lee/hpr-knowledge-base
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Initial commit: HPR Knowledge Base MCP Server

Browse Source 
- MCP server with stdio transport for local use
- Search episodes, transcripts, hosts, and series
- 4,511 episodes with metadata and transcripts
- Data loader with in-memory JSON storage

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
Lee Hanken
2025-10-26 10:54:13 +00:00
commit 7c8efd2228
4494 changed files with 1705541 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

184
hpr_transcripts/hpr0279.txt Normal file
View File

@@ -0,0 +1,184 @@
Episode: 279
Title: HPR0279: cfengine
Source: https://hub.hackerpublicradio.org/ccdn.php?filename=/eps/hpr0279/hpr0279.mp3
Transcribed: 2025-10-07 15:26:29
---
music
Hello everybody and welcome to today's episode which is on CF Engine and from Wikibedia
CF Engine is a policy-based configuration management system written by Mark Burgess
from Oslo University. Now I first come across CF Engine when I started working at Shuburg
Phyllis. A colleague of mine, Ian Seldom, was of great help in explaining not only how
it works but also the philosophy behind it so I thought this would be an excellent opportunity
to get Ian around the table here so that we can explain what CF Engine is and what it does.
So welcome to you. Hello, how long have you been using CF Engine?
I first use CF Engine back in the 90s. Probably about 98 or 99. Has it changed a lot?
Not particularly. No, I mean obviously there's a great, there's the normal sort of function creepy
get where you get a new features coming but the fundamental idea behind CF Engine is remained
pretty much the same since the beginning. So what can you do with it? Oh, maybe we need to
wind back a bit. Why did we ever need CF Engine? We used to have IT departments and in our IT
department we would have a big mini computer. I don't know, a honey well or a DPX20 or whatever.
One machine, one operating system, 60LG uses all on the O Green screens. Then we came into clients
a client sort of models of computing where we suddenly had and the explosion of the internet
where we had hundreds of hosts suddenly and trying to manage our 100 hosts to make sure that
they're all the same, that the host files are the same, that the DNS configurations are the same,
that the Apache web service configurations are the same. It's just a nightmare and it's not what
you want to be using your brain as for. So I think what most system administrators is quite a bit
a clue at the time and at least those quite a bit of clue. Decided to do was to start writing all
sorts of R-Sync scripts and RSH scripts and SSH scripts which went out and tried to manage these
systems and you basically got two big disadvantages here. Firstly every system admin takes a different
approach. Somebody's written one thing in Perl, somebody else has written another thing in Shell
or whatever and so there's no real consistency to how you approach the problem. In the second problem
is that you run your magic installation script but one week later one of your colleagues is going
to change something and your beautiful configuration is no longer consistent. In fact your systems
are diverging and after one or two years they'll become sufficiently divergent. There is
starts to cause your problems. They actually will cause you downtime or system problems or
inaccurate data or information or so. Any other things that we hear in IT just don't want to have.
I come from the school where you take a server, you put in a devian CD, you do a minimal install,
you install the packages you want and then you manage that system by pushing out scripts via
SSH. What's wrong with that idea? Well there's nothing in Perl saying much wrong with the
providing one you're the administrator so there isn't anybody else doing it some other way
and doing it in a way you're not expecting or putting the files in some other place or editing
them by hand so it's that you actually go and regress your previous changes when you run your
script. I'm familiar with this concept. So I mean great if you've got 50 hosts and you manage
them yourself when you're very controlled and you're like nobody else in that system will probably
work. Of course the second problem you've got is that in the meanwhile you've probably been
writing some little web page somewhere or in the modern day parlance a wiki page or some
documentation in your boss's favorite Microsoft Word format or whatever and after about 10 days
that documentation is also completely out of date and it's again it's useless. So CFNG
offers you a way of doing all these things you used to previously do with your ad hoc scripts
but in a structured meta language which means everybody does it the same way it's very very
flexible and it offers a high degree of self documentation which is incredibly useful so
somebody who comes after you can see what you've done and how you've done it. I mean
the value of that is currently understated in actual fact. And the second thing is of course
it has self-healing so your policy within your self CF engine rules that is going to be
continually continuously apply to the service which belong to the class. Okay now a service
which you've defined. Can I just stop you there because I think we need to go back and explain
to everybody what CF engine can do. So I'll give you a rundown from what I found on the web.
