use of org.apache.ignite.internal.processors.cache.query.GridCacheSqlQuery in project ignite by apache.
the class GridReduceQueryExecutor method query.
/**
* @param schemaName Schema name.
* @param qry Query.
* @param keepBinary Keep binary.
* @param enforceJoinOrder Enforce join order of tables.
* @param timeoutMillis Timeout in milliseconds.
* @param cancel Query cancel.
* @param params Query parameters.
* @param parts Partitions.
* @return Rows iterator.
*/
public Iterator<List<?>> query(String schemaName, GridCacheTwoStepQuery qry, boolean keepBinary, boolean enforceJoinOrder, int timeoutMillis, GridQueryCancel cancel, Object[] params, final int[] parts) {
if (F.isEmpty(params))
params = EMPTY_PARAMS;
final boolean isReplicatedOnly = qry.isReplicatedOnly();
// Fail if all caches are replicated and explicit partitions are set.
for (int attempt = 0; ; attempt++) {
if (attempt != 0) {
try {
// Wait for exchange.
Thread.sleep(attempt * 10);
} catch (InterruptedException e) {
Thread.currentThread().interrupt();
throw new CacheException("Query was interrupted.", e);
}
}
final long qryReqId = qryIdGen.incrementAndGet();
final ReduceQueryRun r = new ReduceQueryRun(qryReqId, qry.originalSql(), schemaName, h2.connectionForSchema(schemaName), qry.mapQueries().size(), qry.pageSize(), U.currentTimeMillis(), cancel);
AffinityTopologyVersion topVer = h2.readyTopologyVersion();
List<Integer> cacheIds = qry.cacheIds();
Collection<ClusterNode> nodes = null;
// Explicit partition mapping for unstable topology.
Map<ClusterNode, IntArray> partsMap = null;
// Explicit partitions mapping for query.
Map<ClusterNode, IntArray> qryMap = null;
// Partitions are not supported for queries over all replicated caches.
if (parts != null) {
boolean replicatedOnly = true;
for (Integer cacheId : cacheIds) {
if (!cacheContext(cacheId).isReplicated()) {
replicatedOnly = false;
break;
}
}
if (replicatedOnly)
throw new CacheException("Partitions are not supported for replicated caches");
}
if (qry.isLocal())
nodes = singletonList(ctx.discovery().localNode());
else {
if (isPreloadingActive(cacheIds)) {
if (isReplicatedOnly)
nodes = replicatedUnstableDataNodes(cacheIds);
else {
partsMap = partitionedUnstableDataNodes(cacheIds);
if (partsMap != null) {
qryMap = narrowForQuery(partsMap, parts);
nodes = qryMap == null ? null : qryMap.keySet();
}
}
} else {
qryMap = stableDataNodes(isReplicatedOnly, topVer, cacheIds, parts);
if (qryMap != null)
nodes = qryMap.keySet();
}
if (nodes == null)
// Retry.
continue;
assert !nodes.isEmpty();
if (isReplicatedOnly || qry.explain()) {
ClusterNode locNode = ctx.discovery().localNode();
// Always prefer local node if possible.
if (nodes.contains(locNode))
nodes = singletonList(locNode);
else {
// Select random data node to run query on a replicated data or
// get EXPLAIN PLAN from a single node.
nodes = singletonList(F.rand(nodes));
}
}
}
int tblIdx = 0;
final boolean skipMergeTbl = !qry.explain() && qry.skipMergeTable();
final int segmentsPerIndex = qry.explain() || isReplicatedOnly ? 1 : findFirstPartitioned(cacheIds).config().getQueryParallelism();
int replicatedQrysCnt = 0;
for (GridCacheSqlQuery mapQry : qry.mapQueries()) {
GridMergeIndex idx;
if (!skipMergeTbl) {
GridMergeTable tbl;
try {
tbl = createMergeTable(r.connection(), mapQry, qry.explain());
} catch (IgniteCheckedException e) {
throw new IgniteException(e);
}
idx = tbl.getMergeIndex();
fakeTable(r.connection(), tblIdx++).innerTable(tbl);
} else
idx = GridMergeIndexUnsorted.createDummy(ctx);
// If the query has only replicated tables, we have to run it on a single node only.
if (!mapQry.isPartitioned()) {
ClusterNode node = F.rand(nodes);
mapQry.node(node.id());
replicatedQrysCnt++;
// Replicated tables can have only 1 segment.
idx.setSources(singletonList(node), 1);
} else
idx.setSources(nodes, segmentsPerIndex);
idx.setPageSize(r.pageSize());
r.indexes().add(idx);
}
r.latch(new CountDownLatch(isReplicatedOnly ? 1 : (r.indexes().size() - replicatedQrysCnt) * nodes.size() * segmentsPerIndex + replicatedQrysCnt));
runs.put(qryReqId, r);
try {
cancel.checkCancelled();
if (ctx.clientDisconnected()) {
throw new CacheException("Query was cancelled, client node disconnected.", new IgniteClientDisconnectedException(ctx.cluster().clientReconnectFuture(), "Client node disconnected."));
}
List<GridCacheSqlQuery> mapQrys = qry.mapQueries();
if (qry.explain()) {
mapQrys = new ArrayList<>(qry.mapQueries().size());
for (GridCacheSqlQuery mapQry : qry.mapQueries()) mapQrys.add(new GridCacheSqlQuery("EXPLAIN " + mapQry.query()).parameterIndexes(mapQry.parameterIndexes()));
}
final boolean distributedJoins = qry.distributedJoins();
final Collection<ClusterNode> finalNodes = nodes;
cancel.set(new Runnable() {
@Override
public void run() {
send(finalNodes, new GridQueryCancelRequest(qryReqId), null, false);
}
});
boolean retry = false;
// Always enforce join order on map side to have consistent behavior.
