NAME

logsurfer - process logfiles and perform certain actions

SYNOPSIS

logsurfer [ -c configfile ] [ -d dumpfile ] [ -D ] [ -f ] [ -F ] [ -l start_linenum ] [ -p pidfile ] [ -r start_regex ] [ -s ] [ -t ] [ -e ] [ filename ]

DESCRIPTION

Logsurfer is a program that works on single textlines from a logfile (or from standard input). It compares each line against regular expressions of a set of "rules" and performs certain actions that are defined for this match.

The logsurfer uses "rules" to define actions for messages (single lines). Messages may be collected in "contexts" which are stored in memory. Those contexts have their own default actions (e.g. for timeouts). At starttime you have to define a set of rules in the configuration file. While processing logmessages the logsurfer is able to dynamically add or delete rules, create or destroy contexts, invoke external programs for alerting people or to perform further automated actions etc.

A rule starts with a set of regular expressions. The messageline that is being checked has to match against the first regular expression but must not match the second expression. The string "-" can be used as the second regex if you don't need this "match not" expression. If the messageline meets these requirements the associated action is taken. Currently the following actions are allowed: ignore this message, execute an external program, pass this messageline as standard input to an external program, create or delete a context, generate a "report" or dynamically add a new rule. In addition to these actions rules may have a configuration to delete themself if another message (again specified by two regular expressions) is being processed by the logsurfer or if an optional timeout has occured. Matching of messagelines against rules is usually done until the first match occurs. This can be circumvented by specifying a "continue" keyword in the rule. In this case the program will continue to search for another match in the rule definitions.

A context is a set of messagelines that have been read by the logsurfer. Again the messagelines are specified by two regular expressions. Contexts can be used in "reports" to present a collection of messages associated with a specific action (e.g. all messagelines from one ftp session). Every context has a default action which is launched if either a specified timeout has occured or a configured limit (of linenumbers) has been reached. Default actions may be: ignore (drop) this context, execute an external program, feed all collected logmessages as standard input to an external program or generate a "report".

A report summarizes a number of contexts and pipes the result into an external program.

The normal procedure for each logline is: Store this line in all matching contexts (as defined by their regular expressions), if the self-destroy regex matches, too, then the default action of the context is executed and the context is deleted, find the first matching rule and execute the given action (continue rule matching if the rule has the keyword "continue" --- see logsurfer.conf(4) manpage), check for the stop-pattern of the rule and in case of a match delete the rule, check for timeouts of rules and delete the affected rules, check for context timeouts and execute their default action.

Errors (like broken configuration statements or memory problems) are reported on standard error. The syslog mechanism is currently not being used because this might result in a loop if the logsurfer is processing those syslog-messages. It is suggested to start the logsurfer and redirect the output (incl. error output) to another file. If necessary you might want to start a second logsurfer that is watching the output of the first logsurfer.

If the program exits the following steps are executed:

1): The status of all active rules and contexts is dumped to the dumpfile.
2): If contexts have timeout values (other than 0) then a timeout is send to those functions. This will result in launching the default actions of the affected contexts.
3): Rules and contexts are deleted from memory (cleanup). This is only important for debugging purposes and should have no effect.
4): If there are still external programs running the logsurfer will wait up to 20 seconds to catch them.
5): The logsurfer exits with a return value of zero (0).

OPTIONS

-c configfile: Specify another configuration file for the logsurfer. The default is "/usr/local/etc/logsurfer/logsurfer.conf".
-d dumpfile: Specify another dumpfile for the logsurfer. The logsurfer writes the internal information about rules and contexts to this file if the program exists or a special signal was sent to the process (see NOTES). The default is "/dev/null".
-D: Daemonize. Fork and run the logsurfer process in the background. Any output sent to stdin, stdout and stderr are redirected to the bitbucket.
-f: Normally the logsurfer will process the input once and exit at the end (after processing all the input lines). By using the "-f" option the program will enter a special "follow" mode if the end of file has been reached. In this case the program will sleep for one second before trying to read a new line. If a new line has been added to the end of the input, then this line is processed. Otherwise the program will sleep again for one second. This mode is similar to the "-f" option of the tail(1) command.
-F: Acts like the "-f" option, but follows the logfile across rotations by checking if the inode of the file has changed and re-opening the log.
-l start_linenum: Start logfile processing at the given linenumber.
-p pidfile: Write the processid of the logsurfer process to the given filename. This can be used to simplify sending signals to the logsurfer program (for example after shifting logfiles).
-r start_regex: Start logfile processing with the first line that matches the given regular expression.
-s: Normally the logsurfer prints some information to stderr if the program exists. If you specify the "-s" (silent) option, then those messages are not printed.
-t: When logsurfer exits, the -t option will instruct it to explicitly expire all contexts which have either a relative or absolute timeout configured. Without the -t option, contexts will not be timed-out and will be removed silently without triggering their actions.
-e: When Logsurfer starts reading the log file this option will instruct it to seek to the end of the file before scanning starts. Without this option, Logsurfer will start scanning at the beginning of the file.

FILES

/usr/local/etc/logsurfer/logsurfer.conf: default configuration file
/dev/null: dump of the rules and contexts

NOTES

This program may consume a lot of memory. Resource control should be done by the operating system (for example by setting limits). The program was designed to fall back to a defined state if some resources are not available.

Please do not run the logsurfer with a privileged userid (like root). You usually do not need the extra privileges. If your logfiles are not world-readable change them to be group readable and start the logsurfer with a non-privileged account that is a member of that group. Keep in mind, that the logsurfer might execute external programs (depending on your configuration file) which are not designed to run with higher privileges. The logsurfer program will print a warning message if it's started with root privilege (this feature can be disabled at compile time).

Be careful with external programs. The logsurfer is able to process any input (for example arbitrary long lines) but some other programs may crash or produce other inpredictable results if feeded with long lines.

The following signals have the specified effect when sent to the logsurfer program using the kill(1) command.

SIGHUP: Reopens the logfile (to be analyzed). Processing starts with the first line in this file. This can be used after "shifting" logfiles and creating new logfiles.
SIGUSR1: Dump the internal state of all active rules and the collected contexts to the dumpfile "/dev/null".
SIGTERM: Dumps the state to the dumpfile (see above) and exits.

BUGS

Quoting of regular expressions isn't that easy. If you use variables ($0 - $9) in dynamic rules, then you have to keep in mind, that certain characters (matching your regular expresion for that variable) may again contain special chars, that are interpreted as regex-parts. Example: If you use one regular expresion to match a hostname and you are using this again as part of a regular expression, then the dots "." separating parts of the domainname will match any char (not only the dot).

If the logsurfer has been compiled with the "sendmail flush hack" and the logsurfer exists, then there might still be some mails in the local sendmail queue (not flushed). Delivery of such mail maybe delayed until the next time sendmail flushes the queue.