Hello,

this is a new program, which demonstrates the problem.
(Although I dont know if libcurl should be used this way)

What does the program do:

My idea is to not start 1000 requests in parallel, but use a set
of requests "on the fly".
Evertime a request has finished, it should add a new one,
until 1000 requests are done.

On my tests I dont get a predictible behaviour. Sometimes
it works without timeouts, then the first call directly timeouts.

It looks like, that if I return the size of the copied bytes in dev_null
The request is not set to the state CURLE_OK and I cannot remove
it from the msgqueue.

regards,

Martin