Docker in Production
About Host networking
[2015/8/19] Docker Meetup Tokyo #5
Yu Yamanaka (@yuurelx)
DevOps engineer at peroli, Inc.
By the way,
do you know largest curation media in Japan?
The largest one is NAVER matome by LINE
And MERY
PV = 250 million
UU = 19 millon
Web Media + Native Apps + Ad Platform + ??
We are using Docker in our Production!
Main Topic
About
Host networking
We are deploying containers with Host net. mode
Because it makes
easier to lay out network and faster than default mode
1. About networking modes
2. Host net. is easier
3. Host net. is faster
Subtopics
1. Docker Networking modes
- Bridge (default)
- Host (--net=host)
- Container
- None
Bridge mode
(From: "http://www.agilegroup.co.jp/technote/docker-network-in-bridge.html")
Host mode
Using host machine's network stack!
2. Host net. is easier
Bridge mode
We need to pass host's IP to conatiners...
$ HOST_IP=`ip addr show eth0 | grep -oE 'inet [^/]+' | cut -d ' ' -f 2`
$ docker run -d -E DB_HOST=$HOST_IP awesome_app:latest
config/database.yml
------------------------------
production:
host: <%= ENV['DB_HOST'] %>
username: user
password: secret
...
------------------------------
etc...What a hassle!
Host mode
Only specifying '127.0.0.1'
like a native one!
$ docker run -d --net=host awesome_app:latest
config/database.yml
------------------------------
production:
host: 127.0.0.1
username: user
password: secret
...
------------------------------
etc...127.0.0.1
Of course, elegant solution is containerizing all software on a server and link each other.
"Great idea, but not so good in reality"
3. Host net. is faster
Native vs Docker (bridge net.)
Apache Bench
container
100%
110%
About 10% slower...
1000 reqs
(by 65 sessions)
DB
Native vs Docker (host net.)
Apache Bench
container
100%
100%
No overhead!
DB
FASTER IS BETTER
Conclusion
Host net. mode makes
easier to lay out network and faster than default mode
But it sacrifices Docker's portability and independence. (e.g. port collision)
You should use this with caution!
Let's deploy containers
to your production!
Thank you for your attention!
Yu Yamanaka (@yuurelx)
DevOps engineer at peroli, Inc.
