Outdated system requirements? #3053
Comments
|
Hello! Yeah, sadly the excessive resource consumption is a known problem and everyone's not happy about it. There's an issue already about it though #1763. You can also see this issue if you're interested: #2539. But, some status update:
The engineers behind relay is aware of this... horrifying RAM usage. Pinging @Dav1dde since he said he wants to work on this after Metrics' done.
There's a ongoing work to migrate Snuba from Python into Rust. You can see some
The one that usually takes up much RAM resource is Kafka + Zookeeper, there's an ongoing work to migrate into Kafka with KRaft (see #2501) which removes the Zookeeper. But for my own instance at office, I replaced Kafka and Zookeeper with Redpanda, it works, but if something goes wrong, you gotta figure it out yourself. I'm going to close this issue, you may continue your rant on the issues I linked above, or continue it here. I can reopen this issue if you want to, unless the Github bot locked this issue, as I don't have enough power to unlock it. Oh, additional note: If you wish for any changes that you think does not align with your experience on setting up Sentry, feel free to add a comment here on this issue, or on this issue: #2267. We're (or basically I am) working on making the docs better and not confusing. |
|
I'm sorry if this felt like a rant, that is not what I wanted. What I wanted to know is if these RAM usage are what is expected, and, if so, if this chapter (Self-Hosted Sentry: Recommended system resource) can be updated? GitLab has a nice page where they list reference architectures, and what instance(s)/resources are needed at some users count: https://docs.gitlab.com/ee/administration/reference_architectures/ |
No, don't be sorry. I also occasionally rant about how cumbersome sentry's deployment is, and I'm helping maintain the self-hosted repo lol.
Yes, we could update it. I like how GitLab does it, I used to handle my previous company's Gitlab instance as well. But for me, usually the proper ones comes with something along the line of: For AWS, it's recommended to use m5.xlarge, for GCP it's recommended to use ..., since I'm currently running sentry on my company's VMware and on a Contabo instance. I'm thinking of reopening this issue and make it as one of the stuff that needs to be documented (by adding a "Documentation" issue label). What do you think about that? |
Self-Hosted Version
24.4.1
CPU Architecture
x86_64
Docker Version
26.1.0
Docker Compose Version
2.26.1
Steps to Reproduce
Install Sentry
Expected Result
Sentry works properly with 4-16GB of RAM
Actual Result
Sentry uses 14GB on its own, add a few system services and you're maxing out on RAM, and the host will crash if you don't have any swap configured.
Our instance doesn't feel very big to me, about 2k errors per day and 16k transactions. However, with all the consumers and services, my 16GB host is struggling to keep running.
Sentry now comes with 54 containers (!), half of which (27) being "consumers", 17 being snuba consumers specifically.
Relay uses 1.5GB by its own. Snuba consumers are using a total of 2.75GB, the other ones 3.56GB. 5.47GB are used by the other services, most of which are web, kakfa, worker and clickhouse.
Is this a normal RAM utilization for an instance this size?
Event ID
No response
The text was updated successfully, but these errors were encountered: