/**
* Add the StatsTask as a dependent task of the MoveTask because StatsTask will change the
* Table/Partition metadata. For atomicity, we should not change it before the data is actually
* there done by MoveTask.
*
* @param nd the FileSinkOperator whose results are taken care of by the MoveTask.
* @param mvTask The MoveTask that moves the FileSinkOperator's results.
* @param currTask The MapRedTask that the FileSinkOperator belongs to.
* @param hconf HiveConf
*/
private void addStatsTask(
FileSinkOperator nd, MoveTask mvTask, Task<? extends Serializable> currTask, HiveConf hconf) {
MoveWork mvWork = ((MoveTask) mvTask).getWork();
StatsWork statsWork = null;
if (mvWork.getLoadTableWork() != null) {
statsWork = new StatsWork(mvWork.getLoadTableWork());
} else if (mvWork.getLoadFileWork() != null) {
statsWork = new StatsWork(mvWork.getLoadFileWork());
}
assert statsWork != null : "Error when genereting StatsTask";
statsWork.setStatsReliable(hconf.getBoolVar(ConfVars.HIVE_STATS_RELIABLE));
MapredWork mrWork = (MapredWork) currTask.getWork();
// AggKey in StatsWork is used for stats aggregation while StatsAggPrefix
// in FileSinkDesc is used for stats publishing. They should be consistent.
statsWork.setAggKey(((FileSinkOperator) nd).getConf().getStatsAggPrefix());
Task<? extends Serializable> statsTask = TaskFactory.get(statsWork, hconf);
// mark the MapredWork and FileSinkOperator for gathering stats
nd.getConf().setGatherStats(true);
mrWork.setGatheringStats(true);
nd.getConf().setStatsReliable(hconf.getBoolVar(ConfVars.HIVE_STATS_RELIABLE));
nd.getConf()
.setMaxStatsKeyPrefixLength(hconf.getIntVar(ConfVars.HIVE_STATS_KEY_PREFIX_MAX_LENGTH));
// mrWork.addDestinationTable(nd.getConf().getTableInfo().getTableName());
// subscribe feeds from the MoveTask so that MoveTask can forward the list
// of dynamic partition list to the StatsTask
mvTask.addDependentTask(statsTask);
statsTask.subscribeFeed(mvTask);
}
/**
* Add the StatsTask as a dependent task of the MoveTask because StatsTask will change the
* Table/Partition metadata. For atomicity, we should not change it before the data is actually
* there done by MoveTask.
*
* @param nd the FileSinkOperator whose results are taken care of by the MoveTask.
* @param mvTask The MoveTask that moves the FileSinkOperator's results.
* @param currTask The MapRedTask that the FileSinkOperator belongs to.
* @param hconf HiveConf
*/
private void addStatsTask(
FileSinkOperator nd, MoveTask mvTask, Task<? extends Serializable> currTask, HiveConf hconf) {
MoveWork mvWork = ((MoveTask) mvTask).getWork();
StatsWork statsWork = new StatsWork(mvWork.getLoadTableWork());
MapredWork mrWork = (MapredWork) currTask.getWork();
// AggKey in StatsWork is used for stats aggregation while StatsAggPrefix
// in FileSinkDesc is used for stats publishing. They should be consistent.
statsWork.setAggKey(((FileSinkOperator) nd).getConf().getStatsAggPrefix());
Task<? extends Serializable> statsTask = TaskFactory.get(statsWork, hconf);
// mark the MapredWork and FileSinkOperator for gathering stats
nd.getConf().setGatherStats(true);
mrWork.setGatheringStats(true);
// mrWork.addDestinationTable(nd.getConf().getTableInfo().getTableName());
// subscribe feeds from the MoveTask so that MoveTask can forward the list
// of dynamic partition list to the StatsTask
mvTask.addDependentTask(statsTask);
statsTask.subscribeFeed(mvTask);
}