k3s/tests/docker/test-helpers

#!/bin/bash

# ---

port-used() {
    netstat -tuna | grep -q ":$1 "
}
export -f port-used

# ---

get-port() {
    local port=0
    while
        port=$((10000 + RANDOM % 50000))
        port-used $port
    do continue; done
    echo $port
}
export -f get-port

# ---

fetch-kubeconfig() {(
    set -e -o pipefail
    local num=${1:-1}
    local name=$(cat $TEST_DIR/servers/$num/metadata/name)
    local port=$(cat $TEST_DIR/servers/$num/metadata/port)
    docker cp $name:/etc/rancher/k3s/k3s.yaml - 2>/dev/null | tar -xO 2>/dev/null | sed -e "s/:6443/:$port/g" >$TEST_DIR/servers/$num/kubeconfig.yaml
)}
export -f fetch-kubeconfig

# ---

wait-for-kubeconfig() {
    while ! fetch-kubeconfig $1; do
        echo 'Waiting for kubeconfig to become available...' >&2
        sleep 5
    done
}
export -f wait-for-kubeconfig

# ---

count-ready-nodes() {
    kubectl get nodes -o json \
        | jq '.items[].status.conditions[] | select(.type == "Ready" and .status == "True") | .type' \
        | wc -l \
        | tr -d '[:space:]'
}
export -f count-ready-nodes

# ---

wait-for-nodes() {
    while [[ $(count-ready-nodes) -ne $1 ]]; do
        echo 'Waiting for nodes to be ready...' >&2
        sleep 5
    done
}
export -f wait-for-nodes

# ---

pod-ready() {
    kubectl get pods -n kube-system -o json \
        | jq ".items[].status | select(.containerStatuses != null) | .containerStatuses[] | select(.name == \"$1\") | .ready" 2>/dev/null
}
export -f pod-ready

# ---

wait-for-services() {
    for service in "$@"; do
        while [[ "$(pod-ready $service | sort -u)" != 'true' ]]; do
            echo "Waiting for service $service to be ready..." >&2
            sleep 5
        done
        echo "Service $service is ready"
    done
}
export -f wait-for-services

# ---

wait-for-db-connection() {
    if [ -z "$DB_CONNECTION_TEST" ]; then
      echo 'DB_CONNECTION_TEST is not defined' >&2
      return 1
    fi
    while ! $DB_CONNECTION_TEST 2>/dev/null; do
        echo 'Waiting for database to become available...' >&2
        sleep 5
    done
}
export -f wait-for-db-connection

# ---

verify-valid-version() {
    docker exec "$@" 2>&1 | tee .version.tmp
    # check for bad strings in the version output, including '.' in the build metadata
    if grep -oiE '.*(dev|head|unknown|fail|refuse|\+[^"]*\.).*' .version.tmp; then
        return 1
    fi
}
export -f verify-valid-version

# ---

verify-valid-versions() {
    verify-valid-version $1 kubectl version
    verify-valid-version $1 ctr version
    verify-valid-version $1 crictl version
}
export -f verify-valid-versions

# ---

dump-logs() {
    local testID=$(basename $TEST_DIR)
    echo "#---------------------------------"
    echo "#- Begin: logs for run ($testID)"
    echo
    local server=$(cat $TEST_DIR/servers/1/metadata/name)
    docker exec $server kubectl get pods -A -o wide >$TEST_DIR/logs/kubectl-get-pods.txt
    docker exec $server kubectl get nodes -o wide >$TEST_DIR/logs/kubectl-get-nodes.txt
    docker exec $server kubectl describe pods -A >$TEST_DIR/logs/kubectl-describe-pods.txt

