Re: UDAF on AWS Hive
Thanks Zheng, and thanks for your great support to this list. I took your idea and wrote the following code that worked for me...I'm no Java whiz...so it's probably fairly inefficient. I do get to talk to the Amazon folks from time to time, so I'll definitely mention my interest in upgrading the Hive version. Thanks again. Matt package com.company.hadoop.hive.udaf; import org.apache.hadoop.hive.ql.exec.UDAF; import org.apache.hadoop.hive.ql.exec.UDAFEvaluator; import org.apache.hadoop.io.Text; import org.apache.hadoop.io.IntWritable; import java.util.Arrays; public class UDAFGroupConcat extends UDAF{ public static class GroupConcatStringEvaluator implements UDAFEvaluator { private Text mOutput; private boolean mEmpty; public GroupConcatStringEvaluator() { super(); init(); } public void init() { mOutput = null; mEmpty = true; } public boolean iterate(Text o, IntWritable N) { if (o!=null) { if(mEmpty) { mOutput = new Text(N+ +o.toString()); mEmpty = false; } else { String temp = mOutput.toString() + \t + N + + o.toString(); String[] split = temp.split(\t); Arrays.sort(split); String sorted = split[0]; for (int i = 1; i split.length; i++) { sorted = sorted + \t + split[i]; } mOutput.set(sorted); } } return true; } public Text terminatePartial() {return mEmpty ? null : mOutput;} public boolean merge(Text o) { if (o!=null) { if(mEmpty) { mOutput = new Text(o.toString()); mEmpty = false; } else { String temp = mOutput.toString() + \t + o.toString(); String[] split = temp.split(\t); Arrays.sort(split); String sorted = split[0]; for (int i = 1; i split.length; i++) { sorted = sorted + \t + split[i]; } mOutput.set(sorted); } } return true; } public Text terminate() {return mEmpty ? null : mOutput;} } } On Fri, Apr 2, 2010 at 4:11 PM, Matthew Bryan gou...@gmail.com wrote: I'm writing a basic group_concat UDAF for the Amazon version of Hiveand it's working fine for unordered groupings. But I can't seem to get an ordered version working (filling an array based on an IntWritable passed alongside). When I move from using Text return type on terminatePartial() to either Text[] or a State class I start getting errors: FAILED: Error in semantic analysis: org.apache.hadoop.hive.ql.metadata.HiveException: Cannot recognize return type class [Lorg.apache.hadoop.io.Text; from public org.apache.hadoop.io.Text[] com.company.hadoop.hive.udaf.UDAFGroupConcatN$GroupConcatNStringEvaluator.terminatePartial() or FAILED: Error in semantic analysis: org.apache.hadoop.hive.ql.metadata.HiveException: Cannot recognize return type class com.company.hadoop.hive.udaf.UDAFGroupConcatN$UDAFGroupConc atNState from public com.company.hadoop.hive.udaf.UDAFGroupConcatN$UDAFGroupConcatNState com.company.hadoop.hive.udaf.UDAFGroupConcatN$GroupConcatNStringEvaluator.terminatePartial () What limits are there on the return type of terminatePartial()shouldn't it just have to match the argument of merge and nothing more? Keep in mind this is the Amazon version of Hive (0.4 I think) I put both versions of the UDAF below, ordered and unordered. Thanks for your time. Matt # Working Unordered /*QUERY: select user, event, group_concat(details) from datatable group by user,event;*/ package com.company.hadoop.hive.udaf; import org.apache.hadoop.hive.ql.exec.UDAF; import org.apache.hadoop.hive.ql.exec.UDAFEvaluator; import org.apache.hadoop.io.Text; public class UDAFGroupConcat extends UDAF{ public static class GroupConcatStringEvaluator implements UDAFEvaluator { private Text mOutput; private boolean mEmpty; public GroupConcatStringEvaluator() { super(); init(); } public void init() { mOutput = null; mEmpty = true; }
