Slow performance identified
Hi Support Team,
We are currently evaluating the performance of our messaging system deployed using Docker-Compose and have encountered some concerns regarding message delivery times. Below are the details of our setup and test results:
1. We have set up a Docker-Compose environment as defined in the attached YAML configuration file (dc-novu.yml).
2. The system architecture includes a publisher and a subscriber, which interact through a topic managed by a headless service (headlessService.ts, as detailed in the attached examples) (test.sh).
3. We conducted a test where we sent 100 sequential messages to the topic and measured the latency from the creation (which we read from
message.createdAt) to the time reception of each message.
4. The test results indicate varied latencies: approximately 10% of the messages are received with a delay of around 1 second, 20% with about 35 ms, and the remainder fall between these two extremes.
5. The test was performed on a machine equipped with 4 CPU cores and 20GB of RAM specifically assigned to WSL. During the test, the CPU usage was consistently at 100%.
We are concerned that these results may indicate a performance bottleneck. Could you please review our configuration and suggest any potential improvements or adjustments that might help optimize performance?
Thank you for your assistance.
Best regards, Grigory3 Replies
Hi @Grigoriy, I'm an AI assistant for Novu. While a team member reviews this question, I'll see if I can find relevant sources for your question.
-----------------------------
While I generate an answer, here are some sources that might be helpful:
* AWS
* Introduction to Novu
* Deploy with Docker
* Messages
-----------------------------
Answer
Sorry, I wasn't able to find any information about your question.
@Zac Clifton
@Grigoriy Sorry for the slow responce, we recommend running a more CPUs in your set up. At minimum we recommend 8 cpus. You can look at our docs for this https://docs.novu.co/self-hosting-novu/introduction
We are are aware that topics are not fully optimized however it is something that we do plan to work on in the future.
I would also recommend running more workers then the docker compose file has in it and would run it around 3-6 workers.
If you run the test again with these changes I am happy to give more feedback on this.