    for node in $TEST_DIR/*/*; do
        [ -d "$node" ] || continue
        local name=$(cat $node/metadata/name 2>/dev/null)
        [ "$name" ] || continue
        mkdir -p $node/logs
        local hostname=$(docker exec $name hostname)
        docker logs $name >$node/logs/system.log 2>&1
        if [[ ! -z "$hostname" && $name == k3s-* ]]; then
            docker exec $server kubectl describe node/$hostname >$node/logs/kubectl-describe-node.txt
            docker cp $name:/var/lib/rancher/k3s/agent/containerd/containerd.log $node/logs/containerd.log 2>/dev/null
            docker exec $name crictl pods >$node/logs/crictl-pods.txt
            docker exec $name crictl ps -a >$node/logs/crictl-ps.txt
            docker exec $name crictl ps -a -o json >$node/metadata/crictl-ps.json
            for container in $(jq -r '.containers[].id' <$node/metadata/crictl-ps.json); do
                local cname=$(jq -r '.containers[] | select(.id == "'$container'") | .metadata.name' <$node/metadata/crictl-ps.json)
                docker exec $name crictl logs $container >$node/logs/$cname-$container.log 2>&1
            done
        fi
        if [ "$1" == "skip-output" ]; then
            continue
        fi 
        for log in $node/logs/*.log; do
            echo
            echo "#- Tail: $log"
            tail -5 $log
            echo "#- Done: $log"
            echo
        done
    done

    if [ "$1" == "skip-output" ]; then
        return
    fi 

    for txt in $TEST_DIR/logs/*.txt $TEST_DIR/*/*/logs/*.txt; do
        echo
        echo "#- Cat: $txt"
        cat $txt
        echo "#- Done: $txt"
        echo
    done
    echo
    echo "#- Finish: logs for run ($testID)"
    echo "#---------------------------------"
    echo
    ./scripts/log-upload $TEST_DIR
}
export -f dump-logs

# ---

retrieve-sonobuoy-logs() {
    local status=passed
    local code=0
    local testStatus=$(sonobuoy status 2>&1)
    cat <<< $testStatus

    if ! grep -q -E '\s+e2e\s+complete\s+passed\s+' <<< $testStatus; then
        status=failed
        code=1
    fi

    mkdir -p $TEST_DIR/sonobuoy
    local logTarball="$(sonobuoy retrieve $TEST_DIR/sonobuoy)"
    if [ -f "$logTarball" ]; then
        tar -xz -f "$logTarball" -C $TEST_DIR/sonobuoy
        rm "$logTarball"
    else
        rm -rf $TEST_DIR/sonobuoy
    fi

    local e2eLog=$TEST_DIR/sonobuoy/plugins/e2e/results/global/e2e.log
    if [ ! -s $e2eLog ]; then
        return 1
    fi
    if [ -n "$LOG_OUTPUT" ]; then
        cp $e2eLog $(sed -e "s/-STATUS-/-$status-/g" <<< "$LOG_OUTPUT")
    fi
    tail -11 $e2eLog
    awk '/^Summarizing .* Failures?:$/,0' $e2eLog
    return $code
}
export -f retrieve-sonobuoy-logs

# ---

test-wait() {
    wait $1 || echo "test wait exit code $?"
    local delay=15
    sleep $delay
    while sonobuoy status | grep -q -E ' +e2e +running +'; do
        sleep $delay
    done
    sleep $delay
    if sonobuoy status | grep -q -E ' +e2e +complete +passed +'; then
        return 0
    fi
    return 1
}
export -f test-wait

# ---

sonobuoy-test() {
    if [ "$ARCH" = 'arm' ]; then
        echo "Aborting sonobuoy tests, images not available for $ARCH"
        return 0
    fi
    echo 'Starting sonobuoy tests'

    sonobuoy run \
        --config=scripts/sonobuoy-config.json \
        --plugin-env=e2e.E2E_USE_GO_RUNNER=true \
        --kubernetes-version=${VERSION_K8S} \
        --wait=90 \
        "$@" &

    local sonobuoyPID=$!
    local code=0
    time timeout --foreground 60m bash -c test-wait $sonobuoyPID || code=$?
    echo "Sonobuoy finished with code $code"
    retrieve-sonobuoy-logs
    return $code
}
export -f sonobuoy-test

# ---

test-cleanup() {
    local code=$?
    set +e -x
    echo 'Cleaning up...'
    trap - EXIT INT TERM
    if has-function test-post-hook; then
        test-post-hook $code
        code=$?
    fi
    if [[ $code -ne 0 ]]; then
        dump-logs
    fi
    for name in $TEST_DIR/*/*/metadata/name; do
        [ -f "$name" ] || continue
        local container=$(cat $name)
        echo "Removing container $container"
        docker rm -f -v $container
    done
    if has-function test-cleanup-hook; then
        test-cleanup-hook
    fi
    echo
    if [ "$TEST_CLEANUP" = true ]; then
        echo "Removing test directory $TEST_DIR"
        rm -rf $TEST_DIR
    fi
    [ -f "$PROVISION_LOCK" ] && rm $PROVISION_LOCK
    echo -n "Test $(basename $TEST_DIR) "
    if [ $code -eq 0 ]; then
        echo "passed."
    else
        echo "failed."
    fi
    echo
    exit $code
}
export -f test-cleanup