You can check file permissions, ownership and fix them. It can restart fields, demons,
it can install software, it can edit files, it can execute commands, it can configure networks,
it can compressively files and it can so essentially you can make scripts that can do all these
things and basically manage it. You can define rules such which do these things. Okay and theoretically
you can do anything with CF engine because ultimately there is always the action type shell commands
and in action type shell commands you can remember what the heck you like. It requires a slightly
different way of thinking and the thing you shouldn't forget is of course all of these things can
be combined. So quite often you're not looking at doing a single thing. So for instance you have a
web server. Now your web server first you have to create some directories, you have to set the
permissions correctly on the directories, you have to send in a web server configuration file.
When the web server configuration file updates you want to restart the demon. You want to make sure
the illusion is certain place and the elogs get rotated each evening and when after a certain time
they get deleted. So in actual fact what you do is using something like seven different functionalities
of CF engine in a single workflow. So if I change these directories or if I update or write
update the web server configuration I need to restart the demon. If the demon is no longer running
because it's crashed for instance I need to restart the demon. If it is the first quarter after
midnight each day I need to go and rotate the logs and restart the demon. At four o'clock in the
morning I need to go and delete any files which are more than 222 days old or whatever. And all of
that is just simply done for you. These are the sort of things which classically kill systems
that disks get full because like files I get rotated properly. You've got a farm of 14 Apache
web server instance and each web server configuration is slightly different than the other one.
Your permissions are maybe inconsistent so one of your web servers may have very very liberal
permissions which allows your server to be hacked or content you prefer wasn't there to appear there
or obviously it doesn't allow your users to actually upload files which maybe they should be
allowed to. So they have a 1 in 14 chance of failing or something. With this sort of approach those
sort of problems that disappear. It installs on basically all the versions of Unix that there is
and through single init you can install it on Windows. Correct. There is actually a native
compile for Windows but it requires some patching and it's a bit messy.
I've got three servers. I want to manage two one as an admin server.
Oh well yeah if Mark was here Mark the author he probably wouldn't like I'm going to describe
this because the way we use CF Engine is to use it in a centralized way. So we have a policy
host which defines how everything else is going to look. Of course CF Engine doesn't have to work
that way. It can have multiple policy hosts. It can be very very distributed.
That I'll leave as an exercise to the reader. There are some links I believe which come with this
broadcast with people can read more on that but we'll look at it from the centralized point of view
from well because effectively we're a data center where I manage services companies so we're not
interested in giving people freedom. We're interested in making sure things are right and
to keep working. The centralized model works very well for us and so effectively you're going to
define a policy host or multiple policy hosts for redundancy. Some of our customers require
that we're able to work in across geographical distance into separate data centers etc.
Within this you then have a repository. Repository can contain your files and your rules which
you're going to use. So let's see if I can just distribute. That's just a directory tree. It is
just a directory tree which within your server configuration you've said I will allow CF Engine
clients to access files within these direct trees. Put your control files in there and
the other files come in, do the other clients come in and get them or do we push them out or
how does that work? It's on to not a pool basis so effectively this is one of one of Marx's key
things is that the idea is that the server never does something to a client without the client's
permission. So there is no there's no idea of duress or force or I'm the boss and this is how it's
going to be. The clients ask for the new configuration and so and then they and and when they get
the configuration they then run the rules locally and that may involve getting some files of the
policy host it may it may involve brunison edit files rules or changing some permissions or
updating the local password file any number of things. This is actually the point that got me in
the old presentation. When we were talking about a presentation here just for the listeners
company flew in Mark and he got the presentation here but it was this idea that got me from my old
old view of putting the CD in the server it was they you know they he had a diagram of a black hole
and he had a diagram of of amounted and the idea was that the CF engine way is you're always
going to try and get to the to the center so like in a pinball machine you throw the pinball
up and no matter what happens your server will return to a well-known configuration.
