dgryski / boogie Goto Github PK

We want to be able to introspect the servers, these are the easiest ways. We will probably need to add extra commands for getting at the state of the system, but that can come later in another bug.

This thing is running arbitrary commands on servers. We need to be able to effectively figure out what had happened and trace through the command flow for post-mortem debugging.

Just leaving this here as a reminder for now. git grep in the code for notes.

Certain clients should be limited in the commands they can send. We need to figure out what our ACL system is going to look like.

Not all masters can talk to all agents. We need to figure out what the routing layer and the master<->master commands look like.

To authenticate the masters to the agents, we need certs and TLS integration.

Quoting doesn't work properly and we need it to, for obvious reasons:

ml36806:bcli rbastic$ ./bcli -h "127.0.0.1:8081" 'ls /'
2014/08/03 20:09:15 req= {[127.0.0.1:8081] [ls /] 10}
2014/08/03 20:09:15 response: {1407089355500008967}
2014/08/03 20:09:18 host: localhost:8081
stderr:
stdout:
err: exec: "ls /": stat ls /: no such file or directory
exit: 0

VERSUS

20:11 ml36806:bcli rbastic$ ./bcli -h "127.0.0.1:8081" ls /

which would not support pipes.

We are going to use client certs for authentication. We need to check they've been signed by the right CA.

If we have a lot of concurrent requests, we might be creating lots of network connections. A stream multiplexer like muxado would combine those all into a single connection.

Let's leave this until we know if this is a problem in practice.

https://github.com/inconshreveable/muxado

Something like snowflake. We need multiple masters to be able to generate unique ids without centralization. We also don't want to drain /dev/random every time.

Everything is still hardcoded at the moment. Pull all the settings into configuration files.