[issue20924] openssl init 100% CPU utilization on Windows

Changes by Christian Heimes : -- resolution: -> out of date status: open -> closed ___ Python tracker ___

[issue20924] openssl init 100% CPU utilization on Windows

Mark Lawrence added the comment: Is this still considered valid considering from https://www.python.org/downloads/release/python-279/ The entirety of Python 3.4's ssl module has been backported for Python 2.7.9. See PEP 466 for justification. ? -- nosy: +BreamoreBoy

[issue20924] openssl init 100% CPU utilization on Windows

STINNER Victor added the comment: Please also report the Windows version you are using. I don't see the answer to this question -- title: openssl init 100% CPU utilization - openssl init 100% CPU utilization on Windows ___ Python tracker

[issue20924] openssl init 100% CPU utilization

Antoine Pitrou added the comment: Thanks. It would be nice if you could investigate why the underlying NT socket is in non-blocking mode (AFAIK, this issue hasn't been reported before). Also, it would be nice if you could check with Python 3 as well. --

[issue20924] openssl init 100% CPU utilization

Roman O. Vlasov added the comment: Antoine, Martin, thank you for your replies. You was right: NT socket was not in blocking mode (in 1st case). I didn't knew how to determine socket mode in NT, so I explicitly set socket mode to blocking in _ssl.c before calling SSL_do_handshake() for test:

[issue20924] openssl init 100% CPU utilization

Martin v. Löwis added the comment: As Antoine says: if the socket is *really* in blocking mode, then SSL_do_handshake should block until the handshake if fully complete. It should not return SSL_ERROR_WANT_READ in this case. If you have managed to build Python for yourself to diagnose this

[issue20924] openssl init 100% CPU utilization

Roman O. Vlasov added the comment: To reproduce the 100% CPU load problem, we used a simple python TLS client on separate linux PC with Traffic control utility (tc): tc qdisc change dev eth0 root netem delay 5000ms After in-depth analyzing, we realized that _ssl.c behaves differently

[issue20924] openssl init 100% CPU utilization

Antoine Pitrou added the comment: How come OpenSSL returns SSL_ERROR_WANT_READ if the socket is blocking? Perhaps Windows returns a non-blocking socket for accept()? Have you tried with Python 3? -- nosy: +giampaolo.rodola type: resource usage - behavior

[issue20924] openssl init 100% CPU utilization

bob bob added the comment: loewis: This is correct stack: 0021ee80 10036d19 0176d259 0013 msvcr90!memset+0x3d 0021ee9c 10022a22 0021ef1c 0176d218 1001bf62 _ssl!SHA1_Final+0x49 0021eea8 1001bf62 0021ef0c 0021ef1c 0146 _ssl!OPENSSL_cleanse+0x92 0021eebc 10010319 0021ef0c 0021ef1c

[issue20924] openssl init 100% CPU utilization

New submission from bob bob: We have written a server on Python 2.7.6 that uses openssl 0.9.8y running on Windows box. Time to time our server freezes utilizing 100% CPU on ssl_init (C routine). The following is the process stack: 0021ee80 10036d19 0176d259 0013 msvcr90+0x3b35d

[issue20924] openssl init 100% CPU utilization

Antoine Pitrou added the comment: Have you tried reporting this to the OpenSSL project? This doesn't sound like a bug in Python. Also, 0.9.8y is the current latest version in the 0.9.8 line, and we are unlikely to switch the 2.7 Windows builds to OpenSSL 1.x, I think (but ultimately that's

[issue20924] openssl init 100% CPU utilization

Changes by STINNER Victor victor.stin...@gmail.com: -- nosy: +haypo ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue20924 ___ ___ Python-bugs-list

[issue20924] openssl init 100% CPU utilization

Martin v. Löwis added the comment: Antoine: init_ssl would be the entry function of _ssl.c, not code from OpenSSL. bob: It's doubtful though that the stack trace is right, since init_ssl doesn't call itself. Did you install the Python PDB files for 2.7.6 before obtaining this stack trace?