Warehouse Schemas

Warehouse tables

The table below describes the schema in Segment Warehouses:

<source>.aliases A table with all of your alias method calls. This table will include all of the traits you identify users by as top-level columns, for example <source>.aliases.email.
<source>.groups A table with all of your group method calls. This table will include all of the traits you record for groups as top-level columns, for example <source>.groups.employee_count.
<source>.accounts CURRENTLY IN BETA A table with unique group method calls. Group calls are upserted into this table (updated if an existing entry exists, appended otherwise). This table holds the latest state of a group.
<source>.identifies A table with all of your identify method calls. This table will include all of the traits you identify users by as top-level columns, for example <source>.identifies.email.
<source>.users A table with unique identify calls. identify calls are upserted on user_id into this table (updated if an existing entry exists, appended otherwise). This table holds the latest state of of a user. The id column in the users table is equivalent to the user_id column in the identifies table. Also note that this table won’t have an anonymous_id column since a user can have multiple anonymousIds. To get at a user’s anonymousIds, you’ll need to query the identifies table. If you observe any duplicates in the users table, contact us.
<source>.pages A table with all of your page method calls. This table will include all of the properties you record for pages as top-level columns, for example <source>.pages.title.
<source>.screens A table with all of your screen method calls. This table will include all of the properties you record for screens as top-level columns, for example <source>.screens.title.
<source>.tracks A table with all of your track method calls. This table will only include a few standardized properties that are all common to all events: anonymous_id, context_*, event, event_text, received_at, sent_at, and user_id. This is because every event that you send to Segment has completely different properties. For querying by the custom properties, use the <source>.<event> tables instead.
<source>.<event> For track calls, each event like Signed Up or Order Completed also has it’s own table (eg. initech.clocked_in) with columns for each of the event’s distinct properties (eg. initech.clocked_in.time).

Identifies table

Your identifies table is where all of your .identify() method calls are stored. Query it to find out user-level information. It has the following columns:

anonymous_id The anonymous ID of the user.
context_<key> Non-user-related context fields sent with each identify call.
id The unique ID of the identify call itself.
received_at When the identify call was received by Segment.
sent_at When the identify call was triggered by the user.
user_id The unique ID of the user.
<trait> Each trait of the user you record is created as it’s own column, and the column type is automatically inferred from your data. For example, you might have columns like email and first_name.

Querying the Identifies table

To see a list of the columns in the identifies table for your <source> run:

SELECT column_name AS Columns
FROM columns
WHERE schema_name = '<source>'
AND table_name = 'identifies'
ORDER by column_name
Columns
anonymous_id
context_ip
email

Your identifies table is where you can do all sorts of querying about your users and their traits. For example, if you wanted to see the number of unique users you’ve seen on your site each day:

SELECT DATE(sent_at) AS Day, COUNT(DISTINCT(user_id)) AS Users
FROM <source>.identifies
GROUP BY day
ORDER BY day

Groups table

Your groups table is where all of your group method calls are stored. Query it to find out group-level information. It has the following columns:

anonymous_id The anonymous ID of the user.
context_<key> Non-user-related context fields sent with each group call.
group_id The unique ID of the group.
id The unique ID of the group call itself.
received_at When the group call was received by Segment.
sent_at When the group call was triggered by the user.
user_id The unique ID of the user.
<trait> Each trait of the group you record is created as it’s own column, and the column type is automatically inferred from your data. For example, you might have columns like email and name.

Querying the Groups table

To see a list of the columns in the groups table for your <source> run:

SELECT column_name AS Columns
FROM columns
WHERE schema_name = '<source>'
AND table_name = 'groups'
ORDER by column_name
Columns
anonymous_id
context_ip

To see a list of all of the groups using your product run:

SELECT name AS Company
FROM <source>.groups
GROUP BY name
Company
Comcast
Rdio
Warner Brothers

Pages and Screens tables

Your pages and screens tables are where all of your page and screen method calls are stored. Query it to find out information about page views or screen views. It has the following columns:

anonymous_id The anonymous ID of the user.
context_<key> Non-user-related context fields sent with each page or screen call.
id The unique ID of the page or screen call itself.
received_at When the page or screen call was received by Segment.
sent_at When the page or screen call was triggered by the user.
user_id The unique ID of the user.
<property> Each property of your pages or screens is created as it’s own column, and the column type is automatically inferred from your data. For example, you might have columns like referrer and title.

