executed in the context of this span.
    TraceSurroundings trace = MTC.support(ctx.tracing().create(SQL_QRY_EXEC_REQ, MTC.span()).addTag(SQL_QRY_TEXT, () -> qrys.stream().map(GridCacheSqlQuery::query).collect(Collectors.joining("; "))));
    try {
        if (topVer != null) {
            // Reserve primary for topology version or explicit partitions.
            reserved = h2.partitionReservationManager().reservePartitions(cacheIds, topVer, parts, node.id(), reqId);
            if (reserved.failed()) {
                sendRetry(node, reqId, segmentId, reserved.error());
                return;
            }
        }
        // Prepare query context.
        DistributedJoinContext distributedJoinCtx = null;
        if (distributedJoins && !replicated) {
            distributedJoinCtx = new DistributedJoinContext(topVer, partsMap, node.id(), reqId, segmentId, pageSize);
        }
        qctx = new QueryContext(segmentId, h2.backupFilter(topVer, parts, treatReplicatedAsPartitioned), distributedJoinCtx, mvccSnapshot, reserved, true);
        qryResults = new MapQueryResults(h2, reqId, qrys.size(), mainCctx, lazy, qctx);
        // qctx is set, we have to release reservations inside of it.
        reserved = null;
        if (distributedJoinCtx != null)
            qryCtxRegistry.setShared(node.id(), reqId, qctx);
        if (nodeRess.put(reqId, segmentId, qryResults) != null)
            throw new IllegalStateException();
        if (nodeRess.cancelled(reqId)) {
            qryCtxRegistry.clearShared(node.id(), reqId);
            nodeRess.cancelRequest(reqId);
            throw new QueryCancelledException();
        }
        // Run queries.
        int qryIdx = 0;
        boolean evt = mainCctx != null && mainCctx.events().isRecordable(EVT_CACHE_QUERY_EXECUTED);
        for (GridCacheSqlQuery qry : qrys) {
            H2PooledConnection conn = h2.connections().connection(schemaName);
            H2Utils.setupConnection(conn, qctx, distributedJoins, enforceJoinOrder, lazy);
            MapQueryResult res = new MapQueryResult(h2, mainCctx, node.id(), qry, params, conn, log);
            qryResults.addResult(qryIdx, res);
            try {
                res.lock();
                // If we are not the target node for this replicated query, just ignore it.
                if (qry.node() == null || (segmentId == 0 && qry.node().equals(ctx.localNodeId()))) {
                    String sql = qry.query();
                    Collection<Object> params0 = F.asList(qry.parameters(params));
                    PreparedStatement stmt = conn.prepareStatement(sql, H2StatementCache.queryFlags(distributedJoins, enforceJoinOrder));
                    H2Utils.bindParameters(stmt, params0);
                    MapH2QueryInfo qryInfo = new MapH2QueryInfo(stmt, qry.query(), node, reqId, segmentId, runningQryId);
                    ResultSet rs = h2.executeSqlQueryWithTimer(stmt, conn, sql, timeout, qryResults.queryCancel(qryIdx), dataPageScanEnabled, qryInfo, maxMem);
                    if (evt) {
                        ctx.event().record(new CacheQueryExecutedEvent<>(node, "SQL query executed.", EVT_CACHE_QUERY_EXECUTED, CacheQueryType.SQL.name(), mainCctx.name(), null, qry.query(), null, null, params, node.id(), null));
                    }
                    assert rs instanceof JdbcResultSet : rs.getClass();
                    if (qryResults.cancelled()) {
                        rs.close();
                        throw new QueryCancelledException();
                    }
                    res.openResult(rs, qryInfo, ctx.tracing());
                    final GridQueryNextPageResponse msg = prepareNextPage(nodeRess, node, qryResults, qryIdx, segmentId, pageSize, dataPageScanEnabled);
                    if (msg != null)
                        sendNextPage(node, msg);
                } else {
                    assert !qry.isPartitioned();
                    qryResults.closeResult(qryIdx);
                }
                qryIdx++;
            } finally {
                try {
                    res.unlockTables();
                } finally {
                    res.unlock();
                }
            }
        }
        if (!lazy)
            qryResults.releaseQueryContext();
    } catch (Throwable e) {
        if (qryResults != null) {
            nodeRess.remove(reqId, segmentId, qryResults);
            qryResults.close();
            // If a query is cancelled before execution is started partitions have to be released.
            if (!lazy || !qryResults.isAllClosed())
                qryResults.releaseQueryContext();
        } else
            releaseReservations(qctx);
        if (e instanceof QueryCancelledException)
            sendError(node, reqId, e);
        else {
            SQLException sqlEx = X.cause(e, SQLException.class);
            if ((sqlEx != null && sqlEx.getErrorCode() == ErrorCode.STATEMENT_WAS_CANCELED) || (qryResults != null && qryResults.cancelled() && e instanceof QueryMemoryTracker.TrackerWasClosedException))
                sendQueryCancel(node, reqId);
            else {
                GridH2RetryException retryErr = X.cause(e, GridH2RetryException.class);
                if (retryErr != null) {
                    final String retryCause = String.format("Failed to execute non-collocated query (will retry) [localNodeId=%s, rmtNodeId=%s, reqId=%s, " + "errMsg=%s]", ctx.localNodeId(), node.id(), reqId, retryErr.getMessage());
                    sendRetry(node, reqId, segmentId, retryCause);
                } else {
                    QueryRetryException qryRetryErr = X.cause(e, QueryRetryException.class);
                    if (qryRetryErr != null)
                        sendError(node, reqId, qryRetryErr);
                    else {
                        if (e instanceof Error) {
                            U.error(log, "Failed to execute local query.", e);
                            throw (Error) e;
                        }
                        U.warn(log, "Failed to execute local query.", e);
                        sendError(node, reqId, e);
                    }
                }
            }
        }
    } finally {
        if (reserved != null)
            reserved.release();
        if (trace != null)
            trace.close();
    }
}

