I need to do two separate matrix multiplications (S*A and P*A) inside my mappers and emit the results of both. I know i can do that easily with two mapreduce job, but inorder to save running time I need to do them in one job. So what I do is that after doing both multiplications I put both of outputs in context object, but with different keys so that I can distinguish them inside reducer:
LongWritable One = new LongWritable();
One.set(1);
context.write(One, partialSA);
LongWritable two = new LongWritable();
two.set(2);
context.write(two, partialPA);
In reduce, I just need to add all partialSA matrices together and add all partialPA matrices together too. The problem is that If I use combiner, the emitted keys I receive inside combiner are 0 and 1 instead of 1 and two!!! And if i dont use combiner, inside reducer I receive 0 and 1 as keys instead of 1 and 2.
Why is this happening? what is the problem?
Here is the exact cleanup function of my mapper:
public void cleanup(Context context) throws IOException, InterruptedException{
LongWritable one = new LongWritable();
one.set(1);
LongWritable two = new LongWritable();
two.set(2)
context.write(one, partialSA);
context.write(two, partialPA);
}
Here is the reducer() code:
public void reduce(LongWritable key, Iterable<MatrixWritable> values, Context context) throws IOException, InterruptedException{
System.out.println("*** In reduce() **** "+key.get());
Iterator<MatrixWritable> itr = values.iterator();
if(key.get() == 1){
while(itr.hasNext()){
SA.addMatrices(itr.next());
}
}else if(key.get() == 2){
while(itr.hasNext()){
PA.addMatrices(itr.next());
}
}
}
Related
the speed of the last reduce is very slow. the other reduce
the number of my map and reduce is follows
the number of map is 18784, the number of reduce is 1500
the average of time for each reduce about 1'26, but the last reduce is about 2h
i try to change the number of reduce and reduce the size of job. but nothing changed
the last reduce
as for my partition
public int getPartition(Object key, Object value, int numPartitions) {
// TODO Auto-generated method stub
String keyStr = key.toString();
int partId= String.valueOf(keyStr.hashCode()).hashCode();
partId = Math.abs(partId % numPartitions);
partId = Math.max(partId, 0);
return partId;
//return (key.hashCode() & Integer.MAX_VALUE) % numPartitions;
}
I had similar experience, in my case it was due to only one reduce was doing processing all data. This happens due to data skewness. Take a look counters at reducers that is already been processed and the one that is taking lot of time, you will likely see more data is being handled by the reducer that is taking lot of time.
You might want to look into this.
Hadoop handling data skew in reducer
Very probably you are facing skew data problem.
Or your keys are not very well distributed or your getPartition is generating the issue. Itś not clear form me why you are creating a string from the hash code of the string and then getting the hash code for this new string. My suggestion is that first try with the default partition and then look inside the distribution of your keys.
In fact, when you process the large amount of data, you should set the class of Combiner. And if you want to changes encoding you should reset the Reduce function.
for example.
public class GramModelReducer extends Reducer<Text, LongWritable, Text, LongWritable> {
private LongWritable result = new LongWritable();
public void reduce(Text key, Iterable<LongWritable> values,Context context) throws IOException, InterruptedException {
long sum = 0;
for (LongWritable val : values) {
sum += val.get();
}
result.set(sum);
context.write(new Text(key.toString().getBytes("GB18030")), result);
}
}
class GramModelCombiner extends Reducer<Text, LongWritable, Text, LongWritable> {
public void reduce(Text key, Iterable<LongWritable> values,Context context) throws IOException, InterruptedException {
long sum = 0;
for (LongWritable val : values) {
sum += val.get();
}
context.write(key, new LongWritable(sum));
}
}
I have an input file contains:
id value
1e 1
2e 1
...
2e 1
3e 1
4e 1
And I would like to find the total id of my input file. So In my main, I have declare a list so that when I read the input file, I will insert the line into the list
MainDriver.java
public static Set list = new HashSet();
and I my map
// Apply regex to find the id
...
// Insert id to the list
MainDriver.list.add(regex.group(1)); // add 1e, 2e, 3e ...
and In my reduce, I try to use the list as
public void reduce(WritableComparable key, Iterator values,
OutputCollector output, Reporter reporter) throws IOException
{
...
output.collect(key, new IntWritable(MainDriver.list.size()));
}
So I expect the value print out the file, in this case will be 4. But it actually prints out 0.
I have verify that regex.group(1) would extract valid id. So I have no clue why the size of my list is 0 in the reduce process.
The mappers and reducers run on separate JVMs (and often separate machines altogether) both from each other and from the driver program, so there is no common instance of your list Set variable that all of those methods can concurrently read and write to.