Querying the Pages and Screens tables

To see a list of the columns in the pages table for your <source> run:

SELECT column_name AS Columns
FROM columns
WHERE schema_name = '<source>'
AND table_name = 'pages'
ORDER by column_name
Columns
anonymous_id
context_ip
referrer
...

The pages table can give you interesting information about page views that happen on your site, for example you can see the number of page views grouped by day:

SELECT DATE(sent_at) AS Day, COUNT(*) AS Views
FROM <source>.pages
GROUP BY day
ORDER BY day
Day Views
2015-01-14 2,203,198
2015-01-15 2,393,020
2015-07-21 1,920,290

Tracks table

Your tracks table is where all of your track method calls are stored. Query it to find out information about the events your users have triggered. It has the following columns:

anonymous_id The anonymous ID of the user.
context_<key> Non-user-related context fields sent with each track call.
event The slug of the event name, mapping to an event-specific table.
event_text The name of the event.
id An ID attached to the event at execution time and used for deduplication at the server level.
received_at When the track call was received by Segment.
sent_at When the track call was triggered by the user.
user_id The unique ID of the user.

Querying the Tracks table

Your tracks table is a rollup of all of the different event-specific tables, for quick querying of just a single type. For example, you could see the count of how many unique users signed up each day:

SELECT DATE(sent_at) AS Day, COUNT(DISTINCT(user_id)) AS Users
FROM segment.tracks
WHERE event = 'signed_up'
GROUP BY day
ORDER BY day
Day Users
2015-01-14 25,198
2015-01-15 31,020
2015-07-21 19,290

Event Tables

Your event tables are a series of table for each custom event you record to Segment. We break them out into their own tables because the properties, and thus the columns, differ for each event. Query these tables to find out information about specific properties of your custom events. They have the following columns:

</tr>
anonymous_id The anonymous ID of the user.
context_<key> Non-user-related context fields sent with each track call.
event The slug of the event name, so you can join the tracks table.
event_text The name of the event.
id The unique ID of the track call itself.
received_at When the track call was received by Segment.
sent_at When the track call was triggered by the user.
user_id The unique ID of the user.
<property> Each property of your track calls is created as it’s own column, and the column type is automatically inferred from your data.

Querying the Events tables

To see a list of all of the event tables for a given <source> you can run:

SELECT schema as source, "table" as Event
FROM disk
WHERE schema = '<source>'
  AND "table" != 'aliases'
  AND "table" != 'groups'
  AND "table" != 'identifies'
  AND "table" != 'pages'
  AND "table" != 'screens'
  AND "table" != 'tracks'
ORDER BY "table"
source Event
production signed_up
production completed_order

To see a list of the columns in one of your event tables run:

SELECT column_name AS Columns
FROM columns
WHERE schema_name = '<source>'
AND table_name = '<event>'
ORDER by column_name
Columns
anonymous_id
context_ip

Note: If you send us an array, we will stringify it in Redshift. That way you don’t end up having to pollute your events. It won’t work perfectly if you have a lot of array elements but should work decently to store and query those. We also flatten nested objects. 

Tracks vs. Events Tables

To see all of the tables for your organization, you can run this query:

SELECT schema || '.' || "table" AS table, rows
FROM disk
ORDER BY 1

The source.event tables have all of the same columns as the source.track tables, but they also include columns specific to the properties of each event.

So if you’re recording an event like:

analytics.track('Register', {
  plan: 'Pro Annual',
  accountType: 'Facebook'
});

Then you can expect to see columns named plan and account_type as well as the default event, id, etc. That way you can write queries against any of the custom data send in track calls.

Note: Because properties and traits are added as un-prefixed columns to your tables, there is a chance of collision with our reserved column names. For this reason, properties with the same name as reserved column name (eg. user_id) will be discarded.

Your event tables are one of the more powerful datasets in Segment SQL. They allow you to clearly see which actions users are performing when interacting with your product.

