Warehouse Schema

When sending your events to a data warehouse via RudderStack, you don’t need to define a schema for your event data. RudderStack automatically does that for you by following a predefined warehouse schema.

This guide details the structure of this warehouse schema and the columns created in various tables based on different event types.

Schema

RudderStack uses the source name (written in snake case, for example, source_name) to create a schema in your data warehouse.

RudderStack treats the warehouse schema as the source of truth. If you manually make any changes to the schema, for example, add or modify a column, or update a column’s data type, RudderStack honors the modified schema and sends the data to the warehouse accordingly.

The following tables are created in your data warehouse for each RudderStack source connected to it:

The following image highlights the warehouse schema and the relationships between the tables:

All the event properties are stored as top-level columns in the corresponding table. The nested properties are prefixed with the parent key. For example, an event with properties { product: { name: iPhone, version: 11 }} will result in RudderStack creating the columns product_name and product_version.

Standard RudderStack properties

RudderStack sets the following standard properties on all above-mentioned tables:

RudderStack automatically converts the property names from camel case to snake case. For more information on the properties captured at the API level, refer to the RudderStack Event Specification guide.

RudderStack reserves the above-mentioned standard properties. In case of any conflict, RudderStack automatically discards the properties set by the user.

RudderStack drops any non-standard properties declared at the top level of an event.

Identify

For every Identify call, RudderStack creates a record in the identifies table and upserts the records in the users table based on the userId.

In case of Google BigQuery, you can use the views created over the tables to query for unique users in the dataset. Refer to the BigQuery documentation for more details.

A sample identify call is shown below:

rudderanalytics.identify(
  "userId",
  {
    email: "alex@company.com",
    first_name: "Alex",
    last_name: "Keener",
    age: 35,
  },
  {
    context: {
      ip: "0.0.0.0",
    },
    anonymousId: "59b703e3-467a-4a1d-9fe6-da27ed319619",
  }
)

The corresponding schemas created for the identifies and users tables are shown in the following sections:

Table: identifies

Table: users

Note that:

• The users table contains the properties from the latest identify call made for an user. It only has the id column (same as user_id in the identifies table) and does not have the anonymous_id column.
• To obtain a user’s anonymous_id, you can query the identifies table by grouping on the user_id column.

Skip sending data to users table

You can skip sending the event data to the users table by setting the skipUsersTable option to true.

For example, you can add the following transformation that adds an integrations object to your event and skips sending the event data to the users table:

event.integrations = {
    "DELTALAKE": {
        options: {
            skipUsersTable: true
        }
    }
}

Use the following names for your preferred warehouse destination in the above snippet:

Track

For every track call, RudderStack creates a record in the tracks and <event_name> table. In addition to the tracks table columns, the <event_name> table includes the properties set by the user via the properties key.

You can also skip sending the event data to tracks table by setting the skipTracksTable option to true.

A sample track event named Add to Cart is shown below:

rudderanalytics.track(
  "Add to Cart", {
    price: 5,
    currency: "USD",
    product_id: "P12345",
    product_name: "N95 Mask",
  }, {
    context: {
      ip: "0.0.0.0",
    },
    anonymousId: "59b703e3-467a-4a1d-9fe6-da27ed319619",
  }
)

The corresponding schemas created for the tracks and add_to_cart tables are as shown:

Table: tracks

The tracks table stores contextual information like ip, device, etc. which helps you to perform analytics-related operations on top of it.

Skip sending data to tracks table

Similar to users table, you can skip sending the event data to the tracks table by setting skipTracksTable to true.

The transformation that adds the integrations object to your event and skips sending the event data to the tracks table:

event.integrations = {
    "DELTALAKE": {
        options: {
            skipTracksTable: true
        }
    }
}

Use the following names for your preferred warehouse destination in the above snippet:

Table: add_to_cart

The add_to_cart table has the same columns as the tracks table. Additionally, it includes the properties set by the user via the properties key.

Page/Screen

For every page/screen call, RudderStack creates a record in the corresponding pages or screens table.

A sample page event is shown below:

rudderanalytics.page(
  "Cart",
  "Cart Viewed",
  {
    path: "/cart",
    title: "Shopping Cart",
    url: "https://rudderstack.com",
  },
  {
    context: {
      ip: "0.0.0.0",
    },
    anonymousId: "59b703e3-467a-4a1d-9fe6-da27ed319619",
  }
)

The corresponding schema created for the pages/screens table is as shown:

Table: pages/screens

Group

For every group call, RudderStack creates a record in the corresponding groups table.

