@Override
public Value[] next() {
if (resultCursor == null) {
Index idx;
if (distinct || sort != null) {
idx = index;
} else {
idx = table.getScanIndex(session);
}
if (session.getDatabase().getMvStore() != null) {
// sometimes the transaction is already committed,
// in which case we can't use the session
if (idx.getRowCount(session) == 0 && rowCount > 0) {
// this means querying is not transactional
resultCursor = idx.find((Session) null, null, null);
} else {
// the transaction is still open
resultCursor = idx.find(session, null, null);
}
} else {
resultCursor = idx.find(session, null, null);
}
}
if (!resultCursor.next()) {
return null;
}
Row row = resultCursor.get();
return row.getValueList();
}
@Override
public int removeRow(Value[] values) {
Row row = convertToRow(values);
Cursor cursor = find(row);
if (cursor != null) {
row = cursor.get();
table.removeRow(session, row);
rowCount--;
}
return rowCount;
}
private void queryDistinct(ResultTarget result, long limitRows) {
// limitRows must be long, otherwise we get an int overflow
// if limitRows is at or near Integer.MAX_VALUE
// limitRows is never 0 here
if (limitRows > 0 && offsetExpr != null) {
int offset = offsetExpr.getValue(session).getInt();
if (offset > 0) {
limitRows += offset;
}
}
int rowNumber = 0;
setCurrentRowNumber(0);
Index index = topTableFilter.getIndex();
SearchRow first = null;
int columnIndex = index.getColumns()[0].getColumnId();
while (true) {
setCurrentRowNumber(rowNumber + 1);
Cursor cursor = index.findNext(session, first, null);
if (!cursor.next()) {
break;
}
SearchRow found = cursor.getSearchRow();
Value value = found.getValue(columnIndex);
if (first == null) {
first = topTableFilter.getTable().getTemplateSimpleRow(true);
}
first.setValue(columnIndex, value);
Value[] row = {value};
result.addRow(row);
rowNumber++;
if ((sort == null || sortUsingIndex) && limitRows > 0 && rowNumber >= limitRows) {
break;
}
if (sampleSize > 0 && rowNumber >= sampleSize) {
break;
}
}
}
private Cursor find(Row row) {
if (index == null) {
// for the case "in(select ...)", the query might
// use an optimization and not create the index
// up front
createIndex();
}
Cursor cursor = index.find(session, row, row);
while (cursor.next()) {
SearchRow found = cursor.getSearchRow();
boolean ok = true;
Database db = session.getDatabase();
for (int i = 0; i < row.getColumnCount(); i++) {
if (!db.areEqual(row.getValue(i), found.getValue(i))) {
ok = false;
break;
}
}
if (ok) {
return cursor;
}
}
return null;
}
/**
* @param cctx Cache context.
* @param qry Query.
* @param keepPortable Keep portable.
* @return Cursor.
*/
public Iterator<List<?>> query(
GridCacheContext<?, ?> cctx, GridCacheTwoStepQuery qry, boolean keepPortable) {
for (int attempt = 0; ; attempt++) {
if (attempt != 0) {
try {
Thread.sleep(attempt * 10); // Wait for exchange.
} catch (InterruptedException e) {
Thread.currentThread().interrupt();
throw new CacheException("Query was interrupted.", e);
}
}
long qryReqId = reqIdGen.incrementAndGet();
QueryRun r = new QueryRun();
r.pageSize = qry.pageSize() <= 0 ? GridCacheTwoStepQuery.DFLT_PAGE_SIZE : qry.pageSize();
r.idxs = new ArrayList<>(qry.mapQueries().size());
String space = cctx.name();
r.conn = (JdbcConnection) h2.connectionForSpace(space);
AffinityTopologyVersion topVer = h2.readyTopologyVersion();
List<String> extraSpaces = extraSpaces(space, qry.spaces());
Collection<ClusterNode> nodes;
// Explicit partition mapping for unstable topology.