Yeah they're close to the standard way of which would be you have a blind man at the base of
the mountain and you say okay you walk 10 feet this way you go up you cross the bridge you go up
the canyon and if anything happens on if anything changes down through time if the path you know
an avalanche comes and you can no longer pass the path your server is no longer in the same state
that kind of got me that you're never going to have a perfect server. No you can't I mean
that that's the whole point there are four or five good configuration management tools out
the most of them at GPL so that they're entirely free to use and for some some people some
approaches may work better than CF engine. CF engine is all about accepting that there is no
perfect state. Your owe is moving towards conversions and eliminating divergence but the
idea even if you actually build systems from from images and you make them read only
yeah immutable file systems you still got differences you're still going to
differences within your cashiers within your memory footprint within your within your logs
so the systems are never identical so the whole point of controlling systems with CF engine
is identifying what matters and making sure that that is controlled and accepting that the rest
doesn't need to be and to something great actually accepting it doesn't matter yeah it simply
doesn't matter. This is what we wanted to do yeah that's right and there's a lot of things you
just really don't care about in the end and that's yeah that's the CF engine the the CF engine
also if you take convergence has been the primary idea what's brilliant about CF engine of course
is if you already have a farm even if it's a farm of hundreds of servers you can begin very small
with conversions you can say I'm going to write one rule which just simply controls the host file
for all my systems and then you've got you know if you have some sort of imaginary goal of 100
percent convergence you've just gone half a percent to which you want to be you've got a second
rule which you go and make sure all the users all these system administrators public keys are
putting the right place and you then two and a half percent closer to you you're black hole
okay do we not have a security problem where we've got all these machines being controlled by one
central location what's what's the deal here with that yeah of course I introduced this is a
security implication because it means you have to make sure you protect your CF engine policy
host very carefully but yeah and you also have to balance I mean if you if you do not know the
state of of machines within your park and remember a company like Schubert Phyllis we have literally
thousands and thousands of servers working after here I mean we're not going to log on to all of
these servers every day and in actual fact we don't want to we're always aiming for a point we're
a good a good wall behave servers when you never log on to so natural fact is the other way around
some systems like to sit there with the multi-tail watching watching log files go past every day and
logging on to all the machines and give them a little stroke we don't want to do this we just want
them to just work and so if if we go back to the security implication a non-managed system is
of course a huge security problem because you don't know what's there yeah you could you could
if you've got a thousand hosts and ten of them have been in some way hacked owned whatever you
want to call it you could be blissfully unaware of that maybe until you get your bandwidth bill or
something you know whereas whereas of course for the policy host you provide your policy host is
adequately secured man you've made sure that you know who who's allowed that who's using get who
can edit the files you make sure that you've got a CVS control and so on on the on the on the
fire revisions you know exactly who's changed what you're in a far stronger position but you are
indeed introducing a centralized is that machine gets owned then you're in a lot of trouble
but you only have to watch one machine now okay so we've got a server file and we want to control
that what are the steps it's greenfield site okay here's we're going to define a policy host
within our policy host that's where our CF engine server is going to run that's where we're
going to keep our rules files that's where we're going to keep our repository of files we're
going to distribute so that's where all the other servers are going to enter this is the magic
centralized server and I stress again as I said earlier of course this is not the only way of doing
it this just happens the way yeah the idea here we have links and shounals anybody who's
interested can go there and and find out the other ways of doing it but yeah it is one of the
joys of CFNs now that it doesn't tie you down to one model you can use spoke and hope you can
use centralized you can use distributed it it really doesn't matter okay we've got our
CF server derelict yeah that's right and I mean the simplest configuration we're basically going
to go run a CF agent on all of our hosts we wish to control and the CF agent's going to
addy specific timings of what which we which we control log onto the policy host grab the
grab the files execute the rules and in the end in the modern CFNs parlance each policy
statement or rule is considered a promise and it's going to then go through these files and
check each promise and then at the end it'll even tell you that a 100% of promises have been
capped or maybe 90% of promises being capped and 10% needed to be fixed so and the promise can
be anything the most simple thing is going to say my et cetera shadow file must have permissions of
400 yeah and if you're et cetera shadows file has permissions of 644 which would be rather
unusual but yeah it did it's going to change from about to 400 and that's going to be reparing
the broken promise okay and thanks very much it's a lot to digest but again all the links
for this is in the show notes and hope to be pulling in here again to do some more
episodes well that that's great and thanks for your time
thank you for listening to Hack or Public Radio
hpr sponsored by caro.net so head on over to c-a-r-o dot anything for all of those