Skip to main content

Upgrading to 0.16.0

dbt v0.16.0 contains many new features, bug fixes, and improvements. This guide covers all of the important information to consider when upgrading from an earlier version of dbt to 0.16.0.

Articles:

Breaking changes

The following changes may require you to change the code in your dbt project after upgrading to v0.16.0.

Seed type inference

A number of improvements have been made to the type-inference logic in dbt. dbt previously errantly converted string values in seed CSV files like sunday or March into date timestamps in the year 0001. This was obviously incorrect and has now been remedied, but if you relied on this functionality, then this represents a breaking change. See this pull request for more information on the change.

One-argument generate_schema_name deprecation

Support for the one-argument variant of generate_schema_name macros (deprecated in a previous release) are no longer supported. If you are using the one-argument variant of generate_schema_name, see the docs on custom schemas for an example of how to use the two-argument variant of generate_schema_name.

BigQuery partition_by syntax

The partition_by config for BigQuery models now accepts a dictionary containing the following keys:

  • field: The field name in the table to partition by
  • data_type: The data type for the partitioning field (date, timestamp, datetime, int64)
  • range: Only required if the data_type is int64 (for range bucket partitioning)

If a string is provided as the partition_by config for a model on BigQuery, dbt will attempt to parse that string out to a field and data_type representation. A future release of dbt will remove the ability for partition_by configs to be configured using a string.

See the docs on BigQuery partitioning for more information on the updated partition_by syntax for BigQuery models. See also this guide for more information on how dbt leverages this new syntax to make incremental models build faster and cheaper.

Source test argument compilation

Source test arguments are now handled the same way as model test arguments. If you are providing jinja expressions in Source table schema tests, then this is a breaking change for your project. See this pull request for more information on the change.

Debug log timestamp formatting

The format of timestamps in debug logs has changed. Previously, a comma (,) was used to separate the seconds and microseconds in debug log timestamps. If you are programmatically consuming the debug logs emitted by dbt, this could be a breaking change. See this pull request for more information on the change.

Docrefs removed from manifest

docrefs are no longer present in the compiled manifest.json file. If you are programmatically consuming the manifest.json file emitted by dbt and making use of the docrefs field in the manifest, then this is a breaking change. See this pull request for more information no the change.

get_catalog macro interface change

get_catalog macros should now accept two arguments: a Relation pointing to an information schema, and a list of schemas to search for in the supplied information schema. If you are overriding the get_catalog macro in your project, you can find more information about this change in this pull request.

snowflake__list_schemas macro interface change

The snowflake__list_schemas macro should now return an Agate dataframe with a column named "name". If you are overriding the snowflake__list_schemas macro in your project, you can find more information about this change in this pull request.

Snowflake databases wih 10,000 schemas

dbt no longer supports running against Snowflake databases containing more than 10,000 schemas. This is due limitations of the show schemas in database query that dbt now uses to find schemas in a Snowflake database. If your dbt project is running against a Snowflake database containing more than 10,000 schemas, you should not upgrade to dbt v0.16.0.

If this change is applicable to your dbt project, please let us know in a dbt issue or in the dbt Slack.

Python requirements

If you are installing dbt in a Python environment alongside other Python modules, please be mindful of the following changes to dbt's Python dependencies:

  • Added dependency on 'cffi>=1.9,<1.14', to fix snowflake-connector-python issues
  • Changed upper bound 'requests<2.23.0', to fix snowflake-connector-python issues
  • Added dependency on 'idna<2.9' to fix snowflake-connector-python issues
  • Changed snowflake-connector-python to 2.2.1
  • Increased google-cloud-bigquery to >=1.22.0 to add support for integer bucket partitioning
  • Changed upper bound on Jinja2 < 3

New and changed documentation

0