Parquet Time Partitioned Dataset Sink (Deprecated)

Note: Datasets and the Parquet Time Partitioned Dataset Sink are deprecated and will be removed in CDAP 7.0.0.

Sink for a TimePartitionedFileSet that writes data in Parquet format. Every time the pipeline runs, a new partition in the TimePartitionedFileSet will be created based on the logical start time of the run. All data for the run will be written to that partition.

This sink is used whenever you want to write to a TimePartitionedFileSet in Parquet format. For example, you might want to create daily snapshots of a database table by reading the entire contents of the table and writing to this sink.

Configuration

Property	Macro Enabled?	Description

Property	Macro Enabled?	Description
Dataset Name	Yes	Required. Name of the `TimePartitionedFileSet` to which records are written. If it doesn't exist, it will be created.
Dataset Base Path	Yes	Optional. Base path for the `TimePartitionedFileSe`. Defaults to the name of the dataset.
Partition Directory Format	Yes	Optional. Format for the time partition, as used by `SimpleDateFormat`. Defaults to formatting partitions such as `2015-01-01/20-42.142017372000`.
Time Zone	Yes	Optional. The string ID for the time zone to format the date in. Defaults to using UTC. This setting is only used if `filePathFormat` is not null.
Partition Offset	Yes	Optional. Amount of time to subtract from the pipeline runtime to determine the output partition. Defaults to 0m. The format is expected to be a number followed by an 's', 'm', 'h', or 'd' specifying the time unit, with 's' for seconds, 'm' for minutes, 'h' for hours, and 'd' for days. For example, if the pipeline is scheduled to run at midnight of January 1, 2016, and the offset is set to '1d', data will be written to the partition for midnight Dec 31, 2015."
Clean Partitions Older Than		Optional. Configures the sink to delete partitions older than a specified date-time after a successful run. If set, when a run successfully finishes, the sink will subtract this amount of time from the runtime and delete any partitions for time partitions older than that. The format is expected to be a number followed by an 's', 'm', 'h', or 'd' specifying the time unit, with 's' for seconds, 'm' for minutes, 'h' for hours, and 'd' for days. For example, if the pipeline is scheduled to run at midnight of January 1, 2016, and this property is set to 7d, the sink will delete any partitions for time partitions older than midnight Dec 25, 2015.
Compression Codec	No	Default is None. Determines the compression codec to use on the resulting data. Valid values are None, Snappy, GZip, and LZO.
Output Schema	Yes	Required. The Avro schema of the record being written to the sink as a JSON Object.

Example

This example will write to a TimePartitionedFileSet named 'users':

Property	Value

Property	Value
Dataset Name	`users`
Partition Directory Format	`yyyy-MM-dd/HH-mm`
Time Zone	`America/Los_Angeles`
Compression Codec	`Snappy`

It will write data in Parquet format using the given schema. Every time the pipeline runs, a new partition in the TimePartitionedFileSet will be created based on the logical start time of the run with the output directory ending with the date formatted as specified. All data for the run will be written to that partition and compressed using the Snappy codec.

For example, if the pipeline was scheduled to run at 10:00am on January 1, 2015 in Los Angeles, a new partition will be created with year 2015, month 1, day 1, hour 10, and minute 0, and the output directory for that partition would end with 2015-01-01/10-00.