# Chapter 8. Making Pig Fly

Who says Pigs can’t fly? Knowing how to optimize your Pig Latin scripts can make a significant difference in how they perform. Pig is still a young project and does not have a sophisticated optimizer that can make the right choices. Instead, consistent with Pig’s philosophy of user choice, it relies on you to make these choices. Beyond just optimizing your scripts, Pig and MapReduce can be tuned to perform better based on your workload. And there are ways to optimize your data layout as well. This chapter covers a number of features you can use to help Pig fly.

Before diving into the details of how to optimize your Pig Latin, it is worth understanding what items tend to create bottlenecks in Pig jobs:

Input size

It does not seem that a massively parallel system should be I/O bound. Hadoop’s parallelism reduces I/O bound but does not entirely remove it. You can always add more map tasks. However, the law of diminishing returns comes into effect. Additional maps take more time to start up, and MapReduce has to find more slots in which to run them. If you have twice as many maps as you have slots to run them, it will take twice your average map time to run all of your maps. Adding one more map in that case will actually make it worse because the map time will increase to three times the average. Also, every record that is read might need to be decompressed and will need to be deserialized.

Shuffle size

By shuffle size I mean the data that is moved from your map tasks to your reduce tasks. All of this data has to be serialized, sorted, moved over the network, merged, and deserialized. Also, the number of maps and reduces matters. Every reducer has to go to every mapper, find the portion of the map’s output that belongs to it, and copy that. So if there are m maps and r reduces, the shuffle will have m x r network connections. And if reducers have too many map inputs to merge in one pass, they will have to do a multipass merge, reading the data from and writing it to disk multiple times (see “Combiner Phase” for details).

Output size

Every record written out by a MapReduce job has to be serialized, possibly compressed, and written to the store. When the store is HDFS, it must be written to three separate machines before it is considered written.

Intermediate results size

Pig moves data between MapReduce jobs by storing it in HDFS. Thus the size of these intermediate results is affected by the input size and output size factors mentioned previously.

Memory

Some calculations require your job to hold a lot of information in memory, for example, joins. If Pig cannot hold all of the values in memory simultaneously, it will need to spill some to disk. This causes a significant slowdown, as records must be written to and read from disk, possibly multiple times.

## Writing Your Scripts to Perform Well

There are a number of things you can do when writing Pig Latin scripts to help reduce the bottlenecks discussed earlier. It may be helpful to review which operators force new MapReduce jobs in Chapters 5 and 6.

### Filter Early and Often

Getting rid of data as quickly as possible will help your script perform better. Pushing filters higher in your script can reduce the amount of data you are shuffling or storing in HDFS between MapReduce jobs. Pig’s logical optimizer will push your filters up whenever it can. In cases where a filter has multiple predicates joined by and, and one or more of the predicates can be applied before the operator preceding the filter, Pig will split the filter at the and and push the eligible predicate(s). This allows Pig to push parts of the filter when it might not be able to push the filter as a whole. Table 8-1 describes when these filter predicates will and will not be pushed once they have been split.

Table 8-1. When Pig pushes filters
 Preceding operator Filter will be pushed before? Comments cogroup Sometimes The filter will be pushed if it applies to only one input of the cogroup and does not contain a UDF. cross Sometimes The filter will be pushed if it applies to only one input of the cross. distinct Yes filter No Will seek to merge them with and to avoid passing data through a second operator. This is done only after all filter pushing is complete. foreach Sometimes The filter will be pushed if it references only fields that exist before and after the foreach, and foreach does not transform those fields. group Sometimes The filter will be pushed if it does not contain a UDF. join Sometimes The filter will be pushed if it applies to only one input of the join, and if the join is not outer for that input. load No mapreduce No mapreduce is opaque to Pig, so it cannot know whether pushing will be safe. sort Yes split No store No stream No stream is opaque to Pig, so it cannot know whether pushing will be safe. union Yes

Also, consider adding filters that are implicit in your script. For example, all of the records with null values in the key will be thrown out by an inner join. If you know that more than a few hundred of your records have null key values, put a filter input by key is not null before the join. This will enhance the performance of your join.

### Project Early and Often

For earlier versions of Pig, we told users to employ foreach to remove fields they were not using as soon as possible. As of version 0.8, Pig’s logical optimizer does a fair job of removing fields aggressively when it can tell that they will no longer be used:

-- itemid does not need to be loaded, since it is not used in the script
txns        = load 'purchases' as (date, storeid, amount, itemid);
todays      = filter txns by date == '20110513'; -- date not needed after this
bystore     = group todays by storeid;
avgperstore = foreach bystore generate group, AVG(todays.amount);

However, you are still smarter than Pig’s optimizer, so there are situations where you can tell that a field is no longer needed but Pig cannot. If AVG(todays.amount) were changed to COUNT(todays) in the preceding example, Pig would not be able to determine that, after the filter, only storeid and amount were required. It cannot see that COUNT does not need all of the fields in the bag it is being passed. Whenever you pass a UDF the entire record (udf(*)) or an entire complex field, Pig cannot determine which fields are required. In this case, you will need to put in the foreach yourself to remove unneeded data as early as possible.