# ---

test-setup() {
    export TEST_DIR=$(mktemp -d '/tmp/XXXXXX')
    trap test-cleanup EXIT INT TERM

    mkdir -p $TEST_DIR/metadata
    if [ "$LABEL" ]; then
        exec > >(awk "{ printf \"[\033[36m${LABEL}\033[m] %s\n\", \$0; fflush() }") \
            2> >(awk "{ printf \"[\033[35m${LABEL}\033[m] %s\n\", \$0; fflush() }" >&2)
        echo "$LABEL" >$TEST_DIR/metadata/label
    fi

    mkdir -p $TEST_DIR/logs
    exec > >(tee -a $TEST_DIR/logs/test.log) \
        2> >(tee -a $TEST_DIR/logs/test.log >&2)

    if [ -z "$K3S_IMAGE" ]; then
        . ./scripts/version.sh
        TAG=${TAG:-${VERSION_TAG}${SUFFIX}}
        REPO=${REPO:-rancher}
        IMAGE_NAME=${IMAGE_NAME:-k3s}
        export K3S_IMAGE=${REPO}/${IMAGE_NAME}:${TAG}
    fi

    if [ -z "$K3S_IMAGE" ]; then
        echo 'K3S_IMAGE environment variable should be defined'
        return 1
    fi

    SERVER_MINOR=$(awk -F. '{print $2}' <<<${K3S_IMAGE_SERVER:-$K3S_IMAGE})
    AGENT_MINOR=$(awk -F. '{print $2}' <<<${K3S_IMAGE_AGENT:-$K3S_IMAGE})
    if [ $NUM_AGENTS -gt 0 ] && [ $AGENT_MINOR -gt $SERVER_MINOR ]; then
        echo "Agent minor version cannot be higher than server - not supported by Kubernetes version skew policy"
        exit 0
    fi

    echo ${RANDOM}${RANDOM}${RANDOM} >$TEST_DIR/metadata/secret
}
export -f test-setup

# ---

inc-count() {
    local count=$(find $TEST_DIR -type d -mindepth 2 -maxdepth 2 -regex ".*/$1/[0-9]+" -printf '%f\n' | sort -nr | head -1)
    count=$((count+1))
    mkdir -p $TEST_DIR/$1/$count/metadata
    echo $count
}
export -f inc-count

# ---

has-function() {
  [[ -n "$1" && $(type -t $1) == "function" ]]
} 2> /dev/null
export -f has-function

# ---

run-function() {
    has-function $1 || return 0
    "$@"
}
export -f run-function

# ---

provision-server() {
    local count=$(inc-count servers)
    local testID=$(basename $TEST_DIR)
    local name=$(echo "k3s-server-${count}-${testID,,}" | tee $TEST_DIR/servers/$count/metadata/name)
    local port=$(timeout --foreground 5s bash -c get-port | tee $TEST_DIR/servers/$count/metadata/port)
    local SERVER_INSTANCE_ARGS="SERVER_${count}_ARGS"

    run-function server-pre-hook $count

    docker run \
        -d --name $name \
        --hostname $name \
        --privileged \
        -p 127.0.0.1:$port:6443 \
        -p 6443 \
        -e K3S_TOKEN="$(cat $TEST_DIR/metadata/secret)" \
        -e K3S_DEBUG=true \
        ${SERVER_DOCKER_ARGS:-} \
        ${REGISTRY_CLUSTER_ARGS:-} \
        ${K3S_IMAGE_SERVER:-$K3S_IMAGE} server ${ARGS} ${SERVER_ARGS} ${!SERVER_INSTANCE_ARGS}

    local ip=$(docker inspect --format '{{ .NetworkSettings.IPAddress }}' $name | tee $TEST_DIR/servers/$count/metadata/ip)
    local url=$(echo "https://$ip:6443" | tee $TEST_DIR/servers/$count/metadata/url)

    echo "Started $name @ $url"
    run-function server-post-hook $count
}
export -f provision-server

