This is the part most people want to know, so it comes first. The agent installed on your server sends aggregate system statistics, and nothing else:
- Processor use, load averages, and time spent waiting on disk.
- Memory and swap: how much is used, cached and available.
- Disk space and inodes used and available, and read/write operations per second.
- Network throughput in and out, and counts of errors and dropped packets.
- Counts of established and waiting TCP connections.
- How long the machine has been up.
- The machine’s hostname, its operating system name and version, and how many processor cores it has.
- Where your web server keeps its access log, if it uses one we recognise — the file’s location and whether we are able to read it. Not a single line of what is in it.
That last one exists so the dashboard can offer to switch web-traffic monitoring on for you, instead of asking you to edit a configuration file over SSH. Nothing is switched on until you press the button, and the agent only ever looks in a short list of standard locations that is compiled into it.
It does not send the contents of any file, any database, any environment variable, any credential, or anything belonging to your customers. It has no mechanism to do so — it reads the statistics the operating system publishes about itself and nothing else, and it runs as an unprivileged user that could not read most of your data even if it tried.
If you switch on web-traffic monitoring, the agent reads your reverse proxy’s access log locally and reports numbers: requests per second, how many returned an error, and how long responses took. The request paths, visitor IP addresses, user agents and referrers in that log are parsed past and discarded on your machine. None of them are transmitted to us and none are stored.