A sample group call is shown below:

rudderanalytics.group(
  "groupId", {
    email: "alex@keener.com",
    first_name: "Alex",
    last_name: "Keener",
    age: 35,
  }, {
    context: {
      ip: "0.0.0.0",
    },
    anonymousId: "59b703e3-467a-4a1d-9fe6-da27ed319619",
  }
)

The corresponding schema created for the groups table is as shown:

Table: groups

Alias

For every alias call, RudderStack creates a record in the corresponding aliases table.

A sample alias call is shown below:

rudderanalytics.alias("9bb5d4c2", "e6ab2c5e")

Table: aliases

Clock skew considerations

RudderStack considers the time at its end to be absolute and assumes any difference on the client-side. Thus, the client clock skew is relative.

If not specified in the payload explicitly, RudderStack calculates timestamp based on originalTimestamp and sentAt to account for the client clock skew.

As mentioned in the above section:

sent_at > original_timestamp is always true. However, timestamp can be more or less than the original_timestamp. Refer to the cases below for more details.

Case 1: original_timestamp < received_at

The following table demonstrates an example of original_timestamp < received_at(when the client-side time is less than the time registered by RudderStack):

In this case, timestamp will be greater than original_timestamp.

Case 2: original_timestamp > received_at

The following table demonstrates an example of original_timestamp > received_at(when the client-side time is more than the time registered by RudderStack):

In this case, timestamp will be less than original_timestamp.

Accepted timestamp formats

RudderStack recognizes only the following subsets of the ISO 8601 timestamp format:

• 2019-09-26
• 2009-05-19 14:39:22
• 2019-09-26T06:30:12.984Z
• 2020-02-11 04:56:55.175116
• 2019-09-26T06:30:12.984+0530
• 2019-09-26T06:30:12.984+05:30

RudderStack does not recognize any other timestamp format apart from the ones mentioned above.

Reserved keywords

There are some limitations when it comes to using the reserved words in a schema, table, or column names. If these words are used in the event names, traits or properties, RudderStack automatically prefixes an underscore(_) when creating the tables or columns for them in your schema.

Integers are not allowed at the start of the schema or table name. Such schema, column, or table names will be prefixed with an underscore. For e.g., 25dollarpurchase will be changed to _25dollarpurchase.

The following table lists the warehouse-specific documentation references for reserved keywords:

How RudderStack handles data type mismatch

Once RudderStack recognizes and sets a data type for a table column, it will not accept any values for the column that cannot be cast to the specified data type.

The values which cannot be cast are set as NULL in the table and stored in the rudder_discards table.

The rudder_discards table schema is shown below:

row_id is the column which users can use to join with original table and update it as required. As mentioned in the above table, it is set to messageId for all tables except the users table, where it corresponds to userId.

The following snippet highlights a sample event whose properties are discarded due to a data type mismatch:

// intial track call using the RudderStack JavaScript SDK

rudderanalytics.track(
  "Add to Cart",
  {
    price: 5, // originally a int value
    currency: "USD",
    product_id: "P12345",
    product_name: "N95 Mask",
    added_at: "2020-05-19 14:39:22", // originally a datetime value
  },
  {
    context: {
      ip: "0.0.0.0",
    },
    anonymousId: "59b703e3-467a-4a1d-9fe6-da27ed319619",
  }
)

// subsequent track call using the RudderStack JavaScript SDK

rudderanalytics.track(
  "Add to Cart",
  {
    price: "NA", // sent as a string in latest event
    currency: "USD",
    product_id: 789, // sent as int but can be casted into original string data type
    product_name: "N95 Mask",
    added_at: "05/25/2020", // sent as invalid datetime value
  },
  {
    context: {
      ip: "0.0.0.0",
    },
    anonymousId: "59b703e3-467a-4a1d-9fe6-da27ed319619",
  }
)

The subsequent records created in the rudder_discards table for the discarded properties from the second event shown in the following table:

RudderStack also provides observability into why the event was discarded with descriptive messages, for example:

• incompatible schema conversion from string to boolean
• incompatible schema conversion from string to int
• incompatible schema conversion from datetime to string
• incompatible schema conversion from bigint to float

FAQ

Can I change the namespace (schema name) of my data warehouse in RudderStack?

Yes, you can. Although the default namespace will be the source name, RudderStack gives you the option to explicitly set the namespace while setting up your warehouse destination. For more information, refer to the warehouse-specific destination settings for configuring the namespace in the RudderStack dashboard.

Why am I not able to see the properties added at the top level of an event in warehouse destination?

RudderStack drops any non-standard properties (properties apart from the standard properties) declared at the top level of an event. However, you can add such properties in the context or properties section of the event payload.

For a more comprehensive FAQ list, refer to the Warehouse FAQ guide.

Questions? Contact us by email or on Slack