There are a set of proxy environment variables (http_proxy, HTTP_PROXY, https_proxy, HTTPS_PROXY, no_proxy, NO_PROXY) in my OpenShift pods that I did not explicitly include and I do not want them there.
For example
$ oc run netshoot -it --image docker-registry.default.svc:5000/default/netshoot -- bash
If you don't see a command prompt, try pressing enter.
bash-4.4$ env | grep -i proxy | sort
HTTPS_PROXY=http://xx.xx.xx.xx:8081/
HTTP_PROXY=http://xx.xx.xx.xx:8081/
NO_PROXY=.cluster.local,.mydomain.nz,.localdomain.com,.svc,10.xx.xx.xx,127.0.0.1,172.30.0.1,app01.mydomain.nz,app02.mydomain.nz,inf01.mydomain.nz,inf02.mydomain.nz,mst01.mydomain.nz,localaddress,localhost,.edpay.nz
http_proxy=xx.xx.xx.xx:8081
https_proxy=xx.xx.xx.xx:8081
no_proxy=.cluster.local,.mydomain.nz,.localdomain.com,.svc,10.xx.xx.xx,127.0.0.1,172.30.0.1,app01.mydomain.nz,app02.mydomain.nz,inf01.mydomain.nz,inf02.mydomain.nz,mst01.mydomain.nz,localaddress,localhost,.edpay.nzI have yet to track down how those env vars are getting into my pods.
I am not Setting Proxy Environment Variables in Pods.
$ oc get pod netshoot-1-hjp2p -o yaml | grep -A 10 env
[no output]
$ oc get deploymentconfig netshoot -o yaml | grep -A 10 env
[no output]
I am not Creating Pod Presets
$ oc get podpresets --all-namespaces
No resources found.Docker on my master/app nodes have no proxy env vars.
$ grep -i proxy /etc/sysconfig/docker
[no output]Kubelet (openshift-node) on my master/app nodes have no proxy env vars.
$ grep -i proxy /etc/sysconfig/atomic-openshift-node
[no output]Master components on my master nodes have no proxy env vars.
$ grep -i proxy /etc/sysconfig/atomic-openshift-master
[no output]
$ grep -i proxy /etc/sysconfig/atomic-openshift-master-api
[no output]
$ grep -i proxy /etc/sysconfig/atomic-openshift-master-controllers
[no output]Contents of sysconfig files (not including comments)
$ cat /etc/sysconfig/atomic-openshift-master
OPTIONS="--loglevel=0"
CONFIG_FILE=/etc/origin/master/master-config.yaml
$ cat /etc/sysconfig/atomic-openshift-node
OPTIONS=--loglevel=2
CONFIG_FILE=/etc/origin/node/node-config.yaml
IMAGE_VERSION=v3.9.51
$ cat /etc/sysconfig/docker
OPTIONS=' --selinux-enabled --signature-verification=False --insecure-registry 172.30.0.0/16'
if [ -z "${DOCKER_CERT_PATH}" ]; then
DOCKER_CERT_PATH=/etc/docker
fi
ADD_REGISTRY='--add-registry registry.access.redhat.com'
$ cat /etc/sysconfig/atomic-openshift-master-api
OPTIONS=--loglevel=2 --listen=https://0.0.0.0:8443 --master=https://mst01.mydomain.nz:8443
CONFIG_FILE=/etc/origin/master/master-config.yaml
OPENSHIFT_DEFAULT_REGISTRY=docker-registry.default.svc:5000
$ cat /etc/sysconfig/atomic-openshift-master-controllers
OPTIONS=--loglevel=2 --listen=https://0.0.0.0:8444
CONFIG_FILE=/etc/origin/master/master-config.yaml
OPENSHIFT_DEFAULT_REGISTRY=docker-registry.default.svc:5000I'm at a loss as to how those proxy env vars are getting into my pods.
Versions:
We finally figured this out. We had openshift_http_proxy, openshift_https_proxy, and openshift_no_proxy set in our installer inventory variables as per Configuring Global Proxy Options.
We knew that this meant it also implicitly set the openshift_builddefaults_http_proxy, openshift_builddefaults_https_proxy, and openshift_builddefaults_no_proxy installer inventory variables and according to the docs
This variable defines the HTTP_PROXY environment variable inserted into builds using the BuildDefaults admission controller. If you do not define this parameter but define the openshift_http_proxy parameter, the openshift_http_proxy value is used. Set the openshift_builddefaults_http_proxy value to False to disable default http proxy for builds regardless of the openshift_http_proxy value.
What we did not know (and I would argue is not at all clear from the description above), is that setting those installer inventory variables sets the HTTP_PROXY, HTTPS_PROXY, and NO_PROXY env vars permanently within your images.
It's painfully apparent now when we look back on the build logs and see lines like this
...
Step 2/19 : ENV "HTTP_PROXY" "xxx.xxx.xxx.xxx" "HTTPS_PROXY" "xxx.xxx.xxx.xxx" "NO_PROXY" "127.0.0.1,localhost,172.30.0.1,.svc,.cluster.local" "http_proxy" "xxx.xxx.xxx.xxx" "https_proxy" "xxx.xxx.xxx.xxx" "no_proxy" "127.0.0.1,localhost,172.30.0.1,.svc,.cluster.local"
...
We couldn't exclude proxy env vars from the pods because those env vars were set at build time.