Clarification on using new Client()

[NN-Binary]

I would like to control my own 'keep-alive' as in re-using connections whenever I can and be in control of it, currently using the Undici.dispatch way, I would like to know how should I start? I'm thinking of re-using the same undici.Client for each hostname and put them in a Map and just retrieve them.

MAP[Hostname] = new undici.Client('https://www.google.com');

My main question is: If I'm doing so, will it be enough to guarantee that I'm re-using the same TCP connection? I would like to avoid the TLS handshake all the time but being able to just request anything (via dispatch Undici method) in the coming seconds without having to reconnect.

Thank you, still learning and I'm having difficulties

[paganwinter]

I'm interested in knowing about this as well!

[mcollina]

The usage is correct. Client guarantees that only one socket is used at most.

[LucCADORET]

I would like to know if there are a difference in the use of the connections depending on if you use the unidici.request() compared to a client. E.g. if I'm performing 5 requests with unidici.requests on the same endpoint, is it more costly that doint the same with a client ?

[LucCADORET]

Thanks for the info. I'd be happy to contribute once I've fixed my problem (timeouts when running burst of requests in an Azure VM)

[mcollina]

Can you describe more what your problems are?

[LucCADORET]

On an Azure VM, my program is doing lots of outbound API calls at various intervals, it can go up to 800 at the same instant sometimes
However, when the requests spike arrives, I start to see lots of failed queries (header timeouts, or the queries' signatures get invalidated because the signing requests uses an expired timestamp)
I have tried to limit the connections to 50 using an Agent as a global dispatcher, with pipelining, but it still doesn't handle the spikes properly. I suspect that I run into port exhaustion ( https://learn.microsoft.com/en-us/azure/app-service/troubleshoot-intermittent-outbound-connection-errors )
I wanted to try to use a separate agent per queried domain (I query up to 14 different domains) instead of just one global agent, to fine-tune the amount of connections per domain

[mcollina]

I would recommend you to use BalancedPool directly. Note that it emits events that allows you to track if there is capacity in the pool: in this way you can sign &send your requests only when there is immediate capacity.

[LucCADORET]

Re-implementing with a BalancedPool did not solve our issues, but at this point I am unsure of the exact cause, I need to investigate more. What I did observe though, by looking at the pools stats is that the connected value stays to 1 even though connections in the BalancedPool options is set to 128. Pipelining is set to 10 at the same time, and other values such as running shows more than 1 (max observed is 10). So, not sure what connections means exactly, if connected seems caped to 1