### Set Up Your Joins Properly

Joins are one of the most common data operations, and also one of the costliest. Choosing the correct join implementation can improve your performance significantly. The flowchart in Figure 8-1 will help you make the correct selection.

Once you have selected your join implementation, make sure to arrange your inputs in the correct order as well. For replicated joins, the small table must be given as the last input. For skewed joins, the second input is the one that is sampled for large keys. For the default join, the rightmost input has its records streamed through, whereas the other input(s) have their records for a given key value materialized in memory. Thus if you have one join input that you know has more records per key value, you should place it in the rightmost position in the join. For merge join, the left input is taken as the input for the MapReduce job, and thus the number of maps started are based on this input. If one input is much larger than the other, you should place it on the left in order to get more map tasks dedicated to your jobs. This will also reduce the size of the sampling step that builds the index for the right side. For complete details on each of these join implementations, see the sections “Join” and “Using Different Join Implementations”.

### Use Multiquery When Possible

Whenever you are doing operations that can be combined by multiquery, such as grouping and filtering, these should be written together in one Pig Latin script so that Pig can combine them. Although adding extra operations does increase the total processing time, it is still much faster than running jobs separately.

### Choose the Right Data Type

As discussed elsewhere, Pig can run with or without data type information. In cases where the load function you are using creates data that is already typed, there is little you need to do to optimize the performance. However, if you are using the default PigStorage load function that reads tab-delimited files, then whether you use types will affect your performance.

On the one hand, converting fields from bytearray to the appropriate type has a cost. So, if you do not need type information, you should not declare it. For example, if you are just counting records, you can omit the type declaration without affecting the outcome of your script.

On the other hand, if you are doing integer calculations, types can help your script perform better. When Pig is asked to do a numeric calculation on a bytearray, it treats that bytearray as a double because this is the safest assumption. But floating-point arithmetic is much slower than integer arithmetic on most machines. For example, if you are doing a SUM over integer values, you will get better performance by declaring them to be of type integer.

### Select the Right Level of Parallelism

Setting your parallelism properly can be difficult, as there are a number of factors. Before we discuss the factors, a little background will be helpful. It would be natural to think more parallelism is always better; however, that is not the case. Like any other resource, parallelism has a network cost, as discussed under the shuffle size performance bottleneck.

Second, increasing parallelism adds latency to your script because there is a limited number of reduce slots in your cluster, or a limited number that your scheduler will assign to you. If 100 reduce slots are available to you and you specify parallel 200, you still will be able to run only 100 reduces at a time. Your reducers will run in two separate waves. Because there is overhead in starting and stopping reduce tasks, and the shuffle gets less efficient as parallelism increases, it is often not efficient to select more reducers than you have slots to run them. In fact, it is best to specify slightly fewer reducers than the number of slots that you can access. This leaves room for MapReduce to restart a few failed reducers and use speculative execution without doubling your reduce time. See “Handling Failure” for information on speculative execution.

Also, it is important to keep in mind the effects of skew on parallelism. MapReduce generally does a good job partitioning keys equally to the reducers, but the number of records per key often varies radically. Thus a few reducers that get keys with a large number of records will significantly lag the other reducers. Pig cannot start the next MapReduce job until all of the reducers have finished in the previous job. So the slowest reducer defines the length of the job. If you have 10G of input to your reducers and you set parallel to 10, but one key accounts for 50% of the data (not an uncommon case), nine of your reducers will finish quite quickly while the last lags. Increasing your parallelism will not help; it will just waste more cluster resources. Instead, you need to use Pig’s mechanisms to handle skew.

## Writing Your UDF to Perform

Pig has a couple of features intended to enable aggregate functions to run significantly faster. The Algebraic interface allows UDFs to use Hadoop’s combiner (see “Combiner Phase”). The Accumulator interface allows Pig to break a collection of records into several sets and give each set to the UDF separately. This avoids the need to materialize all of the records simultaneously, and thus spill to disk when there are too many records. For details on how to use these interfaces, see “Algebraic Interface” and “Accumulator Interface”. Whenever possible, you should write your aggregate UDFs to make use of these features.

Pig also has optimizations to help loaders minimize the amount of data they load. Pig can tell a loader which fields it needs and which keys in a map it needs. It can also push down certain types of filters. For information on this, see “Pushing down projections” and “Loading metadata”.

On your way out of a commercial jet airliner, have you ever peeked around the flight attendant to gaze at all the dials, switches, and levers in the cockpit? This is sort of what tuning Hadoop is like: many, many options, some of which make an important difference. But without the proper skills, it can be hard to know which is the right knob to turn. Table 8-2 looks at a few of the important features.

