UNIX for Advanced & Expert Users

Hi Folks


I have been debugging a script that is called every thirty seconds. Basically it is doing a ps, well two actually, one to file (read by the getline below) and the other into a pipe. The one into the pipe is: -

V_SYSVPS=/usr/sysv/bin/ps

Code:

$V_SYSVPS -p$PIDLIST -o$PSARGS 2>/dev/null | \
nawk -v tmpfile=$TMPFILE -v ta=$(vdate -t -d 197001010000.00 + $CMDTIME) \
                                  -v h=$V_HOSTNAME -v tr=$((COUNT * V_AUTO_MON_FREQUENCY)) -v argfield=$V_AUTO_MON_ARGFIELD \
                                  -v s="MON:1,psm,$V_AUTO_MON_START,$CMDTIME" \
                                  'BEGIN      {
                                                  while (getline < tmpfile)
                                                          l[$1] = $0
                                              }
                                              }'

The problem is that the sysv ps intermittently takes longer than 30 seconds to come back, I have echoed the date to file either side of it and it is definitely blocking on the write to the pipe. I have also trussed the nawk and it hangs on the read from the pipe.

I spotted that the getline does not check for EOF or close the file but it still blocks on the write to the pipe after I fixed it.

Trussing the ps itself shows it goes to the /dev directory for some reason which contains 29000 files and does a statx on each. However most of the time it deals with that in under 2 seconds.

Can anyone tell me if there is anything about the sysv ps that can explain this 30 delay? (Sometimes it takes 20 or so seconds but this completes in time)

Apparently this is using both ps commands because they could not get all the info they wanted from just the one version.