Are burstable cpus good for production? (Type sense cloud)

For low traffic use cases (both search and writes), they can be used in prod. But for anything moderate or substantial you'd want to use non-burst. Low/moderate/large traffic is relative to how much CPU your traffic patterns consume

@Jason Bosco search is a core feature in our app so it's better to go with dedicated cpus? I would say for the beginning we will have 100 concurrent users

In that case you definitely want to go with non burst

How is the burst actually working? It starts automatically for 4 continuing hours or the 4 hours are on high traffic? (2 GB Ram - 2 cpu) For example, the instance works with no load, then we have high load for 5 minutes then again normal. In this case, it will start for 4 on first high load or in batches (for 5 minutes)

No, the 4 hours is a cumulative count over the entire day. If you continually use 100% of CPU it will last only 4 hours before the CPU is throttled to a much lower baseline performance. However, if the high load is sporadic, then it can easily be spread throughout the day.

Isn't the price too high in Typesense cloud? For example, 3 instances with 4 GB RAM and 2 CPU (dedicated) cost 410$ in frankfurt. If I use eg. Digital Ocean for 3 instances with the same specs (4 - 2) the cost is 130$ (incl. LB)

With Typesense Cloud, you get a fully managed Typesense cluster. With DO, you still have to maintain the Typesense cluster yourself. So you would also have to add the cost of engineering time to DO costs, which typically works out to be much higher in net. In Typesense Cloud we are able to amortize the infra engineering costs across all customers, so we take advantage of economies of scale.

@Jason Bosco yes I understand this but the cost is ~ x4. If it would even be 2x we would go with typesense cloud without a thought. It's too high for a startup. It's my opinion

You can configure both using server parameters, but those are the default yes

I have already created a cluster in DO. So in firewall only these 2 ports > 8108 http and 8107 tcp?

Yeah, only two ports. One for http and the other for peering between nodes in the cluster

Perfect. thank you @Jason Bosco