Written by Rainer Gerhards (2007-12-17)
Rsyslog aims to be a drop-in replacement for sysklogd. However, version 3 has some considerable enhancements, which lead to some backward compatibility issues both in regard to sysklogd and rsyslog v1 and v2.
Rsyslog v3 is currently under initial development. Compatibility issues may be resolved, so be sure to check back often. Also, comments and suggestions are appreciated. Feedback right in time can now have a big impact on the route we take ;)
With v2 and below, inputs were automatically started together with rsyslog. In v3, inputs are optional! They come in the form of plug-in modules. At least one input module must be loaded to make rsyslog do any useful work. The config file directives doc briefly lists which config statements are available by which modules.
It is suggested that input modules be loaded in the top part of the config file. Here is an example, also highlighting the most important modules:
$ModLoad immark.so # provides --MARK-- message capability
$ModLoad imudp.so # provides UDP syslog reception
$ModLoad imtcp.so # provides TCP syslog reception and GSS-API (if compiled
to support it)
$ModLoad imuxsock.so # provides support for local system logging (e.g. via
logger command)
$ModLoad imklog.so # provides kernel logging support (previously done by rklogd)
A number of command line options have been removed. New config file directives have been added for them. Once we implement compatibiltiy mode, these options will return, but only if running in non-v3-native mode.
The -m command line option is ignored for the time being. There is no default mark period. If you need a 20 minute mark period you need to
$ModLoad immark.so # wherever this is
$MarkMessageInterval 1800 # 20 minutes
Is also no longer available. Use the $UDPSeverRun <port> config file directives. You can now also set the local address the server should listen to via $UDPServerAddress <ip> config directive.
The following example configures an UDP syslog server at the local address 192.0.2.1 on port 514:
$ModLoad imudp.so
$UDPSeverAddress 192.0.2.1 # this MUST be before the $UDPServerRun directive!
$UDPServerRun 514
"$UDPServerAddress *" means listen on all local interfaces. This is the default if no directive is specified.
Please note that now multiple listeners are supported. For example, you can do the following:
$ModLoad imudp.so
$UDPSeverAddress 192.0.2.1 # this MUST be before the $UDPServerRun directive!
$UDPServerRun 514
$UDPSeverAddress * # all local interfaces
$UDPServerRun 1514
These config file settings run two listeners: one at192.0.2.1:514 and one on port 1514, which listens on all local interfaces.
Please note that with pre-v3 rsyslogd, a service database lookup was made when a UDP server was started and no port was configured. Only if that failed, the IANA default of 514 was used. For TCP serves, this lookup was never done and 514 always used if no specific port was configured. For consitency, both TCP and UDP now use port 514 as default. If a lookup is desired, you need to specify it in the "Run" directive, e.g. "$UDPServerRun syslog".
klogd has (finally) been replaced by a loadable input module. To enable klogd functionality, do
$ModLoad imklog.so
A limited set of klogd command line settings is now supported via rsyslog.conf. That set of configuration directives is to be expanded. So far, we support:
$klogSymbolsTwice [on/off]
$DebugPrintKernelSymbols [on/off] # spits *a lot* of messages at startup
Either "FixedArray" or "LinkedList" is recommended. "Direct" is available, but should not be used except for a very good reason ("Direct" disables queueing and will potentially lead to message loss on the input side).
I have decided to move the todo list for v3 also into this document. After all, it is kind of a compatibility issue ;) But, more seriously, I think this is a good place to make sure nothing is forgotten. Remember that as of now, rsyslog v3 is in its early stage of development and so there definitely are some nits that will give you grief. This section here tells you what it is.
needs to be implemented. So far, only -c3 is supported. Everything else does not change anything but just provides a warning message. In short: there is no backwards compatibility yet.
Thread termination in mode 1 needs to be looked at. I begin to think that it is OK if we simply cancel the thread, because we already have a different cleanup function. Otherwise, there is a potential for a race condition after we unblocked the TermOK mutex but before select() is entered. If the signal then occurs, we may have a problem. However, I would like to create the other input modules first so that I have more working experience with termination requirements. Again, I think this will lead to changes in thread termination.
So far, this issue is resolved, but there is still some code present that needs to go away (right now disabled). I keep it in because I will make the final decision based on experience gained during creation of input modules.
In an ideal world, the modules should call back only to some well-defined and specifically exported interface functions. Right now, we do not have an ideal world and the input modules make some calls back into the core. This needs to be re-thought but is accepted for the time being.
As already said in "Call Encapsulation", the modules look like such, but are not really able to work stand alone. The worst in this regard are
But the others are not necessarily better. I have moved the code to the modules, because now at least the majority of code is where it belongs to. But there are still a large number of global variables, cross-function calls and the like. This needs to be solved when we try to claim a real modular design. That's not an easy task, but, hey, aren't we up for that...? ;)