|nb.man(7)||NodeBrain Administrator Guide
Version 0.8.17 - August 2014
NAME | TOPICS | DESCRIPTION | GOAL | SCRIPTS | REACTIVE SCRIPTS | DIAGNOSTIC SCRIPTS | STATE AND EVENT MONITORING | THE NAME | FILES | ENVIRONMENT | DIAGNOSTICS | HISTORY | BUGS | AUTHOR | SEE ALSO
|nodebrain - NodeBrain State and Event Monitoring Agent|
Broad topics are covered by separate manual pages. If you are looking for something more specific
than the overview provided here, select the page of interest.
This manual page provides a high level overview of NodeBrain with references to other documentation.
These pages are provided primarily as a convenience for Unix and Linux system administrators
who may have to deal with NodeBrain in some way on their system, but do not necessarily want to build
a new application or write NodeBrain rules. For this reason, the NodeBrain man pages are collectively called the
NodeBrain Administrator Guide. Programmers may find it helpful as well. Never hurts
to have a few man pages to get oriented.
Programmers and system administrators who what to build a new application or just learn more, should reference documents available online at http://www.nodebrain.org.
The primary mission of NodeBrain is to be a light-weight event monitoring agent. In this respect,
NodeBrain fancies itself to be a tool for creating an event monitoring service like Apache is to web applications and
sendmail is to mail delivery. These tools perform a useful service in the background
without the user awareness associated with a web browser or mail client. Although the demand for event monitoring
is far less obvious than the demand for mail and web services, for those few unlucky folks who need
to monitor a stream of events, it would be nice to have a relatively simple, flexible, and free tool.
NodeBrain is at least free, and is trying to be simple and flexible---often
Secondly, a NodeBrain agent would like to share information with peers. If one agent detects a condition of interest, we want it to be able to report the condition to other agents. This requires network communication via TCP/IP, preferably with options for encryption and peer authentication. This is supported by the peer and message modules distributed with NodeBrain.
Hopefully you are now visualizing a network of agents monitoring local streams of events and reporting interesting conditions (generated events) to peers. But we need a way to get events into these agents other than communication with peer agents. Our real sources of information are external. For this, NodeBrain provides modules supporting multiple methods: servant scripts, FIFO files, queues, log file audit, SNMP traps, syslog, etc. It can use STDIN for a batch job or an interactive client, and in both cases communicate with peers. So NodeBrain is also a client that enables the creation of an event stream into an agent. As a batch job, or utility command, it provides a loosely coupled application interface to the agents. As an interactive client, it provides a primitive line mode user interface to agents.
Okay, so we have this tinker-toy component that can be used as an agent, an interactive client and a
utility program. So now we need a way to describe the conditions we want it to respond to.
We need a rule language. Guess what? NodeBrain is a rule language, and nb(1), is
the interpreter of this language. Just when you were getting comfortable with the idea that NodeBrain is
an event monitoring agent, we claim it is an interpreter. Well, actually any program that processes input is an interpreter
to some extent, but NodeBrain is a language you can use to write scripts.
Here is a script called HelloWorld.
Here we execute it.
Imagine a Perl interpreter (you can substitute your favorite interpreter) running as a daemon listening for code to interpret. That is a pretty flexible agent. Perhaps too flexible, but you could write code to impose constraints and structure on the incoming events and the types of actions the agent would be willing to perform for various users or applications. That would just be a Perl implementation of a server of some kind.
So why do we need a new language, this NodeBrain thing? Why do we not just build an agent using an existing interpreter? That seems like a good idea if you can easily do what you want. For example, some people use Prolog as a rule engine. That is nice because of the inferencing capabilities of Prolog. We are just trying to do something slightly different---we think. :) We may have to plead ignorance if proven wrong, but what we are trying to do with NodeBrain does not seem to be a good match for interpretive languages we are familiar with. We should also point out that NodeBrain is not a good substitute for what other interpreters do well. It is good, in the opinion of author, for a relatively specific problem space---the state and event monitoring space. All of our trusty favorite tools can be used in combination with NodeBrain in this problem space.
Monitoring applications typically watch for, and react to, predefined conditions.
Here is a simple reactive NodeBrain script.
define r1 on(a=1 and b=2) x=25;
For simplicity, we are illustrating the concept with an interactive script. In the execution below, user input is highlighted. Normally, assertions will come from external sources.
But wait. There is one more thing NodeBrain would like to add to its resume----diagnostic scripting.
You should think "baby expert system tool". NodeBrain has a diagnostic mode that
enables it to solve for unknown conditions. Here is a trivial and silly example.
define dog on(legs=4 and coat="fur" and tail);
define cow on(legs=4 and coat="leather" and tail);
define person on(legs=2 and (coat="skin" or coat="leather"));
define bird on(legs=2 and coat="feathers" and tail);
define fish on(!legs and coat="scales" and tail);
Let us call this diagnostic script "creature" and see what happens when we execute it. You should imagine that you are asking NodeBrain, with the help of the defined rules, to assist you in identifying a creature at the edge of a swimming pool. To solve this problem, NodeBrain, requests a little information from you. Your responses and NodeBrain conclusion are highlighted.
