This page documents the CI jobs used by the data team in Merge Requests in both the Data Tests and Analytics projects.
CI jobs are grouped by stages.
These jobs are defined in .gitlab-ci.yml.
Runs automatically when the MR opens to be able to run any dbt jobs. Subsequent runs of this job will be fast as it only verifies if the clone exists. This is an empty clone of the prod and prep databases.
Run this if you need to do a real clone of the prod and prep databases. This is a full clone both databases.
Run this if you need to run extract, freshness, or snapshot jobs. Subsequent runs of this job will be fast as it only verifies if the clone exists.
Run this if you only need a clone of the raw tap_postgres schema in order to test changes to postgres pipeline or a manifest file. If the raw clone already exists, this will do nothing.
Run this if you only need a clone of the raw sheetload schema in order to test changes or additions to sheetload. If the raw clone already exists, this will do nothing.
Run this if you need a clone of any other raw schema in order to test changes or additions. Specify which raw schema to clone with the SCHEMA_NAME variable. If the raw clone already exists, this will do nothing.
Run this if you want to force refresh raw, prod, and prep. This does a full clone of raw, but a shallow clone of prep and prod.
These jobs are defined in extract-ci.yml
Run this if you want to test a new boneyard sheetload load. This requires the real prod and prep clones to be available.
Run this if you want to test a new sheetload load. This jobs runs against the clone of RAW. Requires the clone_raw job to have been run.
With the exception of gitlab-ops DB, run this if you're adding or updating a postgres pipeline manifest. Requires a clone of RAW and the MANIFEST_NAME variable, possibly TASK_INSTANCE variable if it's an SCD table.
The MANIFEST_NAME is not the same as the manifest filename. Its the filename except postfix _db_manifest.yaml.
Run this if you're updating the gitlab-ops manifest file. Requires a clone of RAW. Does not require any variables, except possibly a TASK_INSTANCE variable if it's an SCD table. This is separate from the pgp_test job because it requires a CloudSQL Proxy to be running in order to connect to the database.
These jobs are defined in snowflake-dbt-ci.yml
As part of a DBT Model Change MR, you need to trigger a pipeline job to test that your changes won't break anything in production. To trigger these jobs, go to the "Pipelines" tab at the bottom of this MR and click on the appropriate stage (dbt_run or dbt_misc).
These jobs are scoped to the ci target. This target selects a subset of data for the snowplow and version datasets.
Note that job artifacts are available for all dbt run jobs. These include the compiled code and the run results.
These jobs run against the primary RAW database.
Most dbt run jobs can be parameterized with a variable specifying dbt model that requires testing. Watch this intro to see an example of how to set the variable.
The variable DBT_MODELS is a stand-in for any of the examples in the dbt documentation on model selection syntax.
If you are testing changes to tests in the data-tests project, you can pass in DATA_TEST_BRANCH to the manual jobs along with the branch name. This will update the branch in the packages.yml for the data-tests package. This works for any job running dbt test.
You can also add --fail-fast to the end of the model selection to quickly end the dbt call at the first failure. Read the dbt docs for more information.
Specify which model to run with the variable DBT_MODELS
Specify which model to run using an L warehouse with the variable DBT_MODELS
Specify which model to run using an XL warehouse with the variable DBT_MODELS
Specify which model to exclude with the variable DBT_MODELS
Specify which model to exclude using an L warehouse with the variable DBT_MODELS
Specify which model to exclude using an XL warehouse with the variable DBT_MODELS
Specify a dbt model against the clone of the RAW database. This jobs runs against the clone of RAW. Requires the clone_raw job to have been run. This is useful for the following scenarios:
This job tests specific seed file.
Specify seed file with the variable DBT_MODELS.
Specify which snapshot to run with the variable DBT_MODELS.
This jobs runs against the clone of RAW. Requires the clone_raw job to have been run.
These jobs are defined in snowflake-dbt-ci.yml
Runs all of the tests
Runs only data tests
Runs source freshness test. This jobs runs against the clone of RAW. Requires the clone_raw job to have been run.
This job runs automatically and only appears when .sql files are changed. In its simplest form, the job will check to see if any of the currently changed models are queried in Periscope. If they are, the job will fail with a notification to check the relevant dashboard. If it is not queried, the job will succeed.
Current caveats with the job are:
retention_[some_variable])This section explains how the periscope query works.
git clone -b periscope/master --single-branch https://gitlab.com/gitlab-data/periscope.git --depth 1
This clones the periscope project.
git diff origin/$CI_MERGE_REQUEST_TARGET_BRANCH_NAME...HEAD --name-only | grep -iEo "(.*)\.sql" | sed -E 's/\.sql//' | awk -F '/' '{print tolower($NF)}' | sort | uniq > diff.txt
This gets the list of files that have changed from the master branch (i.e. target branch) to the current commit (HEAD). It then finds (grep) only the sql files and substitutes (sed) the .sql with an empty string. Using awk, it then prints the lower-case of the last column of each line in a file (represented by $NF - which is the number of fields), using a slash (/) as a field separator. Since the output is directory/directory/filename and we make the assumption that most dbt models will write to a table named after its file name, this works as expected. It then sorts the results, gets the unique set, and writes it to a file called diff.txt.
periscope_check.py
This recursively searches the entire periscope repo for a string that matches a from|join statement from any of the 3 currently queryable schemas. It does some cleaning on files that match and creates a dictionary of table name mapping to all of the files it is referenced in. It then reads in diff.txt to do a lookup and write to comparison.txt and matches based on the model name.
if (( $(cat comparison.txt | wc -l | tr -d ' ') > 0 )); then echo "Check these!" && cat comparison.txt && exit 1; else echo "All good" && exit 0; fi;
This uses word count (wc) to see how many lines are in the comparison file. If there is more than zero it will print the lines and exit with a failure. If there are no lines it exits with a success.
Runs only schema tests
Runs snapshots. This jobs runs against the clone of RAW. Requires the clone_raw_full job to have been run.
Runs specified model tests with the variable DBT_MODELS
These jobs are defined in .gitlab-ci.yml.
There are several jobs that only appear when .py files have changed. All of them will run automatically on each new commit where .py files are present. Otherwise they are unavailable to run. Other jobs are:
Manual job to do a dry run of Permifrost.
Triggered when there is a change to load/snowflake/roles.yml. Validates that the YAML is correctly formatted.
These jobs are defined in .gitlab-ci.yml.
Runs automatically when MR is merged or closed. Do not run manually.