# ---

provision-agent() {
    local K3S_URL=${K3S_URL:-"$(cat $TEST_DIR/servers/1/metadata/url)"}
    local count=$(inc-count agents)
    local testID=$(basename $TEST_DIR)
    local name=$(echo "k3s-agent-${count}-${testID,,}" | tee $TEST_DIR/agents/$count/metadata/name)
    local AGENT_INSTANCE_ARGS="AGENT_${count}_ARGS"

    run-function agent-pre-hook $count
    docker run \
        -d --name $name \
        --hostname $name \
        --privileged \
        -e K3S_TOKEN=$(cat $TEST_DIR/metadata/secret) \
        -e K3S_URL=$K3S_URL \
        ${AGENT_DOCKER_ARGS:-} \
        ${REGISTRY_CLUSTER_ARGS:-} \
        ${K3S_IMAGE_AGENT:-$K3S_IMAGE} agent ${ARGS} ${AGENT_ARGS} ${!AGENT_INSTANCE_ARGS}

    local ip=$(docker inspect --format '{{ .NetworkSettings.IPAddress }}' $name | tee $TEST_DIR/agents/$count/metadata/ip)

    echo "Started $name"
    run-function agent-post-hook $count
}
export -f provision-agent

# ---

provision-cluster() {
    run-function cluster-pre-hook

    if [ "${ENABLE_REGISTRY}" == 'true' ]; then
      provision-registry-proxy
    fi

    for i in $(seq 1 $NUM_SERVERS); do
        provision-server
        timeout --foreground 120s bash -c "wait-for-kubeconfig $i"
    done
    export KUBECONFIG=$TEST_DIR/servers/${KUBECONFIG_SERVER:-1}/kubeconfig.yaml

    if [ $NUM_AGENTS -gt 0 ]; then
        for _ in $(seq 1 $NUM_AGENTS); do
            provision-agent
        done
    fi

    [ -f "$PROVISION_LOCK" ] && rm $PROVISION_LOCK

    timeout --foreground 2m bash -c "wait-for-nodes $(( NUM_SERVERS + NUM_AGENTS ))"
    timeout --foreground 4m bash -c "wait-for-services $WAIT_SERVICES"

    run-function cluster-post-hook
}
export -f provision-cluster

# ---

provision-registry-proxy() {
    set -e -o pipefail
    local image="docker.io/library/registry:2.8.1"
    local prefix="docker-registry-"
    local registries="docker.io:registry-1.docker.io registry.k8s.io gcr.io quay.io ghcr.io"
    local registries_yaml="$TEST_DIR/registries.yaml"

    echo "mirrors:" > $registries_yaml

    for registry in $registries; do
        IFS=: read registry_name registry_endpoint <<< $registry
        if [ -z "$registry_endpoint" ]; then
            registry_endpoint=$registry_name
        fi

        local name="registry_${registry_name//./_}"
        local status=$(docker inspect $name --format '{{ .State.Status }} {{ .Config.Image }} {{ (index .HostConfig.PortBindings "5000/tcp" 0).HostPort }}' 2>/dev/null || true)

        read state_status config_image hostport <<< $status
        if [ "$state_status" != "running" ] || [ "$config_image" != "$image" ]; then
            hostport=$(timeout --foreground 5s bash -c get-port)
            docker rm --force $name 2>/dev/null || true
            docker run \
                -d --name $name \
                -p 0.0.0.0:$hostport:5000 \
                -v "registry-cache:/var/lib/registry" \
                -e "REGISTRY_HTTP_SECRET=shared-secret" \
                -e "REGISTRY_PROXY_REMOTEURL=https://$registry_endpoint" \
                -e "REGISTRY_STORAGE_CACHE_BLOBDESCRIPTOR=inmemory" \
                -e "REGISTRY_STORAGE_FILESYSTEM_ROOTDIRECTORY=/var/lib/registry/$registry_name" \
                $image
        fi
        echo -e "  $registry_name:\n    endpoint:\n    - http://172.17.0.1:$hostport" >> $registries_yaml
    done

    echo "Using registry mirror with cluster registries.yaml:"
    cat $registries_yaml

    export REGISTRY_CLUSTER_ARGS="-v $registries_yaml:/etc/rancher/k3s/registries.yaml"
}
export -f provision-registry-proxy

# ---

early-exit() {
    printf "\033[33m%s\033[m\n" "$1"
    exit $2
}
export -f early-exit

# ---

run-test() {
    local delay=15
    (
      set +x
      while [ $(count-running-tests) -ge ${MAX_CONCURRENT_TESTS:-4} ]; do
          sleep $delay
      done
    )

    export PROVISION_LOCK=$(mktemp)
    ./tests/docker/test-runner "$@" &
    pids+=($!)

