Connection pooling and long-running requests

[mkjpryor]

I have an application that does a lot of Kubernetes watches, which means a lot of long-lived HTTP connections. This results in a lot of PoolTimeout errors :-)

What is the recommended thing to do here? Massively increase the size of the connection pool? Have a client per connection?

[lovelydinosaur]

Have a client per connection?

No. Having a client-per-connection won't gain you anything.

increase the size of the connection pool?

If you've got a specific use case where you want lots of connections but they're mostly idle, then that seems sensible. (?)

[mkjpryor]

The problem I have is that the connections consumed by these long-lived requests stay ACTIVE (this is a form of long-polling, essentially). So the total number of objects that the thing can possibly handle is limited by the size of the connection pool. This thing needs to be able to handle potentially 1000s objects.

[karpetrosyan]

You can try increasing the size of the connection pool.