Traefik 1.7.12 exiting unexpectedly - swarm mode

Traefik Traefik v1
,
1

I’m using traefik 1.7.12 with no .toml file, just command line parameters and docker labels. We’re using swarm on AWS with cloudwatch logging. Several times the traefik container is stopping , with no log output, and either exit code 0, or exit code 137. Inspecting the stopped container shows OOMKilled: false , there are sufficient resources on the swarm managers. Anyone see this? What could be the problem?

Running traefik as a docker service with the following command line options to the container:

              "Args": [
                "--docker",
                "--docker.swarmmode",
                "--docker.domain=traefik",
                "--docker.watch",
                "--docker.exposedbydefault=false",
                "--web"
            ],

And the labels for the containers:

            "traefik.docker.network": "traefiknet",
            "traefik.enable": "true",
            "traefik.frontend.entryPoints": "http",
            "traefik.frontend.rule": "Host:someprefix.somedomain.io",
            "traefik.port": "80"

The configuration works just fine, everything is correctly routed.

It’s just that maybe once a week (but we did have 3 in a week recently) the traefik process will exit with either (0) or (137).

There is nothing in cloudwatch [docker log] for the traefik daemon.

I know this isn’t much to go on, but would appreciate some help in what to investigate next. Thanks.

1 Like
2

Hi @Harper, when you say There is nothing in cloudwatch [docker log] for the traefik daemon.,
do you mean "zero line of logs", or do you mean "some logs, but nothing related to the error" ?

Also, can you try one of the following options to check if you can get more information?

Finally, if you feel that there might be a resource issue on your swarm managers:

  • What are the limits in memory and CPU (or any other kind) applied to your Traefik's container?
  • You might want to run Traefik on a worker node, and use a "docker socket" forwarder insiode an encrypted network. Example here: https://gist.github.com/dduportal/fe6f07f447e2a88b302b376e36aba934
  • Do you have a monitoring stack of your metrics? It could be interesting to watch the resources usage at machine level (cadvisor/grafana/prometheus might be a good start here if not the case).

Let us know!