One way in MapReduce to count the number of keys is:
Emit (id, 1) from your mapper
(optionally) Sum the 1s for each mapper using a combiner to minimize network and reducer I/O
In the reducer:
In setup() initialize a class-scope numeric variable (int or long presumbly) to 0
In reduce() increment the counter, and ignore the values
In cleanup() emit the counter value now that all keys have been processed
Run the job with a single reducer, so all the keys go to the same JVM where a single count can be made
This is basically ignoring the advantage of using MapReduce in the first place.
Correct me if I'm wrong, but it appears you can map your output from your Mapper by "id", and then in your Reducer you receive something like Text key, Iterator values as the parameters.
You can then just sum up values and output output.collect(key, <total value>);
Example (apologies for using Context rather than OutputCollector, but the logic is the same):
public static class MyMapper extends Mapper<LongWritable, Text, Text, Text> {
private final Text key = new Text("id");
private final Text id = new Text();
public void map(LongWritable key, Text value,
Context context) throws IOException, InterruptedException {
id.set(regex.group(1)); // do whatever you do
context.write(id, countOne);
}
}
public static class MyReducer extends Reducer<Text, Text, Text, IntWritable> {
private final IntWritable totalCount = new IntWritable();
public void reduce(Text key, Iterable<Text> values,
Context context) throws IOException, InterruptedException {
int cnt = 0;
for (Text value : values) {
cnt ++;
}
totalCount.set(cnt);
context.write(key, totalCount);
}
}
If I have only one key. Can I avoid it being sent to only one reducer (and distribute it across multiple reducers)?
I understand that then I might have to have a second map reduce program to combine the reducer outputs?
Is this a good approach? Or please let me know if there is a better way?
I was in a similar situation once. What I did is something like this :
int numberOfReduceCalls = 5
IntWritable outKey = new IntWritable();
Random random = new Random();
public void map(LongWritable key, Text value, Context context)
throws IOException, InterruptedException {
// use a random integer within a limit
outKey.set( random.nextInt(numberOfReduceCalls) );
context.write(outKey, value);
}
I ran a simple wordcount MapReduce example adding combiner with a small change in combiner output, The output of combiner is not merged by reducer. scenario is as follows
Test:
Map -> Combiner ->Reducer
In combiner i added two extra lines to out put a word different and count 1, reducer is not suming the "different" word count. output pasted below.
Text t = new Text("different"); // Added a my own output
context.write(t, new IntWritable(1)); // Added my own output
public class wordcountcombiner extends Reducer<Text, IntWritable, Text, IntWritable>{
#Override
public void reduce(Text key, Iterable<IntWritable> values, Context context) throws IOException, InterruptedException
{
int sum = 0;
for (IntWritable val : values)
{
sum += val.get();
}
context.write(key, new IntWritable(sum));
Text t = new Text("different"); // Added my own output
context.write(t, new IntWritable(1)); // Added my own output
}
}
Input:
I ran a simple wordcount MapReduce example adding combiner with a small change in combiner output, The output of combiner is not merged by reducer. scenario is as follows
In combiner I added two extra lines to out put a word different and count 1, reducer is not suming the "different" word count. output pasted below.
Output:
"different" 1
different 1
different 1
I 2
different 1
In 1
different 1
MapReduce 1
different 1
The 1
different 1
...
How can this happen?
fullcode:
I ran wordcount program with combiner and just for fun i tweaked it in combiner, so i faced this issue.
I have three separate classes for mapper, combiner and reducer.
Driver:
public class WordCount {
public static void main(String[] args) throws IOException, ClassNotFoundException, InterruptedException {
// TODO Auto-generated method stub
Job job = Job.getInstance(new Configuration());
job.setJarByClass(wordcountmapper.class);
job.setJobName("Word Count");
FileInputFormat.addInputPath(job, new Path(args[0]));
FileOutputFormat.setOutputPath(job, new Path(args[1]));
job.setMapperClass(wordcountmapper.class);
job.setCombinerClass(wordcountcombiner.class);
job.setReducerClass(wordcountreducer.class);
job.getConfiguration().set("fs.file.impl", "com.conga.services.hadoop.patch.HADOOP_7682.WinLocalFileSystem");
job.setOutputKeyClass(Text.class);
job.setOutputValueClass(IntWritable.class);
System.exit(job.waitForCompletion(true)? 0 : 1);
}
}
Mapper:
public class wordcountmapper extends Mapper<LongWritable, Text, Text, IntWritable> {
private Text word = new Text();
IntWritable one = new IntWritable(1);
#Override
public void map(LongWritable key, Text value, Context context)
throws IOException, InterruptedException
{
String line = value.toString();
StringTokenizer token = new StringTokenizer(line);
while (token.hasMoreTokens())
{
word.set(token.nextToken());
context.write(word, one);
}
}
}
Combiner:
public class wordcountcombiner extends Reducer<Text, IntWritable, Text, IntWritable>{
#Override
public void reduce(Text key, Iterable<IntWritable> values, Context context) throws IOException, InterruptedException
{
int sum = 0;
for (IntWritable val : values)
{
sum += val.get();
}
context.write(key, new IntWritable(sum));
Text t = new Text("different");
context.write(t, new IntWritable(1));
}
}
Reducer:
public class wordcountreducer extends Reducer<Text, IntWritable, Text, IntWritable>{
#Override
public void reduce(Text key, Iterable<IntWritable> values, Context context) throws IOException, InterruptedException
{
int sum = 0;
for (IntWritable val : values)
{
sum += val.get();
}
context.write(key, new IntWritable(sum));
}
}
The output is normal because you're having two lines doing wrong things :
Why are you having this code
Text t = new Text("different"); // Added my own output
context.write(t, new IntWritable(1)); // Added my own output
In your reducer you're doing the sum and then you're adding to the output different 1 ....
