Sentry 10 High Availability


We’re considering using Sentry 10 to collect events from all our applications.
We strongly need High-availability (mainly for event ingestion) and we’re wondering how to achieve that.

Obviously, the install script deploys all parts on a single host.
So do we need to install using home-made scripts if we want to distribute the components over multiple hosts?

Is there some official recommendations for HA?

  • Is it possible to deploy some parts only?
  • Is it possible to configure Redis/ Zookeeper / Kafka in cluster?
  • Which parts can be scaled up for high-load / high-availability?

Has anybody already set up a Sentry 10 in that way?

Yes, it is called :smiley:

The services in the docker-compose file are the minimum requirements with the exception of Symbolicator, if you don’t intend to process native crash reports.


All of them? :slight_smile: I think you need to be more specific but I honestly want to ask why you are not considering our SaaS offering at this point as I’m quite sure it would be cheaper and more reliable overall at the end.

Thank you very much for your answer.

You’re right, your SaaS option would likely be more reliable and cheaper.
But we’re working for a company processing payment data (card numbers…) so we must enforce strong security rules (especially for PCI-DSS compliance)

That’s why we’re interested in the on premise option which will allow us to run the Sentry service and to benefit its integration capabilities (git repository, jira…).

I doubt that our CISO validate a solution where we authorize a cloud service to interact with our Git, JIRA…

Is it possible to deploy some parts only?

My question was unclear; it was about the install script

As I understand, this script only allows to deploy all parts on a single host (Docker Compose based).
What’s the official way to install a “more customized & distributed” Sentry?
For example:

  • Deploy postgres as Master/Slave
  • Deploy 2 or 3 redis nodes
  • Deploy a 3 nodes Kafka cluster
  • Deploy 2 web servers

Do we need to code all the install scripts by ourselves?
(db migration…)

Since I have this on my plate Q2 latest too, we could cooperate in a GH repo or something, and provide others with building blocks without putting the burden on the Sentry team, while also sabotaging their lifelihood. :wink:

I guess all the people going for on-premise have data-on-premise as their main driving force.

We are PCI-DSS compliant though:

That sure is up to you but many other folks, including Jira itself (well, Atlassian) use Sentry :slight_smile:

Yes, the setup we support and advertise is geared towards simplicity rather than scale. There is no “official” way for a customized Sentry as it would need to change based on your needs. I’d recommend starting small and simple, watch the stats and then tune and scale accordingly. The key bottlenecks would be Kafka, Clickhouse, workers, and Redis I’d say. Postgres should be less of a problem after moving the events off of it (which we already did).

Do we need to code all the install scripts by ourselves?

If you are starting fresh, I’d argue you need very little of it. You’d need the sentry upgrade and Snuba initialization commands but rest wouldn’t be needed. You can use the docker-compose.yml file as a blueprint for how the services connect together.