Merge pull request #62508 from dims/allow-customized-image-for-dockerized-kubelet

Automatic merge from submit-queue. If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.

Support containerized kubelet in CI

**What this PR does / why we need it**:

  * Adjust hyperkube Makefile to allow HYPERKUBE_BIN to be
    specified from outside

  * new DOCKER_ROOT variable to specify --root-dir and associated
    volume for kubelet to know where docker directory lives

  * Save the docker logs when we kill the kubelet container

  * Allow customized image for dockerized kubelet and build a Default
    image if one is not specified

  We should allow a custom build using script we have:
  `VERSION="latest" REGISTRY="k8s.gcr.io" hack/dev-push-hyperkube.sh`

  to be used quickly using DOCKERIZE_KUBELET. To do this we remove the
  hard coded `k8s.gcr.io/kubelet` image and introduce a new environment
  variable `KUBELET_IMAGE`.

  Note that we are switching to hyperkube from kubelet as we have a quick
  and easy script to build hyperkube image and load it into local docker
  daemon. This reduces bad hacks like `docker tag
  k8s.gcr.io/hyperkube-amd64:<tag> k8s.gcr.io/kubelet:latest` being used
  by folks today (see 62057)

  * Better cope with PID 0 when running "docker inspect" to look for
    the process id for kubelet container


**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
Fixes #

**Special notes for your reviewer**:

**Release note**:

```release-note
NONE
```

cluster/images/hyperkube/Makefile

@@ -20,7 +20,7 @@
 REGISTRY?=staging-k8s.gcr.io
 ARCH?=amd64
 OUT_DIR?=_output
-HYPERKUBE_BIN?=$(OUT_DIR)/dockerized/bin/linux/$(ARCH)/hyperkube
+HYPERKUBE_BIN?=$(shell pwd)/../../../$(OUT_DIR)/dockerized/bin/linux/$(ARCH)/hyperkube
 
 BASEIMAGE=k8s.gcr.io/debian-hyperkube-base-$(ARCH):0.10
 TEMP_DIR:=$(shell mktemp -d -t hyperkubeXXXXXX)
@@ -33,7 +33,7 @@ ifndef VERSION
     $(error VERSION is undefined)
 endif
 	cp -r ./* ${TEMP_DIR}
-	cp ../../../${HYPERKUBE_BIN} ${TEMP_DIR}
+	cp ${HYPERKUBE_BIN} ${TEMP_DIR}
 
 	chmod a+rx ${TEMP_DIR}/hyperkube
 

hack/local-up-cluster.sh

@@ -22,6 +22,7 @@ KUBE_ROOT=$(dirname "${BASH_SOURCE}")/..
 DOCKER_OPTS=${DOCKER_OPTS:-""}
 DOCKER=(docker ${DOCKER_OPTS})
 DOCKERIZE_KUBELET=${DOCKERIZE_KUBELET:-""}
+DOCKER_ROOT=${DOCKER_ROOT:-""}
 ALLOW_PRIVILEGED=${ALLOW_PRIVILEGED:-""}
 DENY_SECURITY_CONTEXT_ADMISSION=${DENY_SECURITY_CONTEXT_ADMISSION:-""}
 PSP_ADMISSION=${PSP_ADMISSION:-""}
@@ -31,6 +32,7 @@ KUBELET_AUTHORIZATION_WEBHOOK=${KUBELET_AUTHORIZATION_WEBHOOK:-""}
 KUBELET_AUTHENTICATION_WEBHOOK=${KUBELET_AUTHENTICATION_WEBHOOK:-""}
 POD_MANIFEST_PATH=${POD_MANIFEST_PATH:-"/var/run/kubernetes/static-pods"}
 KUBELET_FLAGS=${KUBELET_FLAGS:-""}
+KUBELET_IMAGE=${KUBELET_IMAGE:-""}
 # many dev environments run with swap on, so we don't fail in this env
 FAIL_SWAP_ON=${FAIL_SWAP_ON:-"false"}
 # Name of the network plugin, eg: "kubenet"
@@ -340,6 +342,13 @@ cleanup_dockerized_kubelet()
   if [[ -e $KUBELET_CIDFILE ]]; then
     docker kill $(<$KUBELET_CIDFILE) > /dev/null
     rm -f $KUBELET_CIDFILE
+
+    # Save the docker logs
+    if [[ -f /var/log/docker.log ]]; then
+      sudo cp /var/log/docker.log ${LOG_DIR}/docker.log
+    elif command -v journalctl &>/dev/null; then
+      journalctl -u docker --no-pager > ${LOG_DIR}/docker.log
+    fi
   fi
 }
 
