Socket leak

[fiddlerwoaroof]

Because DEX:REQUEST defaults to keep-alive t, this call can leak sockets. This has caused issues with a long-running process eventually running out of file descriptors.

https://github.com/fukamachi/aws-sdk-lisp/blob/6d1f66e88764e6226010e98289cf4c35e8f896c9/api.lisp#L44-L52

[ccqpein]

Looks like dexador keep-alive default is nil now. Is this still a issue anymore?

https://github.com/fukamachi/dexador/blob/master/README.markdown#functions

[cdmojoli]

Performance is poor with :keep-alive nil.

From my neck of the woods I get ~650ms per API call without keep-alive and a much better ~176ms with it.

This is a huge difference.

If :keep-alive t is leaking sockets, it is an error in dexador or the code using it.

I will enable it in my installation, and will report if the leak bites me.

[cdmojoli]

FYI, I have been testing :keep-alive t :use-connection-pool t using tcpdump, netstat and listing open file descriptors at /proc/XXXXX/fd and I do not find a leak. Everything is behaving as expected. Granted, this is not a proof of correct behavior.

I am using commit b5ae82b, with SBCL 2.5.0 over Ubuntu 22.04.5 LTS (x64).

[cdmojoli]

As far as I have been using this, there is no socket leak, I encourage applying pull request #39.

BTW, I have found this enlightening comment in DEXADOR code .../dexador-20241012-git/src/backend/usocket.lisp#(defun-careful request):

Have to be a little careful with the fifth value stream we return -- the user may be not aware that keep-alive t without use-connection-pool can leak sockets, so we wrap the returned last value so when it is garbage collected it gets closed. If the user is getting a stream back as BODY, then we instead add a finalizer to that stream to close it when garbage collected