search

SirIle / sirile.github.io

4 stars 1 forks source link

Comments for overlay networking on AWS post #12

Open SirIle opened 8 years ago

SirIle commented 8 years ago

Placeholder for comments.

MBuffenoir commented 8 years ago

Hello and thanks for the post,

Did you find a good way to do service registration in consul (or other KV store) with overlay network. I would like to update automatically my haproxy config with the virtual ip of my frontends, but can not find a way to register them as of now. I tried registrator without success ...

SirIle commented 8 years ago

@MBuffenoir Yeah, I got registrator to work, more in here: http://sirile.github.io/2015/12/14/automatic-scaling-with-docker-19-and-overlay-network-locally-and-in-aws.html. It requires a version of registrator with a certain PR merged, but after that it works.

MBuffenoir commented 8 years ago

Thanks ! Is it kidibox/registrator ? btw, it seems that progrium is asking if someone had done it yet here : https://github.com/gliderlabs/registrator/pull/303

I'm doing the same thing than you, though I don't use logspout anymore, but the docker syslog logdriver... I there still an advantage in using logspout ?

SirIle commented 8 years ago

I did a "docker search registrator" and just happened on kidibox/registrator as it had that PR in the comment and it seems to work. For demonstration purposes I like logspout as my logging is completely pluggable and removable without needing to restart the daemon. I haven't yet looked at the syslog driver all that much, it's on the todo-list. How does the syslog driver behave if the Logstash (or something else) isn't actively listening yet when the daemon starts?

MBuffenoir commented 8 years ago

I used syslog driver with udp, so it does not really matter if you did not start yet your ELK stack ... Haven't tried with tcp. The pluggability of logspout is a plus in testing env, for sure. I think we are in the right direction with this PR, though supporting multiple overlay might be needed at some point. The swarm is getting closer and close to production ready :-)

MBuffenoir commented 8 years ago

Is there any benefits on running one consul per host in the swarm ? I've made it with one instance on my "management" infra node. (I guess this should be a cluster in a prod env). kidibox/registrator worked perfect for me ... Thank you

SirIle commented 8 years ago

The idea is to use the local Consul instance as the DNS server for that node and tie it to the local Consul bridge IP.

kvashishta commented 7 years ago

I am using an on premise machine to do the docker-machine stuff and launched successfully the swarm in AWS using Centos AMI. [root@test232 elastic]# docker-machine ls Error attempting call to get driver name: connection is shut down NAME ACTIVE DRIVER STATE URL SWARM DOCKER ERRORS infra-aws - Error Unknown unexpected EOF infras-aws - amazonec2 Running tcp://52.24.50.186:2376 v1.13.0 swarm-0-aws - amazonec2 Running tcp://35.163.53.38:2376 swarm-0-aws (master) v1.13.0

swarm-2-aws - amazonec2 Running tcp://35.167.172.143:2376 swarm-0-aws v1.13.0 swarm-3-aws * amazonec2 Running tcp://35.162.172.118:2376 swarm-0-aws v1.13.0

swarm-6-aws - amazonec2 Running tcp://35.166.82.33:2376 swarm-0-aws v1.13.1

However when I do a: [root@swarm-0-aws ~]# docker service ls Error response from daemon: This node is not a swarm manager. Use "docker swarm init" or "docker swarm join" to connect this node to swarm and try again. [root@swarm-0-aws ~]# docker service ls Error response from daemon: This node is not a swarm manager. Use "docker swarm init" or "docker swarm join" to connect this node to swarm and try again. [root@swarm-0-aws ~]# swarm-0-aws is the Master, is a "manager" a different concept, please advise.