#!/usr/bin/perl -w ####################################################################################### # monitorLoad.pl # Written by: Brandon Zehm # # Purpose: # To monitor the 15 minute system load average and print # information to the system log. # # Technical Description: # Basically does a `cat /proc/loadavg` and parses the output. If the system load # is higher than the "thresholdWARNING" then it writes the current system load # to a file. If it is higher than "thresholdWARNING" on more than two consecutive # runs then it changes the output message to one of the warning messages. # # Changelog: # 01/20/2004 - v1.20 # - Require a -r option to run. # - Added a -l option for logging to a file # - Textual changes to most error messages. # - It now will try opening /proc/loadavg up to 10 times before giving up. # # 07/25/2002 - v1.03 # Jared Cheney jared_cheney@nonhp-am.exch.hp.com # modified it to write the system load to a temporary file the first time # the load is checked - then the next time it will compare against the first # value and only log an error if the value is rising after two consecutive # readings. 07/29 - update: some system calls were replaced with their perl # counterparts by Brandon. # ####################################################################################### use strict; ## Global Variable(s) my %conf = ( "programName" => 'monitorLoad.pl', "version" => '1.20', "authorName" => 'Brandon Zehm', ## Information about the author or owner of this script. "authorEmail" => 'caspian@dotconf.net', "hostname" => $ENV{'HOSTNAME'}, ## Used in printmsg() for all output. "syslogProgram" => '/usr/local/bin/syslog.pl', "run" => 0, ## Whether we should run script or not "debug" => 0, "stdout" => 0, "syslog" => 1, ## Syslog messages by default (--stdout disables this) "facility" => "USER", "priority" => "INFO", "load" => "", "previousLoad" => "", "comparisonFile" => "/tmp/monitorLoad.tmp", "logging" => '', ## If this is true the printmsg function prints to the log file "logFile" => '', ## If this is specified (form the command line via -l) this file will be used for logging. ## Set some thresholds, this just changes the message printed and the syslog priority. "thresholdWARNING" => '0.5', "thresholdERROR" => 1, "thresholdCRIT" => 2, "thresholdEMERG" => 5, ); ############################# ## ## MAIN PROGRAM ## ############################# ## Initialize initialize(); ## Process Command Line processCommandLine(); ## Open the log file if we need to if ($conf{'logFile'}) { if ($conf{'logging'}) { $conf{'logging'} = 0; close LOGFILE; } if (openLogFile($conf{'logFile'})) { quit("OS-ERROR => Opening the log file [$conf{'logFile'}] returned the error: $!", 1); } } ## Get the load average for (my $counter = 0; $counter <= 10; $counter++, sleep(10)) { quit("OS-ERROR => 10 consecutive errors while trying to read /proc/loadavg", 1) if ($counter >= 10); open(FILE, "/proc/loadavg") or next; $conf{'input'} = ; close FILE; last if $conf{'input'}; } ## Extract the value we need (take the average of all 3 system load values) $conf{'load'} = sprintf("%.02f", (( (split(/\s+/, $conf{'input'}))[0] + (split(/\s+/, $conf{'input'}))[1] + (split(/\s+/, $conf{'input'}))[2]) / 3)); printmsg("OS-DEBUG => The current load is: $conf{'load'}", 1); ## Compare the extracted value to what was previously measured and determine if the value is rising or falling my $directionOfFlow = ""; if (-f $conf{'comparisonFile'}) { ## Get the previous load average open(FILE, "$conf{'comparisonFile'}"); $conf{'previousLoad'} = ; close FILE; chomp $conf{'previousLoad'}; printmsg("OS-DEBUG => The previous load was: $conf{'previousLoad'}", 1); } if ($conf{'previousLoad'}) { if ($conf{'load'} >= $conf{'previousLoad'} ) { $directionOfFlow = "rising"; } elsif ($conf{'load'} < $conf{'previousLoad'} ) { $directionOfFlow = "falling"; } else { $directionOfFlow = "unable to determine"; } } else { $directionOfFlow = "unable to determine"; } ## Overwrite the old measurement with the new one into the compareLoad file. open(FILE, ">$conf{'comparisonFile'}"); print FILE $conf{'load'}; close FILE; ## Decide how urgent this is: ## EMERGENCY ## if ( $conf{'load'} >= $conf{'thresholdEMERG'}) { ## If the system load has risen or remained the same since the last check, then set the priority to EMERG if ($directionOfFlow eq "rising") { $conf{'priority'} = "EMERG"; } ## If the system load has dropped since the last check, then set the priority to WARNING, so a page is not generated. elsif ($directionOfFlow eq "falling") { $conf{'priority'} = "WARNING"; } ## If neither