jwilder / docker-discover Goto Github PK

docker run -d --net host --name docker-discover -e ETCD_HOST=192.168.56.102:4001 -p 1936:1936 -t jwilder/docker-discover

docker logs docker-discover

Error: 'Key not found : /backends'
Error: 'Key not found : /backends'
Error: 'Key not found : /backends'
Error: 'Key not found : /backends'

service on 1936 is unreacheable. Also whoami servers is unreacheable.
If I try to connect directly to a single whoami serv it works fine.

The python script should handle its child processes (reload-haproxy.sh) properly to avoid zombies.

signal.signal(signal.SIGCHLD, signal.SIG_IGN)

works for me.

First of all, these set of scripts, and your blog post, are awesome. I've got everything running as described and it's great. I did run into issues however when using this with private registries.

I'm about 100% sure it has to do with the section of code that looks for 2 / marks to verify if it's a valid registry. With a private registry there will also be a full url. For example:

my.private-registry.com/my/image

Hi,
Thanks for this great job!

I am trying the demo 'whoami', and haproxy works fine. I was able to see backend container id.

But when I deployed my own services on one backend server x.x.x.2

• Container 2600516615c7 from Image service1 exposed port 0.0.0.0:8080->5001/tcp
• Container a6fc817cbca3 from Image service2 exposed port 0.0.0.0:443->5000/tcp
  Same containers are running on another backend server x.x.x.3
• Container 814d49616255 from Image service1 exposed port 0.0.0.0:8080->5001/tcp
• Container 31eb936f2e59 from Image service1 exposed port 0.0.0.0:8080->5001/tcp

docker-discover is deployed on server x.x.x.1

But when call service1 api from address x.x.x.1:8080, I got

connect to x.x.x.1 port 8080: Connection refused

I curl etcd for key 'backends/service1', it showed me 3 sub nodes 'backends/service1/2600516615c7',
'backends/service2/814d49616255', 'backends/service1/port'

And for haproxy config, I see

listen service1
    bind *:5001

    server 814d49616255 x.x.x.2:8080 check inter 2s rise 3 fall 2
    server 2600516615c7 x.x.x.3:8080 check inter 2s rise 3 fall 2

Hello,

First, thanks for all your projects around docker - they are great help.
One issue that I've been noticing is that when I have more than one container of the same service, the haproxy config would start reloading every 10-15 seconds even tho there were no change.

The issue is that etcd doesn't send back the nodes in the same order every time, and docker-discover would detect a change even tho it was just the same nodes, not in the same order.

A way to avoid this would be to have get_services() sort its results so that no change would be detected when nodes are swapped in the etcd response

I wrote a workaround, but I'm no python expert and it's too dirty to put in a pull request, but here it is anyway:

if name == "main":
current_services = {}
while True:
changed = False
try:
services = get_services()

        if not services or services == current_services:
            time.sleep(POLL_TIMEOUT)
            continue
        else:
            numberServices = len(services.items())
            current_numberServices = len(current_services.items())
            # print "numberServices = " + str(numberServices)
            # print "current_numberServices = " + str(current_numberServices)

            if numberServices == current_numberServices:
                #  "No new services... lets check them"
                for i in range(numberServices):
                    if services.items()[i][0] == current_services.items()[i][0]:
                        #  "Same services... Same values?"
                        regex = re.compile('[0-9]+.[0-9]+.[0-9]+.[0-9]+\:[0-9]+')
                        ipServices = sorted(re.findall(regex, str(services.items()[i])))
                        current_ipServices = sorted(re.findall(regex, str(current_services.items()[i])))
                        if ipServices == current_ipServices:
                        else:
                            changed = True
                            break;
                    else:
                        changed = True
                        break;
        if changed == True:
            print "config changed. reload haproxy"
            generate_config(services)
            ret = call(["./reload-haproxy.sh"])
            if ret != 0:
                print "reloading haproxy returned: ", ret
                time.sleep(POLL_TIMEOUT)
                continue
            current_services = services

    except Exception, e:
        print "Error:", e

    time.sleep(POLL_TIMEOUT)

I'm sure all of this could be done with a one liner sorting in get_services tho

Regards,

Exposing ports using -p option is not valid when using the host network stack. When attaching to the host network stack, it's meaningless to ask docker to bind a container port on a host port. The haxproy stats interface is thus not exposed by default

I started a fork for my own personal cloud of servers, but am worried that there isn't a license. Would it be reasonable to request that a license be added so that people whom may be unsure can use this as well?

Thanks! This is a neat project :)

@jwilder