Friday, September 28, 2012

rsyslog string array comparisons

In order to aid some string comparisons, I have just introduced constant string arrays in rsyslog. Look at this construct:

if $msg contains ["event12", "event15","event19"] then
    /home/rgerhards/logfile

This will match if $msg contains any of the three strings given inside the array, so it works like an implicit "or". I have seen this in a couple of customer configs and I am confident that this will ease rsyslog configuration. In later releases, it will also provide room for performance optimizations.

This feature will be part of 7.1.7.

[edit] Starting with rsyslog 7.3.6, string array comparisons are notably faster, especially if the array is large. At most, there are log(n) comparisons required (where n is the array size).

rsyslog's new input() statement: a quick look

I have just completed the new input() config statement. This concludes the major part of the new config format for v6 (v7 will also support an enhanced ruleset() statement). I wanted to give you some quick ideas of how the new format looks in practice. So I have created a small test rsyslog.conf with the old-style directives commented out and followed by the new style ones. I guess it is pretty self-explanatory. Here it is:

#$ModLoad imfile
#$inputfilepollinterval 1
module(load="imfile" pollingInterval="1")

#input(type="imuxsock" )
module(load="imuxsock" syssock.use="off")
input(type="imuxsock" socket="/home/rgerhards/testsock")

$ModLoad imfile
module(load="imfile")
#$InputFileName /tmp/inputfile
#$InputFileTag tag1:
#$InputFileStateFile inputfile-state
#$InputRunFileMonitor
input(type="imfile" file="/tmp/inputfile" tag="tag1:" statefile="inputfile-state")

module(load="imptcp")
#$InputPTCPServerRun 13514
input(type="imptcp" port="13514")

module(load="imtcp" keepalive="on")
#$InputTCPServerSupportOctetCountedFraming off
#$InputTCPServerInputName tcpname
#$InputTCPServerRun 13515
input(type="imtcp" port="13515" name="tcpname" supportOctetCountedFraming="off")

#$UDPServerRun 13514
#$UDPServerRun 13515
input(type="imudp" port="13514")
input(type="imudp" port="13515")

Monday, September 24, 2012

rsyslog plugin for structured Linux kernel logs

Thanks to Milan Bartos, there now is imkmsg, which supports structured kernel logging. This module will become available as part of rsyslog 7.1.5 later this week. The module was created with the help of Red Hat as part of the lumberjack effort for a better and structured logging infrastructure.

Note that the rsyslog v7 engine supports numerous enhancements for simple processing of lumberjack strucutred logs. So imkmsg is a natural fit for that version. As Milan told me, there is still room for improvement inside the module and some nits may exist and be fixed within the next weeks. However, we would appreciate feedback from early adoptors.

Monday, September 17, 2012

setting variables in rsyslog v7

Starting with rsyslog 7.1.3, variables can be manipulated inside rsyslog.conf. Note that the native message properties (like $msg) cannot be modified, but the CEE/lumberjack-type properties (those $!name variables) can. As variable modification is primarily an aid for template generation and modification, we do not consider this a restriction, as native message properties can simply be copied to the CEE tree if this is needed.

Note that CEE/lumberjack properties, as implemented in rsyslog,  can be hierarchical and levels are delimited by the bang sign (based on lumberjack recommendations). So "!uid" is the uid field in the CEE root, whereas "!usr!uid" is the uid field inside the usr container. Nesting can be as deep as desired. Currently, all parts of the CEE tree can be accessed. In later versions, this may require the setting of a global option. If that will happen depends on the feedback we receive.

So: how does it work?

To set a variable, simply use
set varname = expression;
Where expression can be an arbitrary complex expression, just like in an "if" statement. Note the semicolon at the end: this is unfortunately necessary and a diversion from the other config statements. However, this is the price we need to pay to remain upward compatible with the old style config format.

Concrete examples:

set $!usr!level2!var1 = "test"; 
set $!usr!level2!var1 = $msg; # update variable with native MSG field
set $!usr!level2!var2 = 1+1; # set !usr!level2!var2 = 2
set $!usr!level2 = $fromhost; # invalid

The last example is invalid and will not be executed because it tries to replace a complete container (!usr!level2) with a single value. This is rejected and not executed. Note that this problem is not detected during config read time but rather during execution time (in less trivial samples, it can not be reliable detected without execution).

Note that string concatenation is currently NOT supported, but it will be soon in the next releases. Also, full JSON containers cannot yet be copied. If this is tried to, the resulting variable will receive a string representation, which most probably is not what you wanted (and you will get a different result in future versions).

There is also an accompanying "unset" statement to remove a variable that is no longer needed. This is primarily meant to restructure a CEE container. It's syntax simply is:

unset varname;

Again, note the semicolon at the end. A concrete example is

unset !usr!level2!var1;

which removes a single element. But full containers can also be removed:

unset !usr!level2;

Note that all variables are assigned to the message currently being processed. There currently is no way to set global variables (but this may become available based on feedback I see).

Tuesday, September 11, 2012

BSD-Style blocks will go away in rsyslog v7

Rsyslog supports BSD-style blocks since ages. They were a pretty handy tool to group actions together that should act only on remote hosts or log messages from specific programs. However, the v7 config system with its full nesting capabilities provides a much better - and easy to use - way to specify this. If both systems are mixed, the problem is that BSD-style blocks can be used to violate the nesting structure (examples below). Also, support for them adds a lot to rule engine code complexity. And finally, they are also very seldom used, few users even know they exist.

As a result, I have decided to drop support for BSD-style blocks in rsyslog v7 and above. A poll on the mailing list a week ago did not make anybody speak up against that change. So I assume none is hurt. This is especially the case as the conversion of BSD-style blocks to nested config is a very easy one.

Let's look at this example:


!prog1
*.* /var/log/prog1.log
*.* /var/log/prog1again.log
!prog2
*.* /var/log/prog2.log
*.* /var/log/prog2again.log

This code can very simply be replaced by:
 if $programname == 'prog1' then {
   /var/log/prog1.log
    /var/log/prog1again.log
}
if $programname == 'prog2' then {
   /var/log/prog2.log
   /var/log/prog2again.log
}

And if you prefer the more powerful action statments (probably not so much benefit for this use case...), you can write:
 if $programname == 'prog1' then {
   action(type="omfile" file="/var/log/prog1.log")
    action(type="omfile" file="/var/log/prog1again.log")
}
if $programname == 'prog2' then {
   action(type="omfile" file="/var/log/prog2.log")
   action(type="omfile" file="/var/log/prog2again.log")
}
I expect that usually these easy cases happen. HOWEVER, if I had kept support for BSD-style blocks, one could configure things like this:
!prog1
if $msg contains 'test' then  {
                action(type="omfile" file="/var/log/somefile")
                !prog2
                mail.* :mmjsonparse: 
                               & action(type="omfile" file="/var/log/somefile2") 
               !prog3

                               & ~
                !prog4 
                if $msg contains 'test2' then
                               /var/log/logfile
                else
                               /var/log/logfile2
}

Can you make out the actual nesting structure of this config? When, for example, programname needs to be "prog3" and what happens then? IMHO this combination can cause considerable user confusion and frustration. As such, I made a sharp cut and removed it.

My apologies for those that need to do the manual conversion. I am sure the time is well-invested in the long term.

Busy at the moment...

Some might have noticed that I am not as active as usual on the rsyslog project . As this seems to turn out to keep at least for the upcomi...