I have a bunch of text files which are categorized and I would like to create a sequence file for each category in which the key is the category name and the value consists of all the textual content of all the files for the category.
I have a nosql database which has only two columns. Each row represents a file, the first column is the category name and the second one is the absolute address of the text file stored on the HDFS. My mapper reads the database and output pairs in which the key is the category and the value is the absolute address. In the reducer sides, I have the addresses of all the files for each category and I would like to create one sequence files for each category in which the key is the category name and the value consists of the all textual content of all the files belonging to that category.
A simple solution is to iterate through the pairs (in the reducer) and open files one by one and append their content to a String variable and at the end create a sequence file using MultipleOutputs. However as the file sizes may be large appending the content to a single String may not be possible. Is there any way to do this without using a String variable?
Then, since you have all the files in reducer, you can get the content of those files, and append using StringBuilder to save memory, and then discard that StringBuilder reference. If avoiding String is your question, StringBuilder is a quick way. The IO operaion involving the file access and reading is resource intensive. However the data itself, should be ok given the architecture of reducers in hadoop.
You can also think of using a combiner. However, that is mainly used to reduce the traffic between map and reduce. You can take advantage of preparing part of the sequence file, at combiner and then remaining at reducer level. ofcouse this is valid only if the content can be added as it comes and not based on specific order.
Related
I have a bunch of Parquet files containing data where each row has the form [key, data1, data2, data3,...]. I need to know in which file a certain key is located, without actually opening each file and searching. Is it possible to get this from the Parquet metadata?
The keys are formatted as strings.
I already tried accessing the metadata using PyArrow, but didn't get the data I wanted.
Short answer is no.
Longer answer: Parquet has two types of metadata that help in eliminating data, min/max statistics and optionally BloomFilters. With these two you can definitively determine if a file does not contain your key, but can't determine if 100% does (unless your key happens to be a min/max value). Pyarrow currently only really exposes row group statistics and doesn't support BloomFilter reading/writing at all.
Also, if the key is of low enough cardinality then dictionary encoding might be used to encode the column. If all data in a column is dictionary encoded, the it might be possible through some lower level APIs (likely not pyarrow) to retrieve the dictionaries and scan them instead of the entire file.
If you are in control of the writing process then sorting data based on key/limiting the number of keys per file would help make these methods even more efficient.
I have a file with data like
City,Quarter,Classification,Index
Bordeux,Q1,R,3
Krakow,Q1,U,2
Halifax,Q1,U,4
I need to find out the highest Index in each Classification and write them to two separate files. The output should be
Bordeux,Q1,R,3
Halifax,Q1,U,4
How to load the data in Mapper as it requires a key/value pair. In mapper it seems programmer should not do any modification to data. So, how to load it in Context object.
I think the data type of key or value is not changed in Reducer. If so, I'm going to infuse my logic to find the top records, then how to organize into a context object there.
I don't have clue on how to proceed.
Necessary pointers will help me to proceed further.
In your case when you read the file in Mapper the input key is the ObjectId of the line and value is the line itself. So in other words, in each line of the file will be received in Mapper as value field. Now, the output (key,value) of Mapper should be (Classification, Index).
The output of Mapper will become input (key,value) of reducer. So reducer will receive (Classification,Iterable) as input. So for each classification, you can iterate over Index List to get the Max and output of the reducer will be (Classification,Max)
In this case, output key and value type will be same in for Mapper and Reducer.
However, regarding writing it to separate lines: Separate files will be generated only if every key is routed to different reducer instance. So in your case, the total number of reducers should be equal to the total number of unique classifications(Not in good terms of resource utilization though). So, you have to write a custom partitioner to make it happen
Writing parquet data can be done with something like the following. But if I'm trying to write to more than just one file and moreover wanting to output to multiple s3 files so that reading a single column does not read all s3 data how can this be done?
AvroParquetWriter<GenericRecord> writer =
new AvroParquetWriter<GenericRecord>(file, schema);
GenericData.Record record = new GenericRecordBuilder(schema)
.set("name", "myname")
.set("favorite_number", i)
.set("favorite_color", "mystring").build();
writer.write(record);
For example what if I want to partition by a column value so that all the data with favorite_color of red goes in one file and those with blue in another file to minimize the cost of certain queries. There should be something similar in a Hadoop context. All I can find are things that mention Spark using something like
df.write.parquet("hdfs:///my_file", partitionBy=["created_year", "created_month"])
But I can find no equivalent to partitionBy in plain Java with Hadoop.
