bad config file

[isaldarriaga]

I labeled 1 node with cloud.google.com/gke-nodepool=pool-kamailio

kubectl apply -f 01-nats.yaml

kubectl label nodes $(kubectl get nodes | grep gke | awk -F ' ' 'NR==1{print $1}') cloud.google.com/gke-nodepool=pool-kamailio --overwrite

kubectl apply -f kamailio.yaml
kubectl apply -f asterisk.yaml
kubectl apply -f app.yaml

so the DaemonSet launches the pod in one of the cluster nodes.

The "kamailio" container in the kamailio-xxx pod is failing though:

$ kubectl get po

NAME                            READY     STATUS             RESTARTS   AGE
asterisk-1744660152-8msl9       3/3       Running            13         1d
app-3527931237-jg899   1/1       Running            0          4h
kamailio-t9bjr                  2/3       CrashLoopBackOff   6          10m

The error:

$ kubectl logs kamailio-t9bjr -c kamailio

2017-08-24T00:18:25Z gke-foo-cluster-default-pool-72561ba1-992g confd[13]: INFO Backend set to env
2017-08-24T00:18:25Z gke-foo-cluster-default-pool-72561ba1-992g confd[13]: INFO Starting confd
2017-08-24T00:18:25Z gke-foo-cluster-default-pool-72561ba1-992g confd[13]: INFO Backend nodes set to 
2017-08-24T00:18:25Z gke-foo-cluster-default-pool-72561ba1-992g confd[13]: INFO /etc/kamailio/kamailio.cfg has md5sum 17481ad2235ef9698da57c6e90d5ac2b should be 469d897d2dde81f6320357a41712c89f
2017-08-24T00:18:25Z gke-foo-cluster-default-pool-72561ba1-992g confd[13]: INFO Target config /etc/kamailio/kamailio.cfg out of sync
2017-08-24T00:18:25Z gke-foo-cluster-default-pool-72561ba1-992g confd[13]: INFO Target config /etc/kamailio/kamailio.cfg has been updated
 0(1) : <core> [cfg.y:3368]: yyerror_at(): parse error in config file /etc/kamailio/kamailio.cfg, line 35, column 68-74: syntax error
 0(1) : <core> [cfg.y:3368]: yyerror_at(): parse error in config file /etc/kamailio/kamailio.cfg, line 35, column 68-74: invalid host or interface name
 0(1) : <core> [cfg.y:3371]: yyerror_at(): parse error in config file /etc/kamailio/kamailio.cfg, line 35, column 75: 
ERROR: bad config file (3 errors)

Looks like the software expects a specific checksum on the configuration file, then replace it and fails with the default values.

Any workaround?

[Ulexus]

It's complaining about line 35 being invalid. Line 35 is the alias line, which means you're almost certainly failing the hostname or public IP address lookups in the entrypoint script.

I would run, manually, those curl commands and make sure you are getting reasonable results. Failing that, modify the entrypoint script to echo out the resulting environment variables.

[isaldarriaga]

The lookups seems to be working fine from inside a running container in the kamailio pod (i havent' printed the vars from the entrypoint yet):

$ kubectl exec -it kamailio-t9bjr -c rtpproxy -- sh

/ # curl -H 'Metadata-Flavor: Google' http://metadata.google.internal/computeMetadata/v1/instance/network-interfaces/0/access-configs/0/external-ip
35.xxx.191.224

/ # curl -H 'Metadata-Flavor: Google' http://metadata.google.internal/computeMetadata/v1/instance/network-interfaces/0/ip
10.138.0.2

/ # curl -H 'Metadata-Flavor: Google' http://metadata.google.internal/computeMetadata/v1/instance/hostname
gke-foo-cluster-default-pool-72561ba1-992g.c.foo-project.internal

/ # exit

I'm trying to understand better the alias code here.

The kamailio.conf is a dummy file in the source code, but as far as i understand it's copied at runtime from live-demo/kamailio/confd-env/templates/kamailio.cfg.tmpl, right?

(evidenced in the log entry: 2017-08-24T00:18:25Z gke-foo-cluster-default-pool-72561ba1-992g confd[13]: INFO Target config /etc/kamailio/kamailio.cfg has been updated)

In the kamailio.cfg.tmpl line (35) we have:

## 
# Manual addition of aliased hostnames held by the local server.  This
# adds hostname/port tuples to the list of destinations that match the
# 'myself' keyword and are considered local by loose_route().  It is 
# necessary to provide both the hostname and the port (e.g. alias=host:5060)
# for this to work properly.  These entries supplement local aliases
# automatically discovered in /etc/hosts.  Multiple 'alias' entries may 
# be present.

alias=sip.astricon.cycore.io {{getv "/public/ipv4"}} {{getv "/public/hostname"}}

According to the document we need a tuple hostname:port, but only the ip is returned by the curl command. Is this relevant?

[Ulexus]

Aha, you're exactly correct. They do need the port. I wonder why that is missing... oh, well, yes. It is obviously wrong, and it needs the port number

[isaldarriaga]

I changed the alias line in kamailio.cfg.tmpl :

alias="{{getv "/public/hostname"}}:5060"

since GCE FQDNs have dashes, the alias need to be enclosed in double quotes too according to the kamailio specs.

Now the container starts up.