mirror of
https://github.com/apache/impala.git
synced 2026-01-02 03:00:32 -05:00
bd6d2df730
The TableFlattener takes a nested dataset and creates an equivalent unnested dataset. The unnested dataset is saved as Parquet. When an array or map is encountered in the original table, the flattener creates a new table and adds an id column to it which references the row in the parent table. Joining on the id column should produce the original dataset. The flattened dataset should be loaded into Postgres in order to run the query generator (in nested types mode) on it. There is a script that automates generaration, flattening and loading random data into Postgres and Impala: testdata/bin/generate-load-nested.sh -f Testing: - ran ./testdata/bin/generate-load-nested.sh -f and random nested data was generated and flattened as expected. Change-Id: I7e7a8e53ada9274759a3e2128b97bec292c129c6 Reviewed-on: http://gerrit.cloudera.org:8080/5787 Reviewed-by: Alex Behm <alex.behm@cloudera.com> Tested-by: Impala Public Jenkins
This is a tool to convert a nested dataset to an unnested dataset. The source and/or
destination can be the local file system or HDFS.
Structs get converted to a column (with a long name). Arrays and Maps get converted to
a table which can be joined with the parent table on id column.
$ mvn exec:java \
-Dexec.mainClass=org.apache.impala.infra.tableflattener.Main \
-Dexec.arguments="file:///tmp/in.parquet,file:///tmp/out,-sfile:///tmp/in.avsc"
$ mvn exec:java \
-Dexec.mainClass=org.apache.impala.infra.tableflattener.Main \
-Dexec.arguments="hdfs://localhost:20500/nested.avro,file://$PWD/unnested"
There are various options to specify the type of input file but the output is always
parquet/snappy.
For additional help, use the following command:
$ mvn exec:java \
-Dexec.mainClass=org.apache.impala.infra.tableflattener.Main -Dexec.arguments="--help"
This is used by testdata/bin/generate-load-nested.sh.