    (
        set +x
        # busy-wait on the provisioning lock before imposing a final inter-test delay
        while [ -f "$PROVISION_LOCK" ]; do
            sleep 1
        done
        sleep $delay
    )
}
export -f run-test

# ---

cleanup-test-env(){
      export NUM_SERVERS=1
      export NUM_AGENTS=1
      export AGENT_ARGS=''
      export SERVER_ARGS=''
      export WAIT_SERVICES="${all_services[*]}"

      unset AGENT_1_ARGS AGENT_2_ARGS AGENT_3_ARGS AGENT_DOCKER_ARGS K3S_IMAGE_AGENT
      unset SERVER_1_ARGS SERVER_2_ARGS SERVER_3_ARGS SERVER_DOCKER_ARGS K3S_IMAGE_SERVER

      unset -f server-pre-hook server-post-hook agent-pre-hook agent-post-hook cluster-pre-hook cluster-post-hook test-post-hook test-cleanup-hook
}

# ---

count-running-tests(){
      local count=0
      for pid in "${pids[@]}"; do
          if [ $(pgrep -c -P $pid) -gt 0 ]; then
            ((count++))
          fi
      done
      echo "Currently running ${count} tests" 1>&2
      echo ${count}
}
export -f count-running-tests

# ---

e2e-test() {
    local label=$label
    if [ -n "$LABEL_SUFFIX" ]; then
        label="$label-$LABEL_SUFFIX"
    fi
    local logOutput=
    if [ -n "$E2E_OUTPUT" ]; then
        logOutput=$E2E_OUTPUT/$logName
    fi
    if [[ $label =~ ^PARALLEL.* ]]; then
        LABEL=$label LOG_OUTPUT=$logOutput MAX_CONCURRENT_TESTS=3 run-test "$@"
    else
        LABEL=$label LOG_OUTPUT=$logOutput run-test "$@"
    fi

}

# ---

test-run-sonobuoy() {
    local suffix
    if [ "$1" ] && [[ ! $1 =~ ^(serial|parallel)$ ]] ; then
        suffix="-$1"
        export LABEL_SUFFIX=$1
    else
        unset LABEL_SUFFIX
    fi

    cleanup-test-env
    . ./tests/docker/test-setup-sonobuoy$suffix
    
    if [ "$1" = "parallel" ] || [ "$2" = "parallel" ]; then
        label=PARALLEL \
            logName=e2e-STATUS-${ARCH}-parallel.log \
            e2e-test "${sonobuoyParallelArgs[@]}"
        echo "Exit code $? for parallel start"
    fi

    if [ "$1" = "serial" ] || [ "$2" = "serial" ]; then
        label=SERIAL \
            logName=e2e-STATUS-${ARCH}-serial.log \
            e2e-test "${sonobuoySerialArgs[@]}"
        echo "Exit code $? for serial start"
    fi
}
export -f test-run-sonobuoy

# ---

pid-cleanup() {
    local code=$?
    local failCount=0
    set +e
    if [ $code -eq 0 ]; then
        for pid in "${pids[@]}"; do
            wait $pid || code=$?
        done
    fi
    if [ $code -ne 0 ]; then
        for pid in "${pids[@]}"; do
            pkill -P $pid
            wait $pid || failCount=$((failCount+1))
        done
    fi
    trap - EXIT INT TERM
    set +x
    echo
    if [ $failCount -eq 0 ]; then
        printf '\033[32mAll tests passed.\033[m\n'
    else
        printf "\033[31m%s tests failed.\033[m\n" "$failCount"
        if [ "$DRONE_BUILD_EVENT" = 'tag' ]; then
            printf "\033[31mIgnoring test failures on tag.\033[m\n"
            code=0
        else
            code=1
        fi
    fi
    echo
    exit $code
}
export -f pid-cleanup

# ---

pids=()
trap pid-cleanup EXIT INT TERM