rising nor falling, assume that the script just started up or the previous value didn't pass the WARNING threshold, so only set as level WARNING else { $conf{'priority'} = "WARNING"; } } ## CRIT ## elsif ( $conf{'load'} >= $conf{'thresholdCRIT'} ) { ## If the system load has risen or remained the same since the last check, then set the priority to CRIT if ($directionOfFlow eq "rising") { $conf{'priority'} = "CRIT"; } ## If the system load has dropped since the last check, then set the priority to WARNING, so a page is not generated. elsif ($directionOfFlow eq "falling") { $conf{'priority'} = "WARNING"; } ## If neither rising nor falling, assume that the script just started up or the previous value didn't pass the WARNING threshold, so only set as level WARNING else { $conf{'priority'} = "WARNING"; } } ## ERROR ## elsif ( $conf{'load'} >= $conf{'thresholdERROR'} ) { ## If the system load has risen or remained the same since the last check, then set the priority to ERROR if ($directionOfFlow eq "rising") { $conf{'priority'} = "ERROR"; } ## If the system load has dropped since the last check, then set the priority to WARNING, so a page is not generated. elsif ($directionOfFlow eq "falling") { $conf{'priority'} = "WARNING"; } ## If neither rising nor falling, assume that the script just started up or the previous value didn't pass the WARNING threshold, so only set as level WARNING else { $conf{'priority'} = "WARNING"; } } ## WARNING ## elsif ( $conf{'load'} >= $conf{'thresholdWARNING'} ) { $conf{'priority'} = "WARNING"; ## rm the comparisonFile so that the $directionOfFlow doesn't get set unless the load is >= the ERROR threshold at least two times unlink($conf{'comparisonFile'}); } else { $conf{'priority'} = "INFO"; ## rm the comparisonFile so that the $directionOfFlow doesn't get set unless the load is >= the ERROR threshold at least two times unlink($conf{'comparisonFile'}); } ## DEBUG: printmsg("OS-DEBUG => $directionOfFlow is the current direction of flow", 1); ## Generate an output message if ( ($conf{'load'} >= $conf{'thresholdERROR'}) && ($directionOfFlow eq "rising") ) { $conf{'message'} = "OS-$conf{'priority'} => The 15 minute system load average is [$conf{'load'}] and has risen over the last 15 minutes."; } else { $conf{'message'} = "OS-$conf{'priority'} => The 15 minute system load average is [$conf{'load'}]."; } ## Syslog the message (or print it to stdout) printmsg($conf{'message'}, 0); ## Quit quit("",0); ###################################################################### ## Function: help () ## ## Description: For all those newbies ;) ## Prints a help message and exits the program. ## ###################################################################### sub help { print < Synopsis: Checks the 15 minute load average and logs it to the syslog daemon. Usage: $conf{'programName'} [options] Required: -r run script Optional: --stdout print messages to STDOUT rather than the syslog. --facility=[0-11] syslog facility (1/USER is used by default) -l enable logging to the specified file -v verbosity - use multiple times for greater effect EOM quit("", 1); } ###################################################################### ## Function: initialize () ## ## Does all the script startup jibberish. ## ###################################################################### sub initialize { ## Set STDOUT to flush immediatly after each print $| = 1; ## Intercept signals $SIG{'QUIT'} = sub { quit("$$ - EXITING: Received SIG$_[0]", 1); }; $SIG{'INT'} = sub { quit("$$ - EXITING: Received SIG$_[0]", 1); }; $SIG{'KILL'} = sub { quit("$$ - EXITING: Received SIG$_[0]", 1); }; $SIG{'TERM'} = sub { quit("$$ - EXITING: Received SIG$_[0]", 1); }; ## ALARM and HUP signals are not supported in Win32 unless ($^O =~ /win/i) { $SIG{'HUP'} = sub { quit("$$ - EXITING: Received SIG$_[0]", 1); }; $SIG{'ALRM'} = sub { quit("$$ - EXITING: Received SIG$_[0]", 1); }; } ## Make sure the path is sane $ENV{'PATH'} .= ":/bin:/usr/bin:/usr/local/bin:/sbin:/usr/sbin:/usr/local/sbin"; ## Fixup $conf{'hostname'} if ($conf{'hostname'}) { $conf{'hostname'} = $conf{'hostname'}; $conf{'hostname'} =~ s/\..