Files
p2prc-home-server/docs/p2prc-home-server.org
2024-12-10 23:45:46 +00:00

120 lines
5.7 KiB
Org Mode

* P2PRC tracking home server
** Abstract
This is a libary on top of P2PRC to create easier abstractions
to setup servers on spare laptops at home and run applications
which can exibit a monolithic behavoir (Ex: Self hosted
jitsi, Indvidious). The motivation is to ensure users can
share self hosted infrastructure with friends and can design
a really simple replicatable instances of servers.
** Problems being addressed
1. Remembering ports mapped and stateful restart. Rather than the OS remember
information of the processes. The P2P network should also track it and
propogate it to other nodes in the network. This is to ensure when a process
fails there is a formally representable reason which can appropirate corrective
actions.
2. Arbitary instructions used to start a process and should be remembered to reproduce a
process if a node is down.
** What are attempting to do
The section below describes in detail the ideal startegy of addressing the problems stated
above.
*** Building Abstraction
We are proposting to build a library of the Haskell bindings which do the following:
- Creating a P2PRC process and allowing users to modify it's behavoir in functional manner with for instance
immutable datatypes.
- Allowing users to run this abstraction in a reproducable manner or constrainting it to run on precise
sets of machines.
- Should be able to represent the deployment strategy used (ex: Nix, Ansible etc...). This keeps the
deployment agnostic and not vendor specific.
- The intended idea of this is to only work with monolithic applications.
*** Building it for only home server users
- The use case is to ensure it's easier to setup a set of servers and shuffle P2PRC processes around
when needed. We intend to do this by ensuring a process is represented as type.
- Targets initially with the assumption that there is only 1 reliable root node to relay traffic to the make the setup
easier. Future plans will create a DNS layer which can make root nodes and P2PRC processes reproducable.
*** Linient rules
- This library is built for home users, therefore the library will be setup for P2PRC using the [[https://github.com/Akilan1999/p2p-rendering-computation/pull/115][unsafe mode]] which will ensure that all public
keys will be added to the [[https://www.ssh.com/academy/ssh/authorized-keys-file#:~:text=The%20authorized_keys%20file%20in%20SSH,keys%20and%20needs%20proper%20management][SSH auth list]] of all nodes avaliable in the network. While this is not secure it
comes with pre-assumption that all nodes joining the network are home servers and are needed to access each other
with limited rules (set by user permission from the OS side).
- We assume that all P2PRC processes run on a bare-metal machine without any virtualisation.
This is because most users run on machines which they own and we offload to the users
reponsibility to choose a tool that can support reproducable builds like [[https://nixos.org/][Nix]].
[[./P2PRC-Tracking-home-server.png]]
** Plan
Based on the [[https://github.com/Akilan1999/p2p-rendering-computation/tree/master/haskell][Haskell bindings]] to build an extended library (There are still changes needed
in the Haskell bindings before the addon can be built).
*** Handle more cases than machine name
The current haskell library seems to only support 1 type Machine name which is parseable from
the config file. We would need to handle conditions such as handling types such as [[https://github.com/Akilan1999/p2p-rendering-computation/blob/67165d4bf63d82794a1a264edf843295b727c226/config/config.go#L39][Baremetal]] and [[https://github.com/Akilan1999/p2p-rendering-computation/blob/67165d4bf63d82794a1a264edf843295b727c226/config/config.go#L40][UnsafeMode]].
On this proposed library they would both be set to true.
#+BEGIN_SRC haskell
instance FromJSON P2prcConfig where
parseJSON (Object o) = do
machineName <- o .: "MachineName"
pure
$ MkP2prConfig
{ machineName=machineName
}
parseJSON _ = mzero
#+END_SRC
*** Representation of a process
To design a data structure that can represent current set of properties:
1. IP address (string)
2. Port no (int)
3. Task name (string)
4. Task id (int)
5. Encoded deployment script (multi-line string)
6. Command to run deployment script (multi line string)
7. Command to kill deployment script (multi line string)
8. Status (boolean)
9. Domain name (string)
*** Representation of a node:
Follows the implemented Haskell data structure from the bindings:
#+BEGIN_SRC haskell
instance FromJSON ServerInfo where
parseJSON = withObject "ServerInfo" $
\ o -> do
name <- o .: "Name"
ip4str <- o .: "IPV4"
ip6str <- o .: "IPV6"
latency <- o .: "Latency"
download <- o .: "Download"
upload <- o .: "Upload"
serverPort <- o .: "ServerPort"
bmSshPort <- o .: "BareMetalSSHPort"
nat <- o .: "NAT"
mEscImpl <- o .: "EscapeImplementation"
custInfo <- o .: "CustomInformation"
#+END_SRC
*** Function expected to be built:
The following refers to the functions we would be building for our abstraction:
1. SpinProcess(<P2PRC process type>,<Node Type>)
2. KillProcess(<P2PRC process type>)
3. ProcessInformation(<process id> or <domain name>) returns <P2PRC process type>
4. ListProcess() return <P2PRC process type>
** Terms
P2PRC process: A p2prc process refers to potencially an instance (i.e web-server or any task). This process
stores information such as:
- Memory usage.
- Instructions to re-produce that task etc...
Root node: Refers to a node running P2PRC with a public IPV4 address to relay traffic through for nodes
behind NAT and can potencially even act a proxy for nodes(Used for domain name mapping) in the P2P network.