Progressing on faster e2e tests.

Makefile

@@ -24,12 +24,16 @@ PKG := `go list ./... | grep -v /vendor/`
 
 ifeq ($(DEBUG),1)
 	DOCKERFILE = DebugDockerfile
-	DEBUG_POSTFIX := -debug
+	DEBUG_POSTFIX := -debug-$(shell date hhmmss)
 	BUILD_FLAGS += -gcflags "-N -l"
 else
 	DOCKERFILE = Dockerfile
 endif
 
+ifeq ($(FRESH),1)
+  DEBUG_FRESH=$(shell date +"%H-%M-%S")
+endif
+
 ifdef CDP_PULL_REQUEST_NUMBER
 	CDP_TAG := -${CDP_BUILD_VERSION}
 endif
@@ -66,7 +70,7 @@ docker: ${DOCKERDIR}/${DOCKERFILE} docker-context
 	echo "Version ${VERSION}"
 	echo "CDP tag ${CDP_TAG}"
 	echo "git describe $(shell git describe --tags --always --dirty)"
-	cd "${DOCKERDIR}" && docker build --rm -t "$(IMAGE):$(TAG)$(CDP_TAG)$(DEBUG_POSTFIX)" -f "${DOCKERFILE}" .
+	cd "${DOCKERDIR}" && docker build --rm -t "$(IMAGE):$(TAG)$(CDP_TAG)$(DEBUG_FRESH)$(DEBUG_POSTFIX)" -f "${DOCKERFILE}" .
 
 indocker-race:
 	docker run --rm -v "${GOPATH}":"${GOPATH}" -e GOPATH="${GOPATH}" -e RACE=1 -w ${PWD} golang:1.8.1 bash -c "make linux"

e2e/Dockerfile

@@ -14,13 +14,15 @@ RUN apt-get update \
            python3-setuptools \
            python3-pip \
            curl \
+           vim \
     && pip3 install --no-cache-dir -r requirements.txt \
     && curl -LO https://storage.googleapis.com/kubernetes-release/release/v1.18.0/bin/linux/amd64/kubectl \
     && chmod +x ./kubectl \
     && mv ./kubectl /usr/local/bin/kubectl \
-    && apt-get vim
     && apt-get clean \
     && rm -rf /var/lib/apt/lists/*
 
-ENTRYPOINT ["python3", "-m", "unittest", "--failfast", "-v"]
+# working line
-CMD ["discover","--start-directory","/tests"]
+# python3 -m unittest discover -v --failfast -k test_e2e.EndToEndTestCase.test_lazy_spilo_upgrade --start-directory tests
+ENTRYPOINT ["python3", "-m", "unittest"]
+CMD ["discover","-v","--failfast","--start-directory","/tests"]

e2e/exec_into_env.sh

@@ -3,7 +3,7 @@
 export cluster_name="postgres-operator-e2e-tests"
 export kubeconfig_path="/tmp/kind-config-${cluster_name}"
 export operator_image="registry.opensource.zalan.do/acid/postgres-operator:latest"
-export e2e_test_runner_image="registry.opensource.zalan.do/acid/postgres-operator-e2e-tests-runner:latest"
+export e2e_test_runner_image="registry.opensource.zalan.do/acid/postgres-operator-e2e-tests-runner:0.3"
 
 docker run -it --entrypoint /bin/bash --network=host -e "TERM=xterm-256color" \
     --mount type=bind,source="$(readlink -f ${kubeconfig_path})",target=/root/.kube/config \

e2e/run.sh

@@ -9,7 +9,10 @@ IFS=$'\n\t'
 readonly cluster_name="postgres-operator-e2e-tests"
 readonly kubeconfig_path="/tmp/kind-config-${cluster_name}"
 readonly spilo_image="registry.opensource.zalan.do/acid/spilo-12:1.6-p5"
-readonly e2e_test_runner_image="registry.opensource.zalan.do/acid/postgres-operator-e2e-tests-runner:latest"
+readonly e2e_test_runner_image="registry.opensource.zalan.do/acid/postgres-operator-e2e-tests-runner:0.3"
+
+export GOPATH=${GOPATH-~/go}
+export PATH=${GOPATH}/bin:$PATH
 
