On Sun, Apr 28, 2013 at 10:07 PM, Daniel Stenberg <daniel_at_haxx.se>
wrote:
>> Now, we run into issues where individual downloads clog the download
>> pipe because they take too long. This is [despite] the fact that
>> currently we allow up to 1000 parallel downloads.
>
> I don't understand. Why would single transfers "clog" a parallel
> pipeline? What is a parallel pipeline?

Say, there is a queue of 10,000 download requests, where downloads
taking one second and downloads taking five minutes are alternating.
Then, eventually there will be a point in time where the program is busy
only with the lengthy downloads.

> No, not if you're using a recent libcurl. Then the standard timeout
> features should work just fine.

That's great!

Hm, fact is that the program is hanging, somewhere deep inside. So far,
I only once got the chance once to attach `gdb` during a hang. It could
also be an issue with `libev` or another component.

We're using libcurl 7.24.0.
-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette: http://curl.haxx.se/mail/etiquette.html
Received on 2013-04-29