Re: lib/41019: pthreads: high load when using varnish (locking/unlocking causes 98% WCPU)

[Michael Stapelberg <michael+netbsd%stapelberg.de@localhost>]

The following reply was made to PR lib/41019; it has been noted by GNATS.

From: Michael Stapelberg <michael+netbsd%stapelberg.de@localhost>
To: gnats-bugs%NetBSD.org@localhost
Cc: Christos Zoulas <christos%zoulas.com@localhost>
Subject: Re: lib/41019: pthreads: high load when using varnish
        (locking/unlocking causes 98% WCPU)
Date: Sun, 22 Mar 2009 15:50:59 +0100

 Hi Christos,
 
 * [22.03.09 15:38]:
 >  It is not a NetBSD issue:
 >      http://opengroup.org/onlinepubs/007908799/xsh/usleep.html
 Yes, I know, and I didnâ??t claim. I just wanted to emphasize that other 
systems
 do handle bigger values though itâ??s not correct.
 
 
 >  And in your fix you are not using nanosleep because you claim it has
 >  similar issues. I think it does not, so in reality the code should be
 Did you actually try it? Because if you loock at the usleep-implementation, it
 calls nanosleep with useconds * 1000, thatâ??d be 1000 * 1000 * 1000 in our
 example. Now, look at nanosleep(2) and youâ??ll see EINVAL for values greater
 than 1000 millionâ?¦
 
 >  #define U2S 1000000
 >      if (usec >= U2S) {
 >              unsigned int sec = usec / U2S;
 >              usec %= U2S;
 >  
 >              sleep(sec);
 >              usleep(usec);
 >      }
 Thank you, that looks like a good solution. Iâ??ve submitted it to the varnish
 bugtracker.
 
 Best regards,
 Michael