 echo "Clustername: ${cluster_name}"
 echo "Kubeconfig path: ${kubeconfig_path}"
@@ -38,6 +41,7 @@ function start_kind(){
 }
 
 function load_operator_image() {
+  echo "Loading operator image"
   export KUBECONFIG="${kubeconfig_path}"
   kind load docker-image "${operator_image}" --name ${cluster_name}
 }
@@ -53,7 +57,6 @@ function set_kind_api_server_ip(){
 
 function run_tests(){
   echo "Running tests... image: ${e2e_test_runner_image}"
-  echo $@
   # tests modify files in ./manifests, so we mount a copy of this directory done by the e2e Makefile
 
   docker run --rm --network=host -e "TERM=xterm-256color" \
@@ -61,7 +64,7 @@ function run_tests(){
   --mount type=bind,source="$(readlink -f manifests)",target=/manifests \
   --mount type=bind,source="$(readlink -f tests)",target=/tests \
   --mount type=bind,source="$(readlink -f exec.sh)",target=/exec.sh \
-  -e OPERATOR_IMAGE="${operator_image}" "${e2e_test_runner_image}" $@
+  -e OPERATOR_IMAGE="${operator_image}" "${e2e_test_runner_image}" ${E2E_TEST_CASE-} $@
 }
 
 function clean_up(){
@@ -78,6 +81,7 @@ function main(){
   [[ ! -f ${kubeconfig_path} ]] && start_kind
   load_operator_image
   set_kind_api_server_ip
+
   shift
   run_tests $@
   exit 0

e2e/tests/test_e2e.py

@@ -10,7 +10,7 @@ import yaml
 from datetime import datetime
 from kubernetes import client, config
 
-from k8s_api import K8s
+from tests.k8s_api import K8s
 
 SPILO_CURRENT = "registry.opensource.zalan.do/acid/spilo-12:1.6-p5"
 SPILO_LAZY = "registry.opensource.zalan.do/acid/spilo-cdp-12:1.6-p114"
@@ -262,7 +262,7 @@ class EndToEndTestCase(unittest.TestCase):
         k8s = self.k8s
         # update infrastructure roles description
         secret_name = "postgresql-infrastructure-roles"
-        roles = "secretname: postgresql-infrastructure-roles-new,userkey: user, rolekey: memberof, passwordkey: password, defaultrolevalue: robot_zmon"
+        roles = "secretname: postgresql-infrastructure-roles-new, userkey: user, rolekey: memberof, passwordkey: password, defaultrolevalue: robot_zmon"
         patch_infrastructure_roles = {
             "data": {
                 "infrastructure_roles_secret_name": secret_name,
@@ -337,7 +337,6 @@ class EndToEndTestCase(unittest.TestCase):
         k8s.update_config(patch_lazy_spilo_upgrade, step="Init baseline image version")
 
         self.eventuallyEqual(lambda: k8s.get_statefulset_image(), SPILO_CURRENT, "Stagefulset not updated initially")        
-        
         self.eventuallyEqual(lambda: k8s.count_running_pods(), 2, "No 2 pods running")
         self.eventuallyEqual(lambda: len(k8s.get_patroni_running_members(pod0)), 2, "Postgres status did not enter running")
 

pkg/cluster/pod.go

@@ -304,9 +304,16 @@ func (c *Cluster) isSafeToRecreatePods(pods *v1.PodList) bool {
 	 after this check succeeds but before a pod is re-created
 	*/
 
+	for _, pod := range pods.Items {
+		c.logger.Debugf("name=%s phase=%s ip=%s", pod.Name, pod.Status.Phase, pod.Status.PodIP)
+	}
+
 	for _, pod := range pods.Items {
 		state, err := c.patroni.GetPatroniMemberState(&pod)
-		if err != nil || state == "creating replica" {
+		if err != nil {
+			c.logger.Errorf("failed to get Patroni state for pod: %s", err)
+			return false
+		} else if state == "creating replica" {
 			c.logger.Warningf("cannot re-create replica %s: it is currently being initialized", pod.Name)
 			return false
 		}