@@ -778,6 +787,13 @@ function start_kubelet {
       sudo -E "${GO_OUT}/hyperkube" kubelet "${all_kubelet_flags[@]}" >"${KUBELET_LOG}" 2>&1 &
       KUBELET_PID=$!
     else
+
+      # Build the hyperkube container image if necessary
+      if [[ -z "$KUBELET_IMAGE" && -n "$DOCKERIZE_KUBELET" ]]; then
+        HYPERKUBE_BIN="${GO_OUT}/hyperkube" REGISTRY="k8s.gcr.io" VERSION="latest" make -C "${KUBE_ROOT}/cluster/images/hyperkube" build
+        KUBELET_IMAGE="k8s.gcr.io/hyperkube-amd64:latest"
+      fi
+
       # Docker won't run a container with a cidfile (container id file)
       # unless that file does not already exist; clean up an existing
       # dockerized kubelet that might be running.
@@ -799,7 +815,7 @@ function start_kubelet {
       fi
       all_kubelet_flags+=(--containerized)
 
-      docker run --rm --name kubelet \
+      all_kubelet_volumes=(
         --volume=/:/rootfs:ro,rslave \
         --volume=/var/run:/var/run:rw \
         --volume=/sys:/sys:ro \
@@ -807,19 +823,28 @@ function start_kubelet {
         --volume=/var/lib/kubelet/:/var/lib/kubelet:rslave \
         --volume=/dev:/dev \
         --volume=/run/xtables.lock:/run/xtables.lock:rw \
+      )
+
+      if [[ -n "${DOCKER_ROOT}" ]]; then
+        all_kubelet_flags+=(--root-dir="${DOCKER_ROOT}")
+        all_kubelet_volumes+=(--volume="${DOCKER_ROOT}:${DOCKER_ROOT}:rslave")
+      fi
+
+      docker run --rm --name kubelet \
+        "${all_kubelet_volumes[@]}" \
         ${cred_bind} \
         --net=host \
         --pid=host \
         --privileged=true \
         -i \
         --cidfile=$KUBELET_CIDFILE \
-        k8s.gcr.io/kubelet \
+        "${KUBELET_IMAGE}" \
         /kubelet "${all_kubelet_flags[@]}" >"${KUBELET_LOG}" 2>&1 &
       # Get PID of kubelet container.
       for i in {1..3}; do
         echo -n "Trying to get PID of kubelet container..."
         KUBELET_PID=$(docker inspect kubelet -f '{{.State.Pid}}' 2>/dev/null || true)
-        if [ -n "$KUBELET_PID" ]; then
+        if [[ -n ${KUBELET_PID} && ${KUBELET_PID} -gt 0 ]]; then
             echo " ok, $KUBELET_PID."
             break
         else
@@ -989,9 +1014,15 @@ if [[ "${KUBETEST_IN_DOCKER:-}" == "true" ]]; then
   ${KUBE_ROOT}/hack/install-etcd.sh
   export PATH="${KUBE_ROOT}/third_party/etcd:${PATH}"
   KUBE_FASTBUILD=true make ginkgo cross
+
   apt install -y sudo
+  apt-get remove -y systemd
+
   # configure shared mounts to prevent failure in DIND scenarios
   mount --make-rshared /
+
+  # kubekins has a special directory for docker root
+  DOCKER_ROOT="/docker-graph"
 fi
 
 # validate that etcd is: not running, in path, and has minimum required version.