Map<ClusterNode, IntArray> partsMap = null;
if (isPreloadingActive(cctx, extraSpaces)) {
if (cctx.isReplicated()) nodes = replicatedUnstableDataNodes(cctx, extraSpaces);
else {
partsMap = partitionedUnstableDataNodes(cctx, extraSpaces);
nodes = partsMap == null ? null : partsMap.keySet();
}
} else nodes = stableDataNodes(topVer, cctx, extraSpaces);
if (nodes == null) continue; // Retry.
assert !nodes.isEmpty();
if (cctx.isReplicated() || qry.explain()) {
assert qry.explain() || !nodes.contains(ctx.discovery().localNode())
: "We must be on a client node.";
// Select random data node to run query on a replicated data or get EXPLAIN PLAN from a
// single node.
nodes = Collections.singleton(F.rand(nodes));
}
int tblIdx = 0;
final boolean skipMergeTbl = !qry.explain() && qry.skipMergeTable();
for (GridCacheSqlQuery mapQry : qry.mapQueries()) {
GridMergeIndex idx;
if (!skipMergeTbl) {
GridMergeTable tbl;
try {
tbl = createMergeTable(r.conn, mapQry, qry.explain());
} catch (IgniteCheckedException e) {
throw new IgniteException(e);
}
idx = tbl.getScanIndex(null);
fakeTable(r.conn, tblIdx++).setInnerTable(tbl);
} else idx = GridMergeIndexUnsorted.createDummy();
for (ClusterNode node : nodes) idx.addSource(node.id());
r.idxs.add(idx);
}
r.latch = new CountDownLatch(r.idxs.size() * nodes.size());
runs.put(qryReqId, r);
try {
if (ctx.clientDisconnected()) {
throw new CacheException(
"Query was cancelled, client node disconnected.",
new IgniteClientDisconnectedException(
ctx.cluster().clientReconnectFuture(), "Client node disconnected."));
}
Collection<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(), mapQry.parameters()));
}
if (nodes.size() != 1 || !F.first(nodes).isLocal()) { // Marshall params for remotes.
Marshaller m = ctx.config().getMarshaller();
for (GridCacheSqlQuery mapQry : mapQrys) mapQry.marshallParams(m);
}
boolean retry = false;
if (send(
nodes,
new GridQueryRequest(qryReqId, r.pageSize, space, mapQrys, topVer, extraSpaces, null),
partsMap)) {
awaitAllReplies(r, nodes);
Object state = r.state.get();
if (state != null) {
if (state instanceof CacheException) {
CacheException err = (CacheException) state;
if (err.getCause() instanceof IgniteClientDisconnectedException) throw err;
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 (qry.explain()) return explainPlan(r.conn, space, qry);
if (skipMergeTbl) {
List<List<?>> res = new ArrayList<>();
assert r.idxs.size() == 1 : r.idxs;
GridMergeIndex idx = r.idxs.get(0);
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 {
GridCacheSqlQuery rdc = qry.reduceQuery();
// Statement caching is prohibited here because we can't guarantee correct merge index
// reuse.
ResultSet res =
h2.executeSqlQueryWithTimer(
space, r.conn, rdc.query(), F.asList(rdc.parameters()), false);
resIter = new Iter(res);
}
}
for (GridMergeIndex idx : r.idxs) {
if (!idx.fetchedAll()) // We have to explicitly cancel queries on remote nodes.
send(nodes, new GridQueryCancelRequest(qryReqId), null);
}
if (retry) {
if (Thread.currentThread().isInterrupted())
throw new IgniteInterruptedCheckedException("Query was interrupted.");
continue;
}
return new GridQueryCacheObjectsIterator(resIter, cctx, keepPortable);
} catch (IgniteCheckedException | RuntimeException e) {
U.closeQuiet(r.conn);
if (e instanceof CacheException) 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 {
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++)
fakeTable(null, i).setInnerTable(null); // Drop all merge tables.
}
}
}
}