The most important thing to notice in the example above is that NodeBrain never asked us what kind of tail our creature has. It did not have to. It was able to select "person" and reject all other possibilities without knowing anything about the tail. This is how doctors and car mechanics diagnose problems without running every possible expensive test---well, if we are lucky. So the idea is for an agent to detect a condition based on rules and spawn a diagnostic script to determine the underlying problem. We want our diagnostic script to be intelligent enough to only run expensive tests when required. The diagnostic script can take action to resolve a problem, if in fact it determines there is a problem, or simply report its findings back to the agent rules for further decision making.
We have been talking about state and event monitoring without providing a definition. We have been assuming the
reader has an intuitive understanding that is close enough to what we mean. Now we will take the risk of
making it less clear by attempting an explanation.
The first statement below describes a state, and the second statement describes an event.
o The sun came out from behind a cloud.
Both statements below describe states, but taken in sequence they describe an event---the same event as the second statement above.
o The sun is out.
In NodeBrain, a state is represented by a unique combination of values in all the cells. Rules are used to monitor the state of specific sets of cells for "interesting" changes in state. A transition to an interesting state is recognized as an event of interest. In this scenario, we might argue that we are doing state monitoring and deriving events.
Now, let us think about it from the perspective of the source of this information. If the sun is behind a cloud, and you want to work in your internal office (no windows) until the sun comes out, you might ask a co-working with a window seat to let you know if the sun comes out from behind a cloud. When the co-worker comes to your office and says "The sun is out", the statement matches our description of a state. But you are both aware the co-working is really reporting an event, "The sun came out from behind a cloud." If you spent 15 minutes trying to finish up what you are working on and the co-worker returns to repeat the statement, "The sun is out", what does this mean? Does it mean the sun is still out, or that it went behind a cloud and came out again? After all, your request of the co-worker was to "let you know if the sun comes out from behind a cloud."
Suppose your request had been for the co-worker to come to your office every 15 minutes and tell you if the sun is out or if it is behind a cloud. Now the common understanding is that the co-worker (who is unreasonably patient with you) is reporting a state, and that you will detect the state change of interest. As an alternative, you might have an agreement that the co-working only report the state when it changes. In that case, the co-working is detecting the state change (an event) and reporting it as a new state, which you also recognize as an event.
As silly as this example is, NodeBrain supports these variations in the protocol between an agent and a peer (or other source of information). As NodeBrain commands, the statements above that report a state might look like this.
A NodeBrain rule, for monitoring this state might look like the following, assuming we elect to use the highlighted option above.
This rule would respond only twice to the following set of assertions because we are monitoring for a state change to sunny.
assert sunny; # action taken
The statement used earlier for an event report can also be expressed as a NodeBrain command. For this we use the verb ALERT instead of ASSERT.
Here is a rule to monitor for this event. We are now using IF instead of ON.
This rule would respond three times to the following set of alerts because we are monitoring for events, and this represents three events.
One final point here. Earlier we said an event is a change of state, and that we can monitor states to detect events. It is also true that we can monitor events to detect states.
define sunHidden if(type="SunWentBehindCloud") !sunny;
The details are not important here, the point is simply that NodeBrain supports both the notion of a state and an event, and can translate between the two. It is very important to be able to recognize a reported event as just further evidence of a known state. If you have a leaky roof, every drip is an event. However, you do not want to call the roof repair company every time you detect a drip---you want to call them once each time the roof transitions into a state of disrepair, as evidenced by your first detection of a drip. Notice we converted from an event (drip) to a state (roof is in disrepair) and monitored for a state change to recognized a higher level event (roof has transitioned to a state of disrepair).
|It may be obvious, but let us comment on the name, "NodeBrain". First the "Brain" part. Like almost every program, nb(1) is a decision maker. In the animal world, as best we understand it, this function is assigned to brains when not overridden by hearts. So we use the term "brain" to identify our program as a generalized decision maker. To counter the notion that we think it is really smart, versions are given code names like "Numskull", "Knucklehead" and "Dunce", at least in prototype versions. The "Node" part has two explanations. First, nb(1) is designed to be an agent in a network of peers. In this model, NodeBrain is a brain that operates as a node within a network of brains. Secondly, an executing instance of nb(1) is a collection of nodes, each with a set of rules and extended functionality provided by a node module. So NodeBrain is a brain composed of nodes in addition to being a node within a network of brains.|
NodeBrain does not actually care how rule files are named, but people
generally do. Here are some guidelines. We use *.nb as a generic
reference to a file containing NodeBrain commands, but other suffixes may be
used to identify the type of rules in a file.
|Ed Trettevik <email@example.com>|
|nb(1), nb.cfg(7), nb.syn(5), nb.mod(5), nb.lib(3), nbkit(1), nbkit-caboodle(7)|
|NodeBrain 0.8.17||August 2014||nb.man(7)|