int flags = GridH2QueryRequest.FLAG_ENFORCE_JOIN_ORDER;
if (distributedJoins)
flags |= GridH2QueryRequest.FLAG_DISTRIBUTED_JOINS;
if (qry.isLocal())
flags |= GridH2QueryRequest.FLAG_IS_LOCAL;
if (qry.explain())
flags |= GridH2QueryRequest.FLAG_EXPLAIN;
if (isReplicatedOnly)
flags |= GridH2QueryRequest.FLAG_REPLICATED;
GridH2QueryRequest req = new GridH2QueryRequest().requestId(qryReqId).topologyVersion(topVer).pageSize(r.pageSize()).caches(qry.cacheIds()).tables(distributedJoins ? qry.tables() : null).partitions(convert(partsMap)).queries(mapQrys).parameters(params).flags(flags).timeout(timeoutMillis).schemaName(schemaName);
if (send(nodes, req, parts == null ? null : new ExplicitPartitionsSpecializer(qryMap), false)) {
awaitAllReplies(r, nodes, cancel);
Object state = r.state();
if (state != null) {
if (state instanceof CacheException) {
CacheException err = (CacheException) state;
if (err.getCause() instanceof IgniteClientDisconnectedException)
throw err;
if (wasCancelled(err))
// Throw correct exception.
throw new QueryCancelledException();
throw new CacheException("Failed to run map query remotely.", err);
}
if (state instanceof AffinityTopologyVersion) {
retry = true;
// If remote node asks us to retry then we have outdated full partition map.
h2.awaitForReadyTopologyVersion((AffinityTopologyVersion) state);
}
}
} else
// Send failed.
retry = true;
Iterator<List<?>> resIter = null;
if (!retry) {
if (skipMergeTbl) {
List<List<?>> res = new ArrayList<>();
// Simple UNION ALL can have multiple indexes.
for (GridMergeIndex idx : r.indexes()) {
Cursor cur = idx.findInStream(null, null);
while (cur.next()) {
Row row = cur.get();
int cols = row.getColumnCount();
List<Object> resRow = new ArrayList<>(cols);
for (int c = 0; c < cols; c++) resRow.add(row.getValue(c).getObject());
res.add(resRow);
}
}
resIter = res.iterator();
} else {
cancel.checkCancelled();
UUID locNodeId = ctx.localNodeId();
H2Utils.setupConnection(r.connection(), false, enforceJoinOrder);
GridH2QueryContext.set(new GridH2QueryContext(locNodeId, locNodeId, qryReqId, REDUCE).pageSize(r.pageSize()).distributedJoinMode(OFF));
try {
if (qry.explain())
return explainPlan(r.connection(), qry, params);
GridCacheSqlQuery rdc = qry.reduceQuery();
ResultSet res = h2.executeSqlQueryWithTimer(r.connection(), rdc.query(), F.asList(rdc.parameters(params)), // The statement will cache some extra thread local objects.
false, timeoutMillis, cancel);
resIter = new H2FieldsIterator(res);
} finally {
GridH2QueryContext.clearThreadLocal();
}
}
}
if (retry) {
if (Thread.currentThread().isInterrupted())
throw new IgniteInterruptedCheckedException("Query was interrupted.");
continue;
}
return new GridQueryCacheObjectsIterator(resIter, h2.objectContext(), keepBinary);
} catch (IgniteCheckedException | RuntimeException e) {
U.closeQuiet(r.connection());
if (e instanceof CacheException) {
if (wasCancelled((CacheException) e))
throw new CacheException("Failed to run reduce query locally.", new QueryCancelledException());
throw (CacheException) e;
}
Throwable cause = e;
if (e instanceof IgniteCheckedException) {
Throwable disconnectedErr = ((IgniteCheckedException) e).getCause(IgniteClientDisconnectedException.class);
if (disconnectedErr != null)
cause = disconnectedErr;
}
throw new CacheException("Failed to run reduce query locally.", cause);
} finally {
// Make sure any activity related to current attempt is cancelled.
cancelRemoteQueriesIfNeeded(nodes, r, qryReqId, qry.distributedJoins());
if (!runs.remove(qryReqId, r))
U.warn(log, "Query run was already removed: " + qryReqId);
if (!skipMergeTbl) {
for (int i = 0, mapQrys = qry.mapQueries().size(); i < mapQrys; i++) // Drop all merge tables.
fakeTable(null, i).innerTable(null);
}
}
}
}
Aggregations