Introducing usage monitoring

Learn about the new Usage feature available in Xata

Written by

Siobhán Meier, Ryan Slade

Published on

March 20, 2024

Today we're excited to announce a brand new feature to monitor what you're using in your Xata workspace! Understandably so, our users have wanted this visibility for both planning and billing purposes. This beta release lets users explore the data of their usage metrics at a workspace level for the first time. This is just the starting point, we have a fun roadmap ahead for how we can improve visibility and provide actionable insights on top of these metrics.

With this release, Xata tracks and displays the following metrics so that customers can gain insight into their usage of Xata:

  • Data storage: The storage on disk consumed by Postgres
  • Search storage: The storage on disk for dedicated search indexing
  • File storage: The storage on disk consumed by file attachments
  • AI questions: The number of questions sent to the Xata AI ask endpoint
  • Total branches: The current total number of database branches

We are actively working on enhancements to the usage monitoring feature. We will soon expose these metrics on a database branch level as well, and we plan to release further metric insights in time.

#

Feature description

The usage monitoring feature introduces a new Usage page showing detailed metrics over a number of time ranges to choose from. This new page is exclusively available to our pro plan workspaces.

Pro workspace Usage page
Pro workspace Usage page

Additionally, pro plan workspaces receive a metrics summary for the current calendar month on the workspace landing page. This gives users an up-to-date, high level overview of the workspace usage at a glance.

Pro workspace overview
Pro workspace overview

Free plan workspaces can also get a glimpse into their usage on the workspace landing page. This concisely shows the percentage of available free tier resources that have been used in the current calendar month.

Free workspace usage overview
Free workspace usage overview

As you would expect, the data we track comes from a variety of sources and we wanted to collect it all into a single place.

In order to power this feature we had to choose a database that is well-suited to collecting and reporting time series data and meeting our usage requirements:

  • Fast inserts
  • Fast and flexible querying (ideally using SQL)
  • Support for rollup and eviction of old date
  • Efficient on disk storage
  • Support for high cardinality metrics

After investigating and experimenting with many different databases we eventually landed on Clickhouse because it meets all of the above requirements on paper and after some real world experimentation they all held up in practice.

In order to collect the data in (near) real time, we have two collection strategies:

  1. Poll usage data from existing backends (Postgres, Search, Attachments)
  2. React to real time usage events (AI questions, branch changes)

Polling

We have an existing task scheduler and background job processor which we use to periodically spawn collection jobs in each region. These jobs then concurrently fan out to all databases within their region, collecting usage data, collating it and then storing it in Clickhouse using their bulk insert mechanism.

Event processing

For events caused by user actions we rely on the fact we store metadata about these events in DynamoDB. AWS provides a number of mechanisms for reacting to changes in Dynamo data and in our case we send change events to Kinesis which in turn spawns a lambda function which writes the event into Clickhouse. Although there are a more moving parts, it allows us to record these events almost instantaneously as well as providing a level of resilience in the face of failures since events can build up in the queue until Clickhouse is ready to receive them.

On disk compression

One of the most impressive aspects of Clickhouse in practice is how well data is compressed on disk which is thanks in a large part to its decision to store data by column. If values in a column don’t change that frequently over time, which is often the case, then the column data can be very easily compressed.

For example, let’s take the case of tracking Postgres disk usage over time. We’re tracking four pieces of data each time we take a snapshot:

  • User workspace id
  • User branch id
  • Disk used in bytes
  • Timestamp

Because we take snapshots of this data, each row in a single snapshot will all contain the same timestamp. In practice this means that Clickhouse only needs to store the value once as well as a corresponding counter to indicate how many rows it applies to.

Also, because we have chosen to use the Merge Tree backend on Clickhouse, it will occasionally merge snapshots together allowing it to further combine rows for the same workspace or branch together. This allows Clickhouse to efficiently store data that has a relatively low cardinality, but at the same time doesn't stop it from being able to also store high cardinality data.

Clickhouse stores data in what it calls "parts", which equate in practice to a set of files on disk. Here's a snapshot from a real part showing the number of rows and space consumed – both compressed and uncompressed:

rows:                        300.25 million
data_uncompressed_bytes:     12.86 GiB
data_compressed_bytes:       1.13 GiB
primary_key_bytes_in_memory: 2.03 MiB
bytes_on_disk:               1.13 GiB

You can see here that data has been compressed by around 12x. Additionally, because we include the workspace as part of our primary key, the number of actual rows we need to scan when querying is relatively low.

SELECT count(*)
FROM postgres_disk
WHERE workspace_id = 'abc'
1 row in set.
Elapsed: 0.355 sec.
Processed 45.69 thousand rows, 685.32 KB
(128.57 thousand rows/s., 1.93 MB/s.)

Note here that instead of scanning approximately 300 million rows, we only needed to scan around 45k.

Learn more about the development of this feature from the folks that built it, and watch a quick demo to see it in action. Check out our latest meet the makers session here:

Pop into Discord and say hi if you'd like to dig in further 👋

Interested in trying out for yourself? Sign up today! Happy building 🦋

Start free,
pay as you grow

Xata provides the best free plan in the industry. It is production ready by default and doesn't pause or cool-down. Take your time to build your business and upgrade when you're ready to scale.

Free plan includes
  • Single team member
  • 10 database branches
  • High availability
  • 15 GB data storage
Start freeExplore all plans
Free plan includes
  • Single team member
  • 10 database branches
  • High availability
  • 15 GB data storage

Sign up to our newsletter

By subscribing, you agree with Xata’s Terms of Service and Privacy Policy.

Copyright © 2024 Xatabase Inc.
All rights reserved.

Product

Feature requestsPricingStatusAI solutions