@Test public void testReducerNumEstimation() throws Exception { // Skip the test for Tez. Tez use a different mechanism. // Equivalent test is in TestTezAutoParallelism Assume.assumeTrue("Skip this test for TEZ", Util.isMapredExecType(cluster.getExecType())); // use the estimation Configuration conf = HBaseConfiguration.create(new Configuration()); HBaseTestingUtility util = new HBaseTestingUtility(conf); int clientPort = util.startMiniZKCluster().getClientPort(); util.startMiniHBaseCluster(1, 1); String query = "a = load '/passwd';" + "b = group a by $0;" + "store b into 'output';"; PigServer ps = new PigServer(cluster.getExecType(), cluster.getProperties()); PhysicalPlan pp = Util.buildPp(ps, query); MROperPlan mrPlan = Util.buildMRPlan(pp, pc); pc.getConf().setProperty("pig.exec.reducers.bytes.per.reducer", "100"); pc.getConf().setProperty("pig.exec.reducers.max", "10"); pc.getConf().setProperty(HConstants.ZOOKEEPER_CLIENT_PORT, Integer.toString(clientPort)); ConfigurationValidator.validatePigProperties(pc.getProperties()); conf = ConfigurationUtil.toConfiguration(pc.getProperties()); JobControlCompiler jcc = new JobControlCompiler(pc, conf); JobControl jc = jcc.compile(mrPlan, "Test"); Job job = jc.getWaitingJobs().get(0); long reducer = Math.min( (long) Math.ceil(new File("test/org/apache/pig/test/data/passwd").length() / 100.0), 10); Util.assertParallelValues(-1, -1, reducer, reducer, job.getJobConf()); // use the PARALLEL key word, it will override the estimated reducer number query = "a = load '/passwd';" + "b = group a by $0 PARALLEL 2;" + "store b into 'output';"; pp = Util.buildPp(ps, query); mrPlan = Util.buildMRPlan(pp, pc); pc.getConf().setProperty("pig.exec.reducers.bytes.per.reducer", "100"); pc.getConf().setProperty("pig.exec.reducers.max", "10"); ConfigurationValidator.validatePigProperties(pc.getProperties()); conf = ConfigurationUtil.toConfiguration(pc.getProperties()); jcc = new JobControlCompiler(pc, conf); jc = jcc.compile(mrPlan, "Test"); job = jc.getWaitingJobs().get(0); Util.assertParallelValues(-1, 2, -1, 2, job.getJobConf()); final byte[] COLUMNFAMILY = Bytes.toBytes("pig"); util.createTable(Bytes.toBytesBinary("test_table"), COLUMNFAMILY); // the estimation won't take effect when it apply to non-dfs or the files doesn't exist, such as // hbase query = "a = load 'hbase://test_table' using org.apache.pig.backend.hadoop.hbase.HBaseStorage('c:f1 c:f2');" + "b = group a by $0 ;" + "store b into 'output';"; pp = Util.buildPp(ps, query); mrPlan = Util.buildMRPlan(pp, pc); pc.getConf().setProperty("pig.exec.reducers.bytes.per.reducer", "100"); pc.getConf().setProperty("pig.exec.reducers.max", "10"); ConfigurationValidator.validatePigProperties(pc.getProperties()); conf = ConfigurationUtil.toConfiguration(pc.getProperties()); jcc = new JobControlCompiler(pc, conf); jc = jcc.compile(mrPlan, "Test"); job = jc.getWaitingJobs().get(0); Util.assertParallelValues(-1, -1, 1, 1, job.getJobConf()); util.deleteTable(Bytes.toBytesBinary("test_table")); // In HBase 0.90.1 and above we can use util.shutdownMiniHBaseCluster() // here instead. MiniHBaseCluster hbc = util.getHBaseCluster(); if (hbc != null) { hbc.shutdown(); hbc.join(); } util.shutdownMiniZKCluster(); }