Dynamic Partitioning

Overview

When writing data in HCatalog it is possible to write all records to a single partition. In this case the partition column(s) need not be in the output data.

The following Pig script illustrates this:

A = load 'raw' using HCatLoader(); 
... 
split Z into for_us if region='us', for_eu if region='eu', for_asia if region='asia'; 
store for_us into 'processed' using HCatStorer("ds=20110110, region=us"); 
store for_eu into 'processed' using HCatStorer("ds=20110110, region=eu"); 
store for_asia into 'processed' using HCatStorer("ds=20110110, region=asia"); 

In cases where you want to write data to multiple partitions simultaneously, this can be done by placing partition columns in the data and not specifying partition values when storing the data.

A = load 'raw' using HCatLoader(); 
... 
store Z into 'processed' using HCatStorer(); 

...

It is also important to note that all partitions created during a single run are part of one transaction; therefore if any part of the process fails, none of the partitions will be added to the table.

External Tables

Starting in HCatalog 0.5, dynamic partitioning on external tables was broken (HCATALOG-500). This issue was fixed in Hive 0.12.0 by creating dynamic partitions of external tables in locations based on metadata rather than user specifications (HIVE-5011). Starting in Hive 0.13.0, users are able to customize the locations by specifying a path pattern in the job configuration property hcat.dynamic.partitioning.custom.pattern (HIVE-5011).6109). Static partitions for external tables can have user-specified locations in all Hive releases.

For example, in Hive 0.12.0 if a table named user_logs is partitioned by (year, month, day, hour, minute, country) and stored at external location "hdfs://hcat/data/user_logs", then the locations of its dynamic partitions have the standard Hive format which includes keys as well as values, such as:

• hdfs://hcat/data/user_logs/year=2013/month=12/day=21/hour=06/minute=10/country=US

In Hive 0.13.0 and later, hcat.dynamic.partitioning.custom.pattern can be configured to a custom path pattern. For example, the pattern "${year}/${month}/${day}/${hour}/${minute}/${country}" omits keys from the path:

• hdfs://hcat/data/user_logs/2013/12/21/06/10/US

Each dynamic partition column must be present in the custom location path in the format ${column_name}, and the custom location path must consist of all dynamic partition columns. Other valid custom path strings include:

• data/${year}/${month}/${day}/${country}

• ${year}­‐${month}­‐${day}/country=${country}

• output/yr=${year}/mon=${month}/day=${day}/geo=${country}

See HCatalog Configuration Properties for another example. Also see the PDF attachment to HIVE-6019 for details of the implementation.

Hive Hive Dynamic Partitions

Information about Hive dynamic partitions is available here:

...

So this statement...

store A into 'mytable' using HCatStorer("a=1, b=1");

...is equivalent to any of the following statements, if the data has only values where a=1 and b=1:

store A into 'mytable' using HCatStorer();

store A into 'mytable' using HCatStorer("a=1");

store A into 'mytable' using HCatStorer("b=1");

...

For example, let's say a=1 for all values across our dataset and b takes the values 1 and 2. Then the following statement...

store A into 'mytable' using HCatStorer();

...is equivalent to either of these statements:

store A into 'mytable' using HCatStorer("a=1");

split A into A1 if b='1', A2 if b='2';
store A1 into 'mytable' using HCatStorer("a=1, b=1");
store A2 into 'mytable' using HCatStorer("a=1, b=2");

...

A current code example for writing out a specific partition for (a=1, b=1) would go something like this:

Map<String, String> partitionValues = new HashMap<String, String>();
partitionValues.put("a", "1");
partitionValues.put("b", "1");
HCatTableInfo info = HCatTableInfo.getOutputTableInfo(dbName, tblName, partitionValues);
HCatOutputFormat.setOutput(job, info);

...

With dynamic partitioning, we simply specify only as many keys as we know about, or as required. It will figure out the rest of the keys by itself and spray out necessary partitions, being able to create multiple partitions with a single job.