diff --git a/heartbeat/podman-etcd b/heartbeat/podman-etcd index f53b1c2ae..bfac546d8 100755 --- a/heartbeat/podman-etcd +++ b/heartbeat/podman-etcd @@ -1,1551 +1,1588 @@ #!/bin/sh # # The podman etcd HA resource agent creates and launches a etcd podman # container based off a supplied podman image. Containers managed by # this agent are both created and removed upon the agent's start and # stop actions. # # Based on the podman resource agent. # # Copyright (c) 2014 David Vossel # Michele Baldessari # All Rights Reserved. # # This program is free software; you can redistribute it and/or modify # it under the terms of version 2 of the GNU General Public License as # published by the Free Software Foundation. # # This program is distributed in the hope that it would be useful, but # WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. # # Further, this software is distributed without any warranty that it is # free of the rightful claim of any third person regarding infringement # or the like. Any license provided herein, whether implied or # otherwise, applies only to this software file. Patent licenses, if # any, provided herein do not apply to combinations of this program with # other software, or any other product whatsoever. # # You should have received a copy of the GNU General Public License # along with this program; if not, write the Free Software Foundation, # Inc., 59 Temple Place - Suite 330, Boston MA 02111-1307, USA. # ####################################################################### # Initialization: : ${OCF_FUNCTIONS_DIR=${OCF_ROOT}/lib/heartbeat} . ${OCF_FUNCTIONS