*$//; } else { $conf{'hostname'} = "unknown"; } return(0); } ###################################################################### ## Function: processCommandLine () ## ## Processes command line storing important data in global var %conf ## ###################################################################### sub processCommandLine { ############################ ## Process command line ## ############################ my @ARGS = @ARGV; my $numargv = @ARGS; my $counter = 0; for ($counter = 0; $counter < $numargv; $counter++) { if ($ARGS[$counter] =~ s/^--stdout//i) { ## stdout ## $conf{'stdout'} = 1; $conf{'syslog'} = 0; } elsif ($ARGS[$counter] =~ s/^-v+//i) { ## Verbosity ## $conf{'debug'} += (length($&) - 1); } elsif ($ARGS[$counter] =~ /^-l$/) { ## Log File ## $counter++; $conf{'logFile'} = $ARGS[$counter]; } elsif ($ARGS[$counter] =~ s/^--facility=//i) { ## Facility ## $conf{'facility'} = $'; } elsif ($ARGS[$counter] =~ /^-r$/) { ## Run $conf{'run'} = 1; } elsif ($ARGS[$counter] =~ /^-h$|^--help$/i) { ## Help ## help(); } else { ## The Message ## quit("OS-ERROR => The option [$ARGS[$counter]] is unknown. Try --help", 1); } } ## If the user didn't use a -r print the help if ($conf{'run'} == 0) { help(); } return(1); } ############################################################################################### ## Function: printmsg (string $message, int $level) ## ## Description: Handles all messages - ## Depending on the state of the program it will log ## messages to a log file, print them to STDOUT or both. ## ## ## Input: $message A message to be printed, logged, etc. ## $level The debug level of the message. If not defined 0 ## will be assumed. 0 is considered a normal message, ## 1 and higher is considered a debug message. ## ## Output: Prints to STDOUT, to LOGFILE, both, or none depending ## on the state of the program and the debug level specified. ## ## Example: printmsg("ERROR => The file could not be opened!", 0); ############################################################################################### sub printmsg { ## Assign incoming parameters to variables my ( $message, $level ) = @_; ## Make sure input is sane $level = 0 if (!defined($level)); ## Continue only if the debug level of the program is >= message debug level. if ($conf{'debug'} >= $level) { ## Get the date in the format: Dec 3 11:14:04 my ($sec, $min, $hour, $mday, $mon) = localtime(); $mon = ('Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun', 'Jul', 'Aug', 'Sep', 'Oct', 'Nov', 'Dec')[$mon]; my $date = sprintf("%s %02d %02d:%02d:%02d", $mon, $mday, $hour, $min, $sec); ## Syslog the message is needed if ($conf{'syslog'}) { system( qq{$conf{'syslogProgram'} --facility=$conf{'facility'} --priority=$conf{'priority'} "$conf{'programName'}: $message" }); } ## Print to STDOUT always if debugging is enabled, or if conf{stdout} is true. if ( ($conf{'debug'} >= 1) or ($conf{'stdout'} == 1) ) { print "$date $conf{'hostname'} $conf{'programName'}: $message\n"; } ## Print to the log file if $conf{'logging'} is true if ($conf{'logging'}) { print LOGFILE "$date $conf{'hostname'} $conf{'programName'}: $message\n"; } } ## Return 0 errors return(0); } ############################################################################################### ## FUNCTION: ## openLogFile ( $filename ) ## ## ## DESCRIPTION: ## Opens the file $filename and attaches it to the filehandle "LOGFILE". Returns 0 on success ## and non-zero on failure. Error codes are listed below, and the error message gets set in ## global variable $!. ## ## ## Example: ## openFile ("/var/log/scanAlert.log"); ## ############################################################################################### sub openLogFile { ## Get the incoming filename my $filename = $_[0]; ## Make sure our file exists, and if the file doesn't exist then create it if ( ! -f $filename ) { printmsg("NOTICE => The file [$filename] does not exist. Creating it now with mode [0600].", 0); open (LOGFILE, ">>$filename"); close LOGFILE; chmod (0600, $filename); } ## Now open the file and attach it to a filehandle open (LOGFILE,">>$filename") or return (1); ## Put the file into non-buffering mode select LOGFILE; $| = 1; select STDOUT; ## Tell the rest of the program that we can log now $conf{'logging'} = "yes"; ## Return success return(0); } ###################################################################### ## Function: quit (string $message, int $errorLevel) ## ## Description: Exits the program, optionally printing $message. It ## returns an exit error level of $errorLevel to the ## system (0 means no errors, and is assumed if empty.) ## ## Example: quit("Exiting program normally", 0); ###################################################################### sub quit { my %incoming = (); ( $incoming{'message'}, $incoming{'errorLevel'} ) = @_; $incoming{'errorLevel'} = 0 if (!defined($incoming{'errorLevel'})); ## Print exit message if ($incoming{'message'}) { printmsg($incoming{'message'}, 0); } ## Exit exit($incoming{'errorLevel'}); }