Documentation on frontend.namespaceBurst

Community Support
1

Hi, Temporal community-

I have not found any documentation, Slack threads, or community threads on the frontend.namespaceBurst dynamic config value - but please direct me if there are any.

The default namespace RPS limits are too high for my company’s use cases right now, and I want to lower them to help ensure better namespace isolation. I’m likely going to lower frontend.namespaceRPS from the default of 2400 to ~500.

But I also see the frontend.namespaceBurst sibling option, and I have some questions about it to make sure I set the right value for it as well:

  • How is a “burst” defined? Is that simply exceeding the per-namespace, per-instance RPS at a single point in time?
  • Once a burst has been detected, for how long does the burst RPS take effect?
  • After the burst RPS is no longer in effect, is there any kind of cooldown time period until it can take effect again?

Thanks so much,
Christian

3

Hi, sorry for late response,

frontend.namespaceBurst

I think good info on this is in the rate lib, specifically here.

  • How is a “burst” defined? Is that simply exceeding the per-namespace, per-instance RPS at a single point in time?

Burst is the amount of tokens in the bucket. Rate is the refill rate to the bucket.

  • Once a burst has been detected, for how long does the burst RPS take effect?

If I understand question, as long a there is a “token” in the bucket, the call to frontend is not going to be throttled.

  • After the burst RPS is no longer in effect, is there any kind of cooldown time period until it can take effect again?

Don’t think there is a cooldown (see previous question)

4

This is a concise answer that I believe answers everything, thank you, @tihomir !

5

Hey @tihomir, to piggyback off this thread, we’re curious if there’s any consideration to support global namespace burst?
Or if there is an option to turn off frontend.namespaceBurst (setting the value to 0 seems to restrict all traffic)?

We’d like to avoid setting RPS limits that change based on the number of instances configured.