Guido van Rossum [mailto:guido@python.org] wrote:

...

...

...

It doesn't look terribly hard to make the SSL wrapper obey the timeout, by calling select() on the "raw" socket before calling SSL_write or SSL_read. I'm willing to try to get this to work.

That's cool. I don't know much about the SSL_read() API -- does it promise to read exactly the requested number of byte (unless the socket is closed)? Then a single select() before it is called may not be sufficient.

I don't know any more than you do -- I've never looked at the OpenSSL docs until today. It looks like SSL_read may indeed return fewer bytes than requested: http://www.openssl.org/docs/ssl/SSL_read.html

Hm, from that page it looks like the internal implementation may actually repeatedly read from the socket, until it has processed a full 16K block. But I may be mistaken, since it also refers to a non-blocking underlying "BIO", whatever that is. :-(

Yeah, I'm not really sure what that is.

...

But I still think that a single select() is OK. It seems to be working fine in my testing. The select() times out only if there has been no activity on the socket for the entire timeout period, which seems sufficient to me. I don't think it matters that SSL_read may return fewer bytes than requested. Maybe I'm missing something.

If you can see no CPU activity while it's waiting to time out, it's probably okay.

Yes, it seems to work -- no CPU activity and it times out exactly when it's supposed to.

...

Anyhow, I'll do some more testing and then if it still seems to work I'll upload a patch later today.

Great! Please assign to me so I head about it.

I just uploaded the patch and assigned it to you. - Geoff