startToClose staying constant, scheduleToStart increasing under heavy load

Community Support
1

Ive built a temporal app in TypeScript, with some workflows.

I’m using kubernetes autoscaling and metrics, so that more workers are deployed when the scheduleToStart metric increases.

However, when there is heavy load with clients calling for many workflows to be run, the startToClose metric increases and the scheduleToStart metric stays near 0. Which is the opposite of what I would expect, and means that no new workers are deployed to handle the load.

How can I make it so that scheduleToStart metric increases under load? Alternatively, should I be using a different metric to determine autoscaling?

2

Hello @rich_tweed

However, when there is heavy load with clients calling for many workflows to be run, the startToClose metric increases and the scheduleToStart metric stays near 0

Is this for workflow tasks or activity tasks? could you point to the exact metric here Temporal SDK metrics reference | Temporal Documentation?

Can you check your pods CPU and memory utilization?

should I be using a different metric to determine autoscaling?
For reference, you can have a look at this post
Suggested metrics to autoscale Temporal workers on