Structure of Test Framework

Adding Tests

  1. In the Functional / Advanced directory, create a directory for a new test suite (or cd to an existing test suite).
  2. In the test suite directory, add testcases, expected results, and test definition file(s). Optionally include test suite sub directories to organize tests.
  3. Pairs of testcase (ex: query1.sql) and expected result files (ex: query1.e_tsv) are co-located and share the same name.
  4. You could generate expected result files using Postgres or any such database.
  5. In the [Datasources][framework/resources/Datasources] directory, create corresponding datasource directories and copy over any required scripts and datasets required by the tests.
  6. Refer to README for downloading datasets required for the Advanced tests.

Add testcases

A testcase consists a query file and a baseline file containing the expected result. The query file contains a list of SQL statements. Only one query statement that proceeded with a “--@test” line (can't contain “;”) is verified against the baseline. If no query meets that requirement, the middle query is verified. Here is an example:

-- Some setup queries
setup query one;
--@test The following query will be verified.
select * from table;
-- Some teardown queries
teardown query one;

Structure of test definition files.

A test definition file is a JSON file that defines one or a set of tests within a directory. The framework scans for files with .json extension in the specified location(s) and executes all test(s) defined.

The data described in “datasources” section is prepared in several stages. The stages are defined by “mode” field and executed in the following order:

  • rm - remove data specified in “dest” field
  • cp - copy data from “src” in local file system to “dest” in the target fs
  • mkdir - make directory specified in “dest”
  • gen - execute command specified in “src”. In case of a file with .ddl extension it will be send to Drill through JDBC
  • post_rm - the same as “rm” but executed later
  • post_cp - the same as “cp” but executed later
  • dfs_cp - copy data between dfs locations
  • ddl - executes specified ddl file through JDBC

Note that local paths are based on DRILL_TEST_DATA_DIR property (by default it is framework/resources), and all dfs paths are based on DRILL_TESTDATA property (by default it is /drill/testdata)

Special handling for Drill version in regex based baseline verification

The Drill version in a regex based baseline file can be substituted with “{DRILL_VERSION}”. At runtime, it will be replaced with actual Drill version and verified against the running version.

Special handling for order-by tests

The Test Framework is able to verify the order in which rows are returned for some SQL statements that have an order-by clause. If the columns in the order-by clause are also projected (present) in the output, then the Test Framework may be able to check the order of the rows that are returned. Here are some requirements for a query to be validated by the Test Framework:

  1. All columns/fields in the order-by clause must appear in the projection list
  2. Expressions cannot be used in the order-by clause. Things like “order by column.field[2]”. The [2] indicates an expression which is the third element in the field array.
  3. Referencing a field within a json string in a column is more complicated. Most cases are supported. Some cases may not work.
  4. If a query references more than one table, then use aliases for each column in the projection list, and reference these aliases in the order-by clause. Using aliases is a good practice in general when verifying an order-by clause.
  5. The order-by clause cannot be followed by another SQL operation except for limit. If the order-by clause is followed by an offset or collate, for example, it might not work.

Here are some queries that can be validated:

  1. select id, gbyi from complex.json order by id limit 10;

This query can be validated because the results are ordered by the “id” column, and the “id” column is projected in the output. The Test Framework can examine the output and verify that the rows are in order.

  1. select * from (select d.uid uid, flatten( evnts from data.json d order by d.uid) s order by s.evnts.event_time, s.evnts.campaign_id;

This query can be validated because the results are ordered by the data in the “evnts” column, and the “evnts” column is projected in the output. The Test Framework can parse the JSON string in the “evnts” column and examine the event_time and campaign_id values.

These queries cannot be validated:

  1. select t.gbyt,, t.ooa[0].in zeroin, t.ooa[1].fl.f1 flf1, t.ooa[1].fl.f2 flf2, t.ooa[1].in onein, t.ooa[2], t.ooa[2], t.ooa[2] from complex.json t where t.ooa[2] is not null order by t.ooa[2] limit 10;
aaa10nullnullnull10aaa 10bbb 10ccc 10
ooos1000nullnullnull1000aaa 1000bbb 1000ccc 1000
nul10002nullnullnull10002aaa 10002bbb 10002ccc 10002
sba100031000310003.6789154351.010003aaa 10003bbb 10003ccc 10003
str1000810008nullnull10008aaa 10008bbb 10008ccc 10008
fl1000910009nullnull10009aaa 10009bbb 10009ccc 10009
saa1001null1001.678964331.01001aaa 1001bbb 1001ccc 1001
soa10023nullnullnull10023aaa 10023bbb 10023ccc 10023
nul10028null10028.6789154601.010028aaa 10028bbb 10028ccc 10028
ooos10029null10029.6789154611.010029aaa 10029bbb 10029ccc 10029

This query cannot be validated because the order-by has an expression, “t.ooa[2]”. The Test Framework cannot evaluate the array reference “ooa[2]”.

  1. select id from complex.json order by gbyi limit 10;

This query cannot be validated because the “gbyi” column is not projected so the Test Framework cannot determine what the order should be.