Add snapshots to your DAG

Related documentation

• Snapshot configurations
• Snapshot properties
• snapshot command

What are snapshots?

Analysts often need to "look back in time" at previous data states in their mutable tables. While some source data systems are built in a way that makes accessing historical data possible, this is not always the case. dbt provides a mechanism, snapshots, which records changes to a mutable table over time.

Snapshots implement type-2 Slowly Changing Dimensions over mutable source tables. These Slowly Changing Dimensions (or SCDs) identify how a row in a table changes over time. Imagine you have an orders table where the status field can be overwritten as the order is processed.

id	status	updated_at
1	pending	2024-01-01

Now, imagine that the order goes from "pending" to "shipped". That same record will now look like:

id	status	updated_at
1	shipped	2024-01-02

This order is now in the "shipped" state, but we've lost the information about when the order was last in the "pending" state. This makes it difficult (or impossible) to analyze how long it took for an order to ship. dbt can "snapshot" these changes to help you understand how values in a row change over time. Here's an example of a snapshot table for the previous example:

id	status	updated_at	dbt_valid_from	dbt_valid_to
1	pending	2024-01-01	2024-01-01	2024-01-02
1	shipped	2024-01-02	2024-01-02	null

Configuring snapshots

Previewing or compiling snapshots in IDE not supported

It is not possible to "preview data" or "compile sql" for snapshots in dbt Cloud. Instead, run the dbt snapshot command in the IDE.

Configuration best practices

 Use the timestamp strategy where possible

This strategy handles column additions and deletions better than the check strategy.

 Ensure your unique key is really unique

The unique key is used by dbt to match rows up, so it's extremely important to make sure this key is actually unique! If you're snapshotting a source, I'd recommend adding a uniqueness test to your source (example).

How snapshots work

When you run the dbt snapshot command:

• On the first run: dbt will create the initial snapshot table — this will be the result set of your select statement, with additional columns including dbt_valid_from and dbt_valid_to. All records will have a dbt_valid_to = null.
• On subsequent runs: dbt will check which records have changed or if any new records have been created:
  • The dbt_valid_to column will be updated for any existing records that have changed
  • The updated record and any new records will be inserted into the snapshot table. These records will now have dbt_valid_to = null

Note, these column names can be customized to your team or organizational conventions using the snapshot_meta_column_names config.

Snapshots can be referenced in downstream models the same way as referencing models — by using the ref function.

Detecting row changes

Snapshot "strategies" define how dbt knows if a row has changed. There are two strategies built-in to dbt:

• Timestamp — Uses an updated_at column to determine if a row has changed.
• Check — Compares a list of columns between their current and historical values to determine if a row has changed.

Timestamp strategy (recommended)

The timestamp strategy uses an updated_at field to determine if a row has changed. If the configured updated_at column for a row is more recent than the last time the snapshot ran, then dbt will invalidate the old record and record the new one. If the timestamps are unchanged, then dbt will not take any action.

The timestamp strategy requires the following configurations:

Config	Description	Example
updated_at	A column which represents when the source row was last updated	updated_at

Example usage:

Check strategy

The check strategy is useful for tables which do not have a reliable updated_at column. This strategy works by comparing a list of columns between their current and historical values. If any of these columns have changed, then dbt will invalidate the old record and record the new one. If the column values are identical, then dbt will not take any action.

The check strategy requires the following configurations:

Config	Description	Example
check_cols	A list of columns to check for changes, or all to check all columns	["name", "email"]

check_cols = 'all'

The check snapshot strategy can be configured to track changes to all columns by supplying check_cols = 'all'. It is better to explicitly enumerate the columns that you want to check. Consider using a surrogate key to condense many columns into a single column.

Example Usage

Hard deletes (opt-in)

Rows that are deleted from the source query are not invalidated by default. With the config option invalidate_hard_deletes, dbt can track rows that no longer exist. This is done by left joining the snapshot table with the source table, and filtering the rows that are still valid at that point, but no longer can be found in the source table. dbt_valid_to will be set to the current snapshot time.

This configuration is not a different strategy as described above, but is an additional opt-in feature. It is not enabled by default since it alters the previous behavior.

For this configuration to work with the timestamp strategy, the configured updated_at column must be of timestamp type. Otherwise, queries will fail due to mixing data types.

Example Usage

Snapshot query best practices

This section outlines some best practices for writing snapshot queries:

• Snapshot source data

  Your models should then select from these snapshots, treating them like regular data sources. As much as possible, snapshot your source data in its raw form and use downstream models to clean up the data

• Use the source function in your query

  This helps when understanding data lineage in your project.

• Include as many columns as possible

  In fact, go for select * if performance permits! Even if a column doesn't feel useful at the moment, it might be better to snapshot it in case it becomes useful – after all, you won't be able to recreate the column later.

• Avoid joins in your snapshot query

  Joins can make it difficult to build a reliable updated_at timestamp. Instead, snapshot the two tables separately, and join them in downstream models.

• Limit the amount of transformation in your query

  If you apply business logic in a snapshot query, and this logic changes in the future, it can be impossible (or, at least, very difficult) to apply the change in logic to your snapshots.

Basically – keep your query as simple as possible! Some reasonable exceptions to these recommendations include:

• Selecting specific columns if the table is wide.
• Doing light transformation to get data into a reasonable shape, for example, unpacking a JSON blob to flatten your source data into columns.

Snapshot meta-fields

Snapshot tables will be created as a clone of your source dataset, plus some additional meta-fields*.

