Recipe 9.40 Parsing the Process Accounting Log

9.40.1 Problem

You want to extract detailed information such as exit codes from the process accounting log.

9.40.2 Solution

Read and unpack the accounting records with this Perl script:

use POSIX qw(:sys_wait_h);
use constant ACORE => 0x08;      # for $flag, below
$/ = \64;                           # size of each accounting record
while (my $acct = <>) {
        my (        $flag,
                $comm) =
                         unpack("CxS3LS9x2LA17", $acct);
        printf("%s %-16s", scalar(localtime($btime)), $comm);
        printf(" exited with status %d", WEXITSTATUS($exitcode))
                if WIFEXITED($exitcode);
        printf(" was killed by signal %d", WTERMSIG($exitcode))
                if WIFSIGNALED($exitcode);
        printf(" (core dumped)")
                if $flag & ACORE;
        printf("\n"); }

9.40.3 Discussion

Even the dump-acct command [Recipe 9.39] misses some information recorded by the kernel, such as the exit code. This is really the status that would have been returned by wait(2), and includes the specific signal for commands that were killed. To recover this information, attack the accounting records directly with a short Perl script.

Our recipe shows how to read and unpack the records, according to the description in /usr/include/sys/acct.h. When we run the script, it produces a chronological report that describes how each process expired, e.g:

Sun Feb 16 21:23:56 2003 ls          exited with status 0
Sun Feb 16 21:24:05 2003 sleep       was killed by signal 2
Sun Feb 16 21:24:14 2003 grep        exited with status 1
Sun Feb 16 21:25:05 2003 myprogram   was killed by signal 7 (core dumped)

9.40.4 See Also

acct(5). The C language file /usr/include/sys/acct.h describes the accounting records written by the kernel.

    Chapter 9. Testing and Monitoring