A Few of My Favorite Tools

strace

Recently I’ve started branching out some in my debugging style. In the past it was usually adding print statements, reading docs carefully, reading logs, etc. I still mostly add print statements when I’m debugging my own code, but when trying to figure out why some random program isn’t working, instead of reading docs I go straight to strace.

If you don’t already know, strace traces system calls, so it effectively gets between the program and the kernel and lists all the system calls being made. It’s surprising how informative something like strace can be! There’s another thing called ltrace that does the same thing for libraries, but I’ve never used it.

I suspect strace is old hat to most, but using it is a habit I haven’t been in until recently. Usually with a bare strace invocation I can see the problem immediately, IE the thing is trying to read a file that doesn’t exist yet, or listening on some other port.

To use strace all you have to do is prefix the command you are running with strace.

So for example, to trace ls you merely run:

  strace ls

Of course eventually the output from strace is just too much, so you need to filter it. I just filter by system call, but there are probably other ways to do it too. To see just the calls to open that ls makes you can do the following:

  strace -e trace=open

And you can give it multiple syscalls too:

  strace -e trace=open,stat

Oh and one last thing; for some reason I had the misconception that strace requires you run programs “from within” strace and thus you can’t trace already running programs. Fortunately that’s false! The following should work in general:

  sudo strace -p $(pgrep firefox)

If need be you can reconfigure your system so that you don’t need to be root to trace your own programs, but fortunately I don’t need to trace running programs often enough to need that.

sysdig

After YAPC this year I decided to go through and watch some of the talks I’d missed for whatever reason, and I watched Sartak’s DTrace War Stories, despite the fact that DTrace doesn’t work on Linux. In the talk someone asked about this and Sartak mentioned sysdig. I didn’t get my hopes up too much, having tried SystemTap and many other similar tracing facilities for Linux I’ve only been dissapointed.

I decided to take a look nonetheless and what I found was refreshing. Unlike strace, sysdig allows the user to examine much more of the running system than a single process (or tree of processes, with -f.)

sysdig was made based on the rich set of tools we have for network captures, with the goal of making a standard capture format that can later be queried offline, presumably with gui tools or whatever. Unlike DTrace and SystemTap, sysdig initially feels like a much louder strace. I ran sysdig on my laptop and after about 2s of running it had written 1.9M of data, when stored to a plain text file.

I’m not sure it’s worth me going over all the cool things about sysdig, but I will mention the one time I’ve used it so far when strace seemed like too much work (though I think it still could have gotten me what I needed.) I wanted to see the trace of all the processes B that had been started by process A. With sysdig one can use this simple oneliner:

  sudo sysdig proc.pname=daemonproxy

One other neat thing about sysdig; the basic “query language” is merely an expression that is built from a handful of relational operators (=, <, etc), booleans (and, or, etc), and a large list of fields. That’s great for live capturing or limiting the data you record, but if you want to do more complex filtering you can use what is called “chisels” which use Lua and can format the output programmatically, etc.

For those interested in sysdig I’d recommend reading Sysdig vs DTrace vs Strace and Fishing for Hackers. Both are fascinating reads and make it clear what can be done with sysdig.

daemonproxy

Finally I’d like to briefly mention daemonproxy. I have three computers I do about the same stuff on: web browsing, coding, reading email, etc. Two are work machines and one is a laptop. I set nearly all of the settings via scripts in my dotfiles repo but one thing that had been bugging me was that I have a handful of “user services” I need to run all the time. I had them running in their own tmux session but that means when I first log in I have to do something like:

  tmux -2 new-session -s services
  offlineimap
  syncthing
  sudo openvpn ...

It’s not the worst thing in the world, but offlineimap in particular gets hung pretty regularly and I have to open up the tmux session, kill it, and restart it. That’s where daemonproxy comes in. With the help of the author silverdirk I’ve come up with a relatively basic daemonproxy setup that runs all the programs above with logging and very basic supervision. I can run a single command that will start all of them, log their output, and restart them if they go down. So if offlineimap hangs, as it is wont to do, I just killall -3 offlineimap and it will restart automatically.

I’d recommend anyone who is doing anything with daemons at all check out silverdirk’s lighting talk. It really clarified how elegantly these things can work, at least for me.

Posted Mon, Jul 7, 2014

If you're interested in being notified when new posts are published, you can subscribe here; you'll get an email once a week at the most.