Starting in 1.9 or with dbt Cloud Versionless, these column names can be customized to your team or organizational conventions via the snapshot_meta_column_names config.

Field	Meaning	Usage
dbt_valid_from	The timestamp when this snapshot row was first inserted	This column can be used to order the different "versions" of a record.
dbt_valid_to	The timestamp when this row became invalidated.	The most recent snapshot record will have dbt_valid_to set to null.
dbt_scd_id	A unique key generated for each snapshotted record.	This is used internally by dbt
dbt_updated_at	The updated_at timestamp of the source record when this snapshot row was inserted.	This is used internally by dbt

*The timestamps used for each column are subtly different depending on the strategy you use:

For the timestamp strategy, the configured updated_at column is used to populate the dbt_valid_from, dbt_valid_to and dbt_updated_at columns.

Details for the timestamp strategy

Snapshot query results at 2024-01-01 11:00

id	status	updated_at
1	pending	2024-01-01 10:47

Snapshot results (note that 11:00 is not used anywhere):

id	status	updated_at	dbt_valid_from	dbt_valid_to	dbt_updated_at
1	pending	2024-01-01 10:47	2024-01-01 10:47		2024-01-01 10:47

Query results at 2024-01-01 11:30:

id	status	updated_at
1	shipped	2024-01-01 11:05

Snapshot results (note that 11:30 is not used anywhere):

id	status	updated_at	dbt_valid_from	dbt_valid_to	dbt_updated_at
1	pending	2024-01-01 10:47	2024-01-01 10:47	2024-01-01 11:05	2024-01-01 10:47
1	shipped	2024-01-01 11:05	2024-01-01 11:05		2024-01-01 11:05

For the check strategy, the current timestamp is used to populate each column. If configured, the check strategy uses the updated_at column instead, as with the timestamp strategy.

Details for the check strategy

Snapshot query results at 2024-01-01 11:00

id	status
1	pending

Snapshot results:

id	status	dbt_valid_from	dbt_valid_to	dbt_updated_at
1	pending	2024-01-01 11:00		2024-01-01 11:00

Query results at 2024-01-01 11:30:

id	status
1	shipped

Snapshot results:

id	status	dbt_valid_from	dbt_valid_to	dbt_updated_at
1	pending	2024-01-01 11:00	2024-01-01 11:30	2024-01-01 11:00
1	shipped	2024-01-01 11:30		2024-01-01 11:30

Configure snapshots in versions 1.8 and earlier

FAQs

How do I run one snapshot at a time?

To run one snapshot, use the --select flag, followed by the name of the snapshot:

$ dbt snapshot --select order_snapshot

Check out the model selection syntax documentation for more operators and examples.

How often should I run the snapshot command?

Snapshots are a batch-based approach to change data capture. The dbt snapshot command must be run on a schedule to ensure that changes to tables are actually recorded! While individual use-cases may vary, snapshots are intended to be run between hourly and daily. If you find yourself snapshotting more frequently than that, consider if there isn't a more appropriate way to capture changes in your source data tables.

What happens if I add new columns to my snapshot query?

When the columns of your source query changes, dbt will attempt to reconcile this change in the destination snapshot table. dbt does this by:

1. Creating new columns from the source query in the destination table
2. Expanding the size of string types where necessary (eg. varchars on Redshift)

dbt will not delete columns in the destination snapshot table if they are removed from the source query. It will also not change the type of a column beyond expanding the size of varchar columns. That is, if a string column is changed to a date column in the snapshot source query, dbt will not attempt to change the type of the column in the destination table.

Do hooks run with snapshots?

Yes! The following hooks are available for snapshots:

• pre-hooks
• post-hooks
• on-run-start
• on-run-end

Can I store my snapshots in a directory other than the `snapshot` directory in my project?

By default, dbt expects your snapshot files to be located in the snapshots subdirectory of your project.

To change this, update the snapshot-paths configuration in your dbt_project.yml file, like so:

dbt_project.yml

snapshot-paths: ["snapshots"]

Note that you cannot co-locate snapshots and models in the same directory.

Debug Snapshot target is not a snapshot table errors

If you see the following error when you try executing the snapshot command:

Snapshot target is not a snapshot table (missing dbt_scd_id, dbt_valid_from, dbt_valid_to)

Double check that you haven't inadvertently caused your snapshot to behave like table materializations by setting its materialized config to be table. Prior to dbt version 1.4, it was possible to have a snapshot like this:

{% snapshot snappy %}
  {{ config(materialized = 'table', ...) }}
  ...
{% endsnapshot %}

dbt is treating snapshots like tables (issuing create or replace table ... statements) silently instead of actually snapshotting data (SCD2 via insert / merge statements). When upgrading to dbt versions 1.4 and higher, dbt now raises a Parsing Error (instead of silently treating snapshots like tables) that reads:

A snapshot must have a materialized value of 'snapshot'

This tells you to change your materialized config to snapshot. But when you make that change, you might encounter an error message saying that certain fields like dbt_scd_id are missing. This error happens because, previously, when dbt treated snapshots as tables, it didn't include the necessary snapshot meta-fields in your target table. Since those meta-fields don't exist, dbt correctly identifies that you're trying to create a snapshot in a table that isn't actually a snapshot.

When this happens, you have to start from scratch — re-snapshotting your source data as if it was the first time by dropping your "snapshot" which isn't a real snapshot table. Then dbt snapshot will create a new snapshot and insert the snapshot meta-fields as expected.