Re: UDAF on AWS Hive
Hive 0.4 has limited support on complex types in UDAF. If you are looking for an ad-hoc solution, try putting the data into a single Text. It will be great if you can ask AWS guys upgrading Hive to 0.5. 0.5 has over 100 bug fixes and is much more stable. Zheng On Fri, Apr 2, 2010 at 1:11 PM, Matthew Bryan gou...@gmail.com wrote: I'm writing a basic group_concat UDAF for the Amazon version of Hiveand it's working fine for unordered groupings. But I can't seem to get an ordered version working (filling an array based on an IntWritable passed alongside). When I move from using Text return type on terminatePartial() to either Text[] or a State class I start getting errors: FAILED: Error in semantic analysis: org.apache.hadoop.hive.ql.metadata.HiveException: Cannot recognize return type class [Lorg.apache.hadoop.io.Text; from public org.apache.hadoop.io.Text[] com.company.hadoop.hive.udaf.UDAFGroupConcatN$GroupConcatNStringEvaluator.terminatePartial() or FAILED: Error in semantic analysis: org.apache.hadoop.hive.ql.metadata.HiveException: Cannot recognize return type class com.company.hadoop.hive.udaf.UDAFGroupConcatN$UDAFGroupConc atNState from public com.company.hadoop.hive.udaf.UDAFGroupConcatN$UDAFGroupConcatNState com.company.hadoop.hive.udaf.UDAFGroupConcatN$GroupConcatNStringEvaluator.terminatePartial () What limits are there on the return type of terminatePartial()shouldn't it just have to match the argument of merge and nothing more? Keep in mind this is the Amazon version of Hive (0.4 I think) I put both versions of the UDAF below, ordered and unordered. Thanks for your time. Matt # Working Unordered /*QUERY: select user, event, group_concat(details) from datatable group by user,event;*/ package com.company.hadoop.hive.udaf; import org.apache.hadoop.hive.ql.exec.UDAF; import org.apache.hadoop.hive.ql.exec.UDAFEvaluator; import org.apache.hadoop.io.Text; public class UDAFGroupConcat extends UDAF{ public static class GroupConcatStringEvaluator implements UDAFEvaluator { private Text mOutput; private boolean mEmpty; public GroupConcatStringEvaluator() { super(); init(); } public void init() { mOutput = null; mEmpty = true; } public boolean iterate(Text o) { if (o!=null) { if(mEmpty) { mOutput = new Text(o); mEmpty = false; } else { mOutput.set(mOutput.toString()+ +o.toString()); } } return true; } public Text terminatePartial() {return mEmpty ? null : mOutput;} public boolean merge(Text o) {return iterate(o);} public Text terminate() {return mEmpty ? null : mOutput;} } } Not Working Ordered # /*QUERY: select user, event, group_concatN(details, detail_id) from datatable group by user,event;*/ package com.company.hadoop.hive.udaf; import org.apache.hadoop.hive.ql.exec.UDAF; import org.apache.hadoop.hive.ql.exec.UDAFEvaluator; import org.apache.hadoop.io.Text; import org.apache.hadoop.io.IntWritable; public class UDAFGroupConcatN extends UDAF{ public static class GroupConcatNStringEvaluator implements UDAFEvaluator { private Text[] mArray; private boolean mEmpty; public GroupConcatNStringEvaluator() { super(); init(); } public void init() { mArray = new Text[5]; mEmpty = true; } public boolean iterate(Text o, IntWritable N) { if (o!=nullN!=null) { mArray[N.get()].set(o.toString()); mEmpty=false; } return true; } public Text[] terminatePartial() {return mEmpty ? null : mArray;} public boolean merge(Text[] o) { if (o!=null) { for(int i=0; i=5; i++){ if(mArray[i].getLength()==0){ mArray[i].set(o[i].toString()); } } } return true; } public Text[] terminate() {return mEmpty ? null : mArray;} } } -- Yours, Zheng