Utterly surreal OS bug

fivemack · 2011-01-19, 00:03

Code:

nfsslave2@cow:/home/nfsworld/2+956/linalg$ ps -L 11743
  PID   LWP TTY      STAT   TIME COMMAND
11743 11743 ?        Rl   9645:56 ../../msieve-svn/trunk/msieve-td130 -v -nc -t 4
11743 21509 ?        Rl   8653:20 ../../msieve-svn/trunk/msieve-td130 -v -nc -t 4
11743 21510 ?        Rl   9014:08 ../../msieve-svn/trunk/msieve-td130 -v -nc -t 4
11743 21515 ?        Rl   9043:40 ../../msieve-svn/trunk/msieve-td130 -v -nc -t 4
nfsslave2@cow:/home/nfsworld/2+956/linalg$ ps 11743
  PID TTY      STAT   TIME COMMAND
11743 ?        Rl   24260266:53 ../../msieve-svn/trunk/msieve-td130 -v -nc -t 4

(and whilst the times reported by -L go up at a rate of about one minute per minute, the one reported by 'ps' on its own stays at that ridiculous value)

I don't think I need to do anything about this, but it's most peculiar. Process 11743 also doesn't appear in 'top'.

xilman · 2011-01-19, 07:42

Quote:

Originally Posted by fivemack

Code:

nfsslave2@cow:/home/nfsworld/2+956/linalg$ ps -L 11743
  PID   LWP TTY      STAT   TIME COMMAND
11743 11743 ?        Rl   9645:56 ../../msieve-svn/trunk/msieve-td130 -v -nc -t 4
11743 21509 ?        Rl   8653:20 ../../msieve-svn/trunk/msieve-td130 -v -nc -t 4
11743 21510 ?        Rl   9014:08 ../../msieve-svn/trunk/msieve-td130 -v -nc -t 4
11743 21515 ?        Rl   9043:40 ../../msieve-svn/trunk/msieve-td130 -v -nc -t 4
nfsslave2@cow:/home/nfsworld/2+956/linalg$ ps 11743
  PID TTY      STAT   TIME COMMAND
11743 ?        Rl   24260266:53 ../../msieve-svn/trunk/msieve-td130 -v -nc -t 4

(and whilst the times reported by -L go up at a rate of about one minute per minute, the one reported by 'ps' on its own stays at that ridiculous value)

I don't think I need to do anything about this, but it's most peculiar. Process 11743 also doesn't appear in 'top'.

I saw exactly the same thing when doing the LA for a c178 recently. I believe that the author of top(1) didn't expect a single process to clock up so much time and a variable overflowed. I'm almost, but not quite, tempted to check the source of top(1) to find out. The underlying bug is, presumably, in the code that implements the /proc/*/stat interface.

Paul