This table is taken from Tables 6-1 and 6-2 in Hadoop: The Definitive Guide, Second Edition, by Tom White (O’Reilly), used with permission. See those tables for a more complete list of parameters.

Table 8-2. MapReduce performance-tuning properties
 Property name Type Default value Description io.sort.mb int 100 The size, in megabytes, of the memory buffer to use while sorting map output. Increasing this will decrease the number of spills from the map and make the combiner more efficient, but will leave less memory for your map tasks. io.sort.factor int 10 The maximum number of streams to merge at once when sorting files. It is fairly common to increase this to 100. min.num.spills.for.combine int 3 The minimum number of spill files (from the map) needed for the combiner to run. mapred.job.shuffle.input.buffer.percent float 0.7 The proportion of total heap size to be allocated to the map outputs buffer (reducer buffer for storing map outputs) during the copy phase of the shuffle. mapred.job.shuffle.merge.percent float 0.66 The threshold usage proportion for the map outputs buffer (defined by mapred.job.shuffle.input.buffer.percent) for starting the process of merging the outputs and spilling to disk.

Compared to Hadoop, tuning Pig is much simpler. There are a couple of memory-related parameters that will help ensure Pig uses its memory in the best way possible. These parameters are covered in Table 8-3.

Table 8-3. Pig performance-tuning properties
 Property name Type Default value Description pig.cachedbag.memusage float 0.1 Percentage of the heap that Pig will allocate for all of the bags in a map or reduce task. Once the bags fill up this amount, the data is spilled to disk. Setting this to a higher value will reduce spills to disk during execution but increase the likelihood of a task running out of heap. pig.skewedjoin.reduce.memusage float 0.3 Percentage of the heap Pig will use during a skew join when trying to materialize one side in memory. Setting this to a higher value will reduce the number of ways that large keys are split and thus how many times their records must be replicated, but it will increase the likelihood of a reducer running out of memory.

All of these values for Pig and MapReduce can be set using the set option in your Pig Latin script (see “set”) or by passing them with -D on the command line.

## Using Compression in Intermediate Results

As is probably clear by now, some of the biggest costs in Pig are moving data between map and reduce phases and between MapReduce jobs. Compression can be used to reduce the amount of data to be stored to disk and written over the network. By default, compression is turned off, both between map and reduce tasks and between MapReduce jobs.

To enable compression between map and reduce tasks, two Hadoop parameters are used: mapred.compress.map.output and mapred.map.output.compression.codec. To turn on compression, set mapred.compress.map.output to true. You will also need to select a compression type to use. The most commonly used types are gzip and LZO. gzip is more CPU-intensive but compresses better. To use gzip, set mapred.map.output.compression.codec to org.apache.hadoop.io.compress.GzipCodec. In most cases, LZO provides a better performance boost. See the sidebar “Setting Up LZO on Your Cluster” for details. To use LZO as your codec, set mapred.map.output.compression.codec to com.hadoop.compression.lzo.LzopCodec.

Compressing data between MapReduce jobs can also have a significant impact on Pig performance. This is particularly true of Pig scripts that include joins or other operators that expand your data size. To turn on compression, set pig.tmpfilecompression to true. Again, you can choose between gzip and LZO by setting pig.tmpfilecompression.codec to gzip or lzo, respectively. In the testing we did while developing this feature, we saw performance improvements of up to four times when using LZO, and slight performance degradation when using gzip.

## Data Layout Optimization

How you lay out your data can have a significant impact on how your Pig jobs perform. On the one hand, you want to organize your files such that Pig can scan the minimal set of records. For example, if you have regularly collected data that you usually read on an hourly basis, it likely makes sense to place each hour’s data in a separate file. On the other hand, the more files you create, the more pressure you put on your NameNode. And MapReduce operates more efficiently on larger files than it does on files that are less than one HDFS block (64 MB by default). You will need to find a balance between these two competing forces.

Beginning in 0.8, when your inputs are files and they are smaller than half an HDFS block, Pig will automatically combine the smaller sections when using the file as input. This allows MapReduce to be more efficient and start fewer map tasks. This is almost always better for your cluster utilization. It is not always better for the performance of your individual query, however, because you will be losing locality of data reads for many of the combined blocks, and your map tasks may run longer. If you need to turn this feature off, pass -Dpig.noSplitCombination=true on your command line or set the property in your pig.properties file.

When processing gigabytes or terabytes of data, the odds are overwhelming that at least one row is corrupt or will cause an unexpected result. An example is division by zero, even though no records were supposed to have a zero in the denominator. Causing an entire job to fail over one bad record is not good. To avoid these failures, Pig inserts a null, issues a warning, and continues processing. This way, the job still finishes. Warnings are aggregated and reported as a count at the end. You should check the warnings to be sure that the failure of a few records is acceptable in your job. If you need to know more details about the warnings, you can turn off the aggregation by passing -w on the command line.