syslogd with circular buffer

boyska 48d9d2df8c Merge remote-tracking branch 'origin/sig-coolness' 1 day ago
cmd b11c2edfc0 [tail] format locally 1 month ago
formatter b11c2edfc0 [tail] format locally 1 month ago
.drone.yml 5b7ddb62a6 drone CI 3 months ago
README.md b2127fd349 update readme to recent enhancements 1 month ago
hub.go 7c17c971a0 command responses 1 month ago
hub_test.go 499c90ccda test limit 1 month ago

README.md

A syslog daemon implementing circular buffer, in-memory storage.

This is useful when you want to keep some (heavily detailed) log available, but you don't want to log too many things to disk. Remember: logging is useful, but can be dangerous to your users' privacy!

On your "main" syslog, forward (part of the) messages to this one!

Integration examples

In these examples I'll refer to the usage of UNIX sockets. They are more secure than TCP/UDP sockets because they have file permissions, they can be "masked" using mount namespaces, etc. However, circlogd supports udp/tcp sockets easily, so that should not be an issue.

syslog-ng

To integrate into syslog-ng, put this in /etc/syslog-ng/conf.d/circolog.conf

destination d_circolog {
        unix-dgram("/run/circolog-syslog.sock"
                   flags(syslog-protocol)
                  );
};
log { source(s_src); destination(d_circolog); };

and run circologd -syslogd-socket /run/circolog-syslog.sock -query-socket /run/circolog-query.sock

Client

circolog has its own client: circolog-tail. It is intended to resemble tail -f for the most basic options; however, it will include filtering options that are common when you want to read logs, because that's very easy when you have structured logs available.

However, one design point of circolog is to be usable without having a specific client: so the logs are offered on both HTTP and websocket. This means that you can use curl if you want:

curl --unix-socket /run/circolog-query.sock localhost/

will give you everything that circologd has in memory.

If you want to "follow" (as in tail -f) you need to use the websocket interface. However, I don't know of any websocket client supporting UNIX domain socket, so you have two options:

  1. Use circolog-tail
  2. Use circologd with -query-addr 127.0.0.1:9080, add some iptables rule to prevent non-root to access that port, and run ws ws://localhost:9080/ws. You'll get all the "backlog", and will follow new log messages.

HTTP URLs and parameters

When using HTTP, logs are served on /. Valid parameters are:

  • l. This is the amount of lines to send. This is essentially the same as the -n parameter on tail Using l=-1 (the default) means "give me every log message that you have
  • fmt. This selects the output format. When fmt=json is used, each message is returned as JSON structured data. The format of those JSON messages is still unstable. fmt=syslog, the default, outputs messages using "syslog style" (RFC XXXXXX)

To use websocket, request path /ws. The same parameters of / are recognized.

Control daemon

Circologd can be controlled, on some aspects, at run-time. It has 2 mechanisms for that: the easiest, and more limited, is sending a signal with kill; the second, and more powerful, is a control socket, where you can give commands to it. This control socket is just HTTP, so again curl is your friend. In the future a circolog-ctl client will be developed.

Pause

When circologd is paused, every new message it receives is immediately discarded. No exception. The backlog is, however, preserved. This means that you can trigger the event that you want to investigate, pause circolog, then analyze the logs. Pausing might be the easiest way to make circologd only run "when needed".

When circologd resumes, no previous message is lost.

To pause circologd with signals , send a USR1 signal to the main pid. To "resume", send a USR1 again.

To pause with HTTP, send a POST /pause/toggle to your circologd control socket.

Clear

When you clear the circologd's buffer, it will discard every message it has, but will keep collecting new messages.

You can do that with POST /logs/clear