OSX hosts file craps out on long lines
Posted on Jul 28, 2015
•
•
Back Story
sigh
So, we use the following technologies:
- vagrant to run our dev environment
- ansible to provision our dev and prod environments
- Mac OSX as our dev machines (Yosemite - 10.10.4 at time of writing)
Once upon a time, we tried to make our dev environment (vagrant) as ‘live like’ as possible. We used multiple VMs to represent each server in our production tier.
So we could be even more ‘live like’ (and so we didn’t need to remember IP addresses) we used vagrant hostmanager to
automatically update our host machine’s /etc/hosts file with useful hostnames to our vagrant machine IPs, like so:
## vagrant-hostmanager-start id: 5586eff5-227f-47em-b6b9-cc7d775d8160
192.168.24.41 foo1.localhost.com
192.168.24.42 foo2.localhost.com
192.168.24.43 foo3.localhost.com
192.168.24.44 foo4.localhost.com
192.168.24.45 foo5.localhost.com
192.168.24.46 foo6.localhost.com
192.168.24.47 foo7.localhost.com
192.168.24.48 foo8.localhost.com
## vagrant-hostmanager-end
This worked fine for a while, and ansible runs worked great; the best thing was being able to blow the machines away and start fresh with a new environment.
After adding more and more services, the overhead was too much, and our laptops ran out of resource. As a solution, we switched to using a single, bigger Vagrant instance to handle all of our products on one machine.
Vagrant hostmanager started writing the following, single-line hostfile:
## vagrant-hostmanager-start id: 5586eff5-227f-47em-b6b9-cc7d775d8160
192.168.24.41 foo1.localhost.com foo2.localhost.com foo3.localhost.com foo4.localhost.com foo5.localhost.com foo6.localhost.com foo7.localhost.com foo8.localhost.com
## vagrant-hostmanager-end
The Problem
That’s when we started to have trouble. Ansible was taking ages to complete even simple tasks (e.g. copy config
files), but only against certain machines.
For far too long we assumed we had done something wrong in the Ansible roles being run against the slow services.
However, our Ansible code wasn’t the problem. It seems there is a bug in OSX with parsing the hosts file /etc/hosts,
the result of the bug is that some of the hosts, when many are defined on a single line, can randomly take ~30 seconds to
resolve the host over SSH.
sigh again
The Solution
Thankfully, the maintainer of Vagrant Hostmanager was kind enough to fix the issue in version
1.6.0 (vagrant plugin update hostmanager), and now, if you add this to your Vagrantfile config:
config.hostmanager.aliases_on_separate_lines = true
… your hosts file will look like this:
## vagrant-hostmanager-start id: 5586eff5-227f-47em-b6b9-cc7d775d8160
192.168.24.41 foo1.localhost.com
192.168.24.41 foo2.localhost.com
192.168.24.41 foo3.localhost.com
192.168.24.41 foo4.localhost.com
192.168.24.41 foo5.localhost.com
192.168.24.41 foo6.localhost.com
192.168.24.41 foo7.localhost.com
192.168.24.41 foo8.localhost.com
## vagrant-hostmanager-end
… and the problem is solved.