You are writing in the final output of the job a new "1 different" in the reduce function, without doing any kind of aggregation. The reduce function is called once per key, as you can see in the method signature, it takes as arguments a key and the list of values for that key, which means that it is called once for each of the keys.
Since you are using as key a word, and in each call of reduce you are writing to the output "1 different", you will get one of those for each of the words in the input data.
hadoop requires that the reduce method in the combiner writes only the same key that it receives as input. This is required because hadoop sorts the keys only before the combiner is called, it does not re-sort them after the combiner has run. In your program, the reduce method writes the key "different" in addition to the key that it received as input. This means that the key "different" then appears in different positions in the order of keys, and these occurrences are not merged before they get passed to the reducer.
For example:
Assume the sorted list of keys output by the mapper is: "alpha", "beta", "gamma"
Your combiner is then called three times (once for "alpha", once for "beta", once for "gamma") and produces keys "alpha", "different", then keys "beta", "different", then keys "gamma", "different".
The "sorted" (but actually not sorted) list of keys after the combiner has executed is then:
"alpha", "different", "beta", "different", "gamma", "different"
This list does not get sorted again, so the different occurrences of "different" do not get merged.
The reducer is then called separately six times, and the key "different" appears 3 times in the output of the reducer.
I have a Mapper class that is giving a text key and IntWritable value which could be 1 two or three. Depending upon the values I have to write three different files with different keys. I am getting a Single File output with No record in it.
Also, is there any good Multiple Outputs example(with explanation) you could guide me to?
My Driver Class Had this code:
MultipleOutputs.addNamedOutput(job, "name", TextOutputFormat.class, Text.class, IntWritable.class);
MultipleOutputs.addNamedOutput(job, "attributes", TextOutputFormat.class, Text.class, IntWritable.class);
MultipleOutputs.addNamedOutput(job, "others", TextOutputFormat.class, Text.class, IntWritable.class);
My reducer class is:
public static class Reduce extends Reducer<Text, IntWritable, Text, NullWritable> {
private MultipleOutputs mos;
public void setup(Context context) {
mos = new MultipleOutputs(context);
}
public void reduce(Text key, Iterable<IntWritable> values, Context context) throws IOException, InterruptedException {
int sum = 0;
String CheckKey = values.toString();
if("1".equals(CheckKey)) {
mos.write("name", key, new IntWritable(1));
}
else if("2".equals(CheckKey)) {
mos.write("attributes", key, new IntWritable(2));
}
else if("3".equals(CheckKey)) {
mos.write("others", key,new IntWritable(3));
}
/* for (IntWritable val : values) {
sum += val.get();
}*/
//context.write(key, null);
}
#Override
public void cleanup(Context context) throws IOException, InterruptedException {
mos.close();
}
}
P.S I am new to HADOOP/MAP-Reduce Programming.
ArrayList<Integer> l = new ArrayList<Integer>();
l.add(1);
System.out.println(l.toString());
results in "[1]" not 1 so
values.toString()
will not give "1"
Apart from that I just tried to print an Iterable and it just gave a reference, so that is definitely your problem. If you want to iterate over the values do as in the example below:
Iterator<Text> valueIterator = values.iterator();
while (valueIterator.hasNext()){
}
Note that you can only iterate once!
Your problem statement is muddled. What do you mean, "depending on the values"? The reducer gets an Iterable of values, not a single value. Something tells me that you need to move the multiple output code in your reducer inside the loop you have commented out for taking the sum.
Or perhaps you don't need a reducer at all and can take care of this in the map phase. If you are using the reduce phase to end up with exactly 4 files by using a single reduce task, then you can also achieve what you want by flipping the key and value in your map phase and forgetting about MultipleOutputs altogether, because you'll end up with only 3 working reduce tasks, one for each of your int values. To get the 4th one you can output two copies of the record in each map call using a special key to indicate that the output is meant for the normal file, not one of the three special files. Normally I would not recommend such a course of action as you have severe bounds on the level of parallelism you can achieve in the reduce phase when the number of keys is small.
You should also include some anomalous data handling code to the end of your 'if' ladder that increments a counter or something if you encounter a value that is not one of the three you are expecting.