"INN's DOWN!"

It was not the cry that Joseph wanted to hear from the developers' cubicals. I.N.N. (Integration Now & Nightly) was the continuous integration server at Bethlehem LLC - responsible for testing the Perl code needed for their tax auditing application. If Joseph didn't get the system up and running soon, they were going to miss their deadline.

With a sigh, Joseph logged onto INN, and instantly he realized what the problem was. There was no room at the INN root filesystem.

Out of space on a CI server? How is that even possible? But wait, what was this, a directory for each test script run, each containing megabytes of verbose logs, test excel spreadsheets, and pngs of graphs.

Given that the tests were run hundreds of times a day...it was surprising that things hadn't got out of control before now.

The Problem With Temp Dirs

Joseph couldn't blame the developer for having a temp directory. Often a test needs to write out a bunch of things during a test run. Just a few uses of temp directories in tests are:

Providing a place to put test-generated config files that must be created on a per-run basis

A place you can store files your module produces in order to re-read them back in and check they were created properly (or have them hanging around after the test run to poke at if things go wrong)

A place to write copious logs and any other data that might help you work out what caused your test to fail

All in all very useful. But how should the developers code a temp directory in order not to ruin Joseph's afternoon?

A naive approach might be something along the lines of:

#!/usr/bin/perl

use strict;
use warnings;

use Test::More;
use File::Temp qw( tempdir );
my $dir = tempdir();
note "Tempdir is $dir";

...

The problem with this is that a new temp dir is created for each and every run of the test suite. And poor old Joseph found out the hard way that this means on a server that runs the test a lot of times that disk space can be exhausted if the test produces many megabytes of logs or output data.

One simple solution would be to have the test automatically tidy up after itself. This is a trivial change:

#!/usr/bin/perl

use strict;
use warnings;

use Test::More;
use File::Temp qw( tempdir );
my $dir = tempdir( CLEANUP => 1 );
note "Tempdir is $dir";

...

The CLEANUP => 1 causes the test directory to automatically be deleted when the program exits. Of course, the problem is that this happens as soon as the program exits even if the tests fail. You can't go dig through logs for why a failure happened if the directory where the logs have been written to has just been deleted!

Both 'solutions' also suffer from the problem that a different path is created each time the script runs, which can be very tedious to work with when you're trying to repeatedly get a test script to run and have to copy and paste the name of the directory your logs are in this run from the diagnostic output.

Test::TempDir::Tiny to the rescue

Test::TempDir::Tiny isn't any more complicated to use than File::Temp:

#!/usr/bin/perl

use strict;
use warnings;

use Test::More tests => 1;
use Test::TempDir::Tiny qw( tempdir );
my $dir = tempdir();
note "Tempdir is $dir";

...

But now the location of the tests is always the same. For the test t/mary/babyj.t the first tempdir created with tempdir will be created at tmp/t_mary_babyj_t/default_1.

   bash$ perl t/mary/babyj.t
   1..1
   # Tempdir is /home/ci/build/tmp/t_mary_babyj_t/default_1
   not ok 1 - onto this day a child process is forked
   #   Failed test 'onto this day a child process is forked'
   #   at t/mary/babyj.t line 11.
   # Looks like you failed 1 test of 1.

Each time time the script executes the test directory is cleared out automatically. Now we only have one copy of the script output on the server!

But even better, see what happens when the tests in the scripts pass:

   bash$ perl t/mary/babyj.t
   1..1
   # Tempdir is /home/ci/build/tmp/t_mary_babyj_t/default_1
   ok 1 - onto this day a child process is forked
   bash$ ls /home/ci/build/tmp/t_mary_babyj_t/default_1
   ls: /home/ci/build/tmp/t_mary_babyj_t: No such file or directory

The temp directory is removed! If no other test script temp directories remain in tmp that directory is removed too:

   bash$ ls /home/ci/build/tmp
   ls: /home/ci/build/tmp: No such file or directory

This means on a successful run the server is left in a "clean" state. Awesome.

Stuck in the stable

Joseph sent an email to the developers to tell them to switch to Test::TempDir::Tiny while he cleaned up INN. Until there was room on INN the code would be stuck in whatever had been promoted to the stable branch.

See Also

• Test::TempDir::Tiny

• File::Temp