Example 4 with PartitionReservation

use of org.apache.ignite.internal.processors.query.h2.twostep.PartitionReservation in project gridgain by gridgain.

the class GridMapQueryExecutor method onDmlRequest.

/**
 * @param node Node.
 * @param req DML request.
 */
public void onDmlRequest(final ClusterNode node, final GridH2DmlRequest req) {
    int[] parts = req.queryPartitions();
    List<Integer> cacheIds = req.caches();
    long reqId = req.requestId();
    AffinityTopologyVersion topVer = req.topologyVersion();
    PartitionReservation reserved = null;
    MapNodeResults nodeResults = resultsForNode(node.id());
    // We don't use try with resources on purpose - the catch block must also be executed in the context of this span.
    TraceSurroundings trace = MTC.support(ctx.tracing().create(SpanType.SQL_DML_QRY_EXEC_REQ, MTC.span()).addTag(SQL_QRY_TEXT, req::query));
    try {
        reserved = h2.partitionReservationManager().reservePartitions(cacheIds, topVer, parts, node.id(), reqId);
        if (reserved.failed()) {
            U.error(log, "Failed to reserve partitions for DML request. [localNodeId=" + ctx.localNodeId() + ", nodeId=" + node.id() + ", reqId=" + req.requestId() + ", cacheIds=" + cacheIds + ", topVer=" + topVer + ", parts=" + Arrays.toString(parts) + ']');
            sendUpdateResponse(node, reqId, null, "Failed to reserve partitions for DML request. " + reserved.error());
            return;
        }
        IndexingQueryFilter filter = h2.backupFilter(topVer, parts);
        GridQueryCancel cancel = nodeResults.putUpdate(reqId);
        SqlFieldsQuery fldsQry = new SqlFieldsQuery(req.query());
        if (req.parameters() != null)
            fldsQry.setArgs(req.parameters());
        fldsQry.setEnforceJoinOrder(req.isFlagSet(GridH2QueryRequest.FLAG_ENFORCE_JOIN_ORDER));
        fldsQry.setPageSize(req.pageSize());
        fldsQry.setLocal(true);
        if (req.timeout() > 0 || req.explicitTimeout())
            fldsQry.setTimeout(req.timeout(), TimeUnit.MILLISECONDS);
        boolean local = true;
        final boolean replicated = req.isFlagSet(GridH2QueryRequest.FLAG_REPLICATED);
        if (!replicated && !F.isEmpty(cacheIds) && CU.firstPartitioned(ctx.cache().context(), cacheIds).config().getQueryParallelism() > 1) {
            fldsQry.setDistributedJoins(true);
            local = false;
        }
        UpdateResult updRes = h2.executeUpdateOnDataNode(req.schemaName(), fldsQry, filter, cancel, local);
        GridCacheContext<?, ?> mainCctx = !F.isEmpty(cacheIds) ? ctx.cache().context().cacheContext(cacheIds.get(0)) : null;
        boolean evt = local && mainCctx != null && mainCctx.events().isRecordable(EVT_CACHE_QUERY_EXECUTED);
        if (evt) {
            ctx.event().record(new CacheQueryExecutedEvent<>(node, "SQL query executed.", EVT_CACHE_QUERY_EXECUTED, CacheQueryType.SQL.name(), mainCctx.name(), null, req.query(), null, null, req.parameters(), node.id(), null));
        }
        sendUpdateResponse(node, reqId, updRes, null);
    } catch (Exception e) {
        MTC.span().addTag(ERROR, e::getMessage);
        U.error(log, "Error processing dml request. [localNodeId=" + ctx.localNodeId() + ", nodeId=" + node.id() + ", req=" + req + ']', e);
        sendUpdateResponse(node, reqId, null, e.getMessage());
    } finally {
        if (reserved != null)
            reserved.release();
        nodeResults.removeUpdate(reqId);
        if (trace != null)
            trace.close();
    }
}