/**
* File Sink Operator encountered.
*
* @param nd the file sink operator encountered
* @param opProcCtx context
*/
public Object process(
Node nd, Stack<Node> stack, NodeProcessorCtx opProcCtx, Object... nodeOutputs)
throws SemanticException {
GenMRProcContext ctx = (GenMRProcContext) opProcCtx;
ParseContext parseCtx = ctx.getParseCtx();
boolean chDir = false;
Task<? extends Serializable> currTask = ctx.getCurrTask();
FileSinkOperator fsOp = (FileSinkOperator) nd;
boolean isInsertTable = // is INSERT OVERWRITE TABLE
fsOp.getConf().getTableInfo().getTableName() != null
&& parseCtx.getQB().getParseInfo().isInsertToTable();
HiveConf hconf = parseCtx.getConf();
// Mark this task as a final map reduce task (ignoring the optional merge task)
((MapredWork) currTask.getWork()).setFinalMapRed(true);
// If this file sink desc has been processed due to a linked file sink desc,
// use that task
Map<FileSinkDesc, Task<? extends Serializable>> fileSinkDescs = ctx.getLinkedFileDescTasks();
if (fileSinkDescs != null) {
Task<? extends Serializable> childTask = fileSinkDescs.get(fsOp.getConf());
processLinkedFileDesc(ctx, childTask);
return null;
}
// Has the user enabled merging of files for map-only jobs or for all jobs
if ((ctx.getMvTask() != null) && (!ctx.getMvTask().isEmpty())) {
List<Task<MoveWork>> mvTasks = ctx.getMvTask();
// In case of unions or map-joins, it is possible that the file has
// already been seen.
// So, no need to attempt to merge the files again.
if ((ctx.getSeenFileSinkOps() == null) || (!ctx.getSeenFileSinkOps().contains(nd))) {
// no need of merging if the move is to a local file system
MoveTask mvTask = (MoveTask) findMoveTask(mvTasks, fsOp);
if (isInsertTable && hconf.getBoolVar(ConfVars.HIVESTATSAUTOGATHER)) {
addStatsTask(fsOp, mvTask, currTask, parseCtx.getConf());
}
if ((mvTask != null) && !mvTask.isLocal() && fsOp.getConf().canBeMerged()) {
if (fsOp.getConf().isLinkedFileSink()) {
// If the user has HIVEMERGEMAPREDFILES set to false, the idea was the
// number of reducers are few, so the number of files anyway are small.
// However, with this optimization, we are increasing the number of files
// possibly by a big margin. So, merge aggresively.
if (hconf.getBoolVar(ConfVars.HIVEMERGEMAPFILES)
|| hconf.getBoolVar(ConfVars.HIVEMERGEMAPREDFILES)) {
chDir = true;
}
} else {
// There are separate configuration parameters to control whether to
// merge for a map-only job
// or for a map-reduce job
MapredWork currWork = (MapredWork) currTask.getWork();
boolean mergeMapOnly =
hconf.getBoolVar(ConfVars.HIVEMERGEMAPFILES) && currWork.getReducer() == null;
boolean mergeMapRed =
hconf.getBoolVar(ConfVars.HIVEMERGEMAPREDFILES) && currWork.getReducer() != null;
if (mergeMapOnly || mergeMapRed) {
chDir = true;
}
}
}
}
}
String finalName = processFS(fsOp, stack, opProcCtx, chDir);
if (chDir) {
// Merge the files in the destination table/partitions by creating Map-only merge job
// If underlying data is RCFile or OrcFile a BlockMerge task would be created.
LOG.info("using CombineHiveInputformat for the merge job");
createMRWorkForMergingFiles(fsOp, ctx, finalName);
}
FileSinkDesc fileSinkDesc = fsOp.getConf();
if (fileSinkDesc.isLinkedFileSink()) {
Map<FileSinkDesc, Task<? extends Serializable>> linkedFileDescTasks =
ctx.getLinkedFileDescTasks();
if (linkedFileDescTasks == null) {
linkedFileDescTasks = new HashMap<FileSinkDesc, Task<? extends Serializable>>();
ctx.setLinkedFileDescTasks(linkedFileDescTasks);
}
// The child tasks may be null in case of a select
if ((currTask.getChildTasks() != null) && (currTask.getChildTasks().size() == 1)) {
for (FileSinkDesc fileDesc : fileSinkDesc.getLinkedFileSinkDesc()) {
linkedFileDescTasks.put(fileDesc, currTask.getChildTasks().get(0));
}
}
}
return null;
}
private void outputPlan(Task<? extends Serializable> task) {
if (task == null) return;
out.printf("Stage: \n", task.getId());
// real output
Serializable work = task.getWork();
if (work == null) return;
if (work instanceof FetchWork) {
out.println("Fetch");
output(((FetchWork) work).getSource());
} else if (work instanceof MapredLocalWork) {
out.println("MapredLocalWork");
// fetch
try {
out.println("Fetch Part");
Collection<FetchWork> fetchWorkCollect =
((MapredLocalWork) work).getAliasToFetchWork().values();
for (FetchWork f : fetchWorkCollect) {
output(f.getSource());
}
} catch (Exception e) {
out.println("Exception 1");
}
// others
try {
out.println("Other Parts");
Collection<Operator<? extends OperatorDesc>> collect =
((MapredLocalWork) work).getAliasToWork().values();
for (Operator<? extends OperatorDesc> c : collect) {
output(c);
}
} catch (Exception e) {
out.println("Exception 2");
}
} else if (work instanceof MapredWork) {
out.println("MapredWork");
try {
Collection<Operator<? extends OperatorDesc>> collect =
((MapredWork) work).getAllOperators();
for (Operator<? extends OperatorDesc> c : collect) {
// out.println(1);
output(c);
break; // first operator will give out all info s
}
} catch (Exception e) {
out.println("Exception 3");
}
} else {
output(work);
}
// -------other cases--------------------
if (task instanceof ConditionalTask && ((ConditionalTask) task).getListTasks() != null) {
for (Task<? extends Serializable> con : ((ConditionalTask) task).getListTasks()) {
outputPlan(con);
}
}
if (task.getChildTasks() != null) {
for (Task<? extends Serializable> child : task.getChildTasks()) {
outputPlan(child);
}
}
}