In a typical Map-Reduce application, the number of output files will be the same as the number of reduces in your job. So if you want multiple output files, set the number of reduces accordingly:
job.setNumReduceTasks(N);
or alternatively via the system property:
-Dmapreduce.job.reduces=N
I don't think it is possible to have one column per file with the Parquet format. The internal structure of Parquet files is initially split by row groups, and only these row groups are then split by columns.
I have a use case to upload some tera-bytes of text files as sequences files on HDFS.
These text files have several layouts ranging from 32 to 62 columns (metadata).
What would be a good way to upload these files along with their metadata:
creating a key, value class per text file layout and use it to create and upload as sequence files ?
create SequenceFile.Metadata header in each file being uploaded as sequence file individually ?
Any inputs are appreciated !
Thanks
I prefer storing meta data with the data and then designing your application to be meta data driven, as opposed to embedding meta data in the design or implementation of your application which then means updates to metadata require updates to your app. Ofcourse there are limits to how far you can take a metadata driven application.
You can embed the meta data with the data such as by using an encoding scheme like JSON or you could have the meta data along side the data such as having records in the SeqFile specifically for describing meta data perhaps using reserved tags for the keys so as to given metadata its own namespace separate from the namespace used by the keys for the actual data.
As for the recommendation of whether this should be packaged into separate Hadoop files, bare in mind that Hadoop can be instructed to split a file into Splits (input for map phase) via configuration settings. Thus even a single large SeqFile can be processed in parallel by several map tasks. The advantage of having a single hdfs file is that it more closely resembles the unit of containment of your original data.
As for the recommendation about key types (i.e. whether to use Text vs. binary), consider that the key will be compared against other values. The more compact the key, the faster the comparison. Thus if you can store a dense version of the key that would be preferable. Likewise, if you can structure the key layout so that the first bytes are typically NOT the same then it will also help performance. So, for instance, serializing a Java class as the key would not be recommended because the text stream begins with the package name of your class which is likely to be the same as every other class and thus key in the file.
If you want data and its metadata bundled together, then AVRO format is the appropriate one. It allows schema evolution also.
The simplest thing to do is to make the keys and values of the SequenceFiles Text. Pick a meaningful field from your data to make the Key, the data itself is the value as a Text. SequenceFiles are designed for storing key/value pairs, if that's not what your data is then don't use a SequenceFile. You could just upload unprocessed text files and input those to Hadoop.
For best performance, do not make each file terabytes in size. The Map stage of Hadoop runs one job per input file. You want to have more files than you have CPU cores in your Hadoop cluster. Otherwise you will have one CPU doing 1 TB of work and a lot of idle CPUs. A good file size is probably 64-128MB, but for best results you should measure this yourself.
how do we design mapper/reducer if I have to transform a text file line-by-line into another text file.
I wrote a simple map/reduce programs which did a small transformation but the requirement is a bit more elaborate below are the details:
the file is usually structured like this - the first row contains a comma separated list of column names. Second and the rest of the rows specify values against the columns
In some rows the trailing column values might be missing ex: if there are 15 columns then values might be specified only for the first 10 columns.
I have about 5 input files which I need to transform and aggregate into one file. the transformations are specific to each of the 5 input files.
How do I pass contextual information like file name to the mapper/reducer program?
Transformations are specific to columns so how do I remember the columns mentioned in the first row and then correlate and transform values in rows?
Split file into lines, transform (map) each line in parallel, join (reduce) the resulting lines into one file?
You can not rely on the column info in the first row. If your file is larger than a HDFS block, your file will be broken into multiple splits and each split handed to a different mapper. In that case, only the mapper receiving the first split will receive the first row with column info and the rest won't.
I would suggest passing file specific meta data in separate file and distribute it as side data. Your mapper or reducer tasks could read the meta data file.
Through the Hadoop Context object, you can get hold of the name of the file being processed by a mapper. Between all these, I think you have all the context information you are referring to and you can do file specific transformation. Even though the transformation logic is different for different files, the mapper output needs to have the same format.
If you using reducer, you could set the number of reducers to one, to force all output to aggregate to one file.