Because every source has different events, what you can do with them will vary. Here’s an example where you can see how many “Enterprise” users signed up for each day:

SELECT DATE(sent_at) AS Day, COUNT(DISTINCT(user_id)) AS Users
FROM <source>.signed_up
WHERE account_type = 'Enterprise'
GROUP BY day
ORDER BY day
Day Users
2015-01-14 258
2015-01-15 320
2015-07-21 190

Here’s an example that queries the daily revenue for an ecommerce store:

SELECT DATE(sent_at) AS Day, SUM(total) AS Revenue
FROM <source>.completed_order
GROUP BY day
ORDER BY day
Day Revenue
2014-07-19 $2,630
2014-07-20 $1,595
2014-07-21 $2,350

New Columns

Columns are created for new event properties and traits. Segment processes the incoming data in batches, based on either data size or an interval of time. If the table doesn’t exist we lock and create the table. If the table exists but new columns need to be created, we perform a diff and alter the table to append new columns.

Note: We create tables for each of your custom events, and columns for each event’s custom properties. Redshift itself has limits on how many can be created, so we do not allow unbounded event or property spaces in your data. Instead of recording events like “Ordered Product 15”, use a single property of “Product Number” or similar._

When we process a new batch and discover a new column needs to be added, we take the most recent occurrence of a column and choose its datatype.

The datatypes that we support right now are: 

-timestamp -integer  -float -boolean -varchar

Column Sizing

After analyzing the data from dozens of customers we set the string column length limit at 512 characters. Longer strings are truncated. We found this was the sweet spot for good performance and ignoring non-useful data.

We special-case compression for some known columns like event names and timestamps. The others default to LZO. We may add look-ahead sampling down the road, but from inspecting the datasets today this would be unnecessary complexity.

After a column is created, Redshift doesn’t allow altering. Swapping and renaming may work down the road, but this would likely cause thrashing and performance issues. If you would like to change the column size, see our docs here.

Timestamps

There are four timestamps associated with every Segment API call: timestamp, original_timestamp, sent_at and received_at.

All four timestamps are passed through to your Warehouse for every ETL’d event. In most cases the timestamps are fairly close together, but they have different meanings which are important.

timestamp is the UTC-converted timestamp which is set by the Segment library. If you are importing historical events using a server-side library, this is the timestamp you’ll want to reference in your queries!

original_timestamp is the original timestamp set by the Segment library at the time the event is created. Keep in mind, this timestamp can be affected by device clock skew. You can override this value by manually passing in a value for timestamp which will then be relabed as original_timestamp. Generally, this timestamp should be ignored in favor of the timestamp column.

sent_at is the UTC timestamp set by library when the Segment API call was sent. This timestamp can also be affected by device clock skew.

received_at is UTC timestamp set by the Segment API when the API receives the payload from client or server. All tables use received_at for the sort key.

IMPORTANT: We highly recommend using the received_at timestamp for all queries based on time. The reason for this is two-fold. First, the sent_at timestamp relies on a client’s device clock being accurate, which is generally unreliable. Secondly, we set received_at as the sort key in Redshift schemas, which means queries will execute much faster when using received_at. You can continue to use timestamp or sent_at timestamps in queries if received_at doesn’t work for your analysis, but the queries will take longer to complete.

However, received_at does not ensure chronology of events. For queries based on event chronology, timestamp should be used.

Here’s additional documentation on timestamps in the context of our spec.

id

Each row in your database will have an id which is equivalent to the messageId which is passed through in the raw JSON events. The id is a unique message id associated with the row.

uuid and uuid_ts

The uuid column is used to prevent duplicates. You can ignore this column.

The uuid_ts column is used to keep track of when the specific event was last processed by our connector, specifically for deduping and debugging purposes. You can generally ignore this column.

Sort Key

All tables use received_at for the sort key. Amazon Redshift stores your data on disk in sorted order according to the sort key. The Redshift query optimizer uses sort order when it determines optimal query plans.

More Help

How do I send custom data to my warehouse?

How do I give users permissions to my warehouse?

Check out our Frequently Asked Questions about Warehouses and a list of helpful queries to get you started.

This page was last modified: 14 Jul 2020


Get started with Segment

Segment is the easiest way to integrate your websites & mobile apps data to over 300 analytics and growth tools.
or
Create free account