Machine Introspection

Sometimes, computers misbehave. And very often, you want to know why. Let’s look at some tools that help you do that!

But first, let’s make sure you’re able to do introspection. Often, system introspection requires that you have certain privileges, like being the member of a group (like power for shutdown). The root user is the ultimate privilege; they can do pretty much anything. You can run a command as root (but be careful!) using sudo.

What happened?

If something goes wrong, the first place to start is to look at what happened around the time when things went wrong. For this, we need to look at logs.

Traditionally, logs were all stored in /var/log, and many still are. Usually there’s a file or folder per program. Use grep or less to find your way through them.

There’s also a kernel log that you can see using the dmesg command. This used to be available as a plain-text file, but nowadays you often have to go through dmesg to get at it.

Finally, there is the “system log”, which is increasingly where all of your log messages go. On most, though not all, Linux systems, that log is managed by systemd, the “system daemon”, which controls all the services that run in the background (and much much more at this point). That log is accessible through the somewhat inconvenient journalctl tool if you are root, or part of the admin or wheel groups.

For journalctl, you should be aware of these flags in particular:

What is happening?

If something is wrong, or you just want to get a feel for what’s going on in your system, you have a number of tools at your disposal for inspecting the currently running system:

First, there’s top, and the improved version htop, which show you various statistics for the currently running processes on the system. CPU use, memory use, process trees, etc. There are lots of shortcuts, but t is particularly useful for enabling the tree view. You can also see the process tree with pstree (+ -p to include PIDs). If you want to know what those programs are doing, you’ll often want to tail their log files. journalctl -f, dmesg -w, and tail -f are you friends here.

Sometimes, you want to know more about the resources being used overall on your system. dstat is excellent for that. It gives you real-time resource metrics for lots of different subsystems like I/O, networking, CPU utilization, context switches, and the like. man dstat is the place to start.

If you’re running out of disk space, there are two primary utilities you’ll want to know about: df and du. The former shows you the status of all the partitions on your system (try it with -h), whereas the latter measures the size of all the folders you give it, including their contents (see also -h and -s).

To figure out what network connections you have open, ss is the way to go. ss -t will show all open TCP connections. ss -tl will show all listening (i.e., server) ports on your system. -p will also include which process is using that connection, and -n will give you the raw port numbers.

System configuration

There are many ways to configure your system, but we’ll got through two very common ones: networking and services. Most applications on your system tell you how to configure them in their manpage, and usually it will involve editing files in /etc; the system configuration directory.

If you want to configure your network, the ip command lets you do that. Its arguments take on a slightly weird form, but ip help command will get you pretty far. ip addr shows you information about your network interfaces and how they’re configured (IP addresses and such), and ip route shows you how network traffic is routed to different network hosts. Network problems can often be resolved purely through the ip tool. There’s also iw for managing wireless network interfaces. ping is a handy tool for checking how deeply things are broken. Try pinging a hostname (google.com), an external IP address (1.1.1.1), and an internal IP address (192.168.1.1 or default gw). You may also want to fiddle with /etc/resolv.conf to check your DNS settings (how hostnames are resolved to IP addresses).

To configure services, you pretty much have to interact with systemd these days, for better or for worse. Most services on your system will have a systemd service file that defines a systemd unit. These files define what command to run when that services is started, how to stop it, where to log things, etc. They’re usually not too bad to read, and you can find most of them in /usr/lib/systemd/system/. You can also define your own in /etc/systemd/system .

Once you have a systemd service in mind, you use the systemctl command to interact with it. systemctl enable UNIT will set the service to start on boot (disable removes it again), and start, stop, and restart will do what you expect. If something goes wrong, systemd will let you know, and you can use journalctl -u UNIT to see the application’s log. You can also use systemctl status to see how all your system services are doing. If your boot feels slow, it’s probably due to a couple of slow services, and you can use systemd-analyze (try it with blame) to figure out which ones.

Exercises

locate? dmidecode? tcpdump? /boot? iptables? /proc?


Edit this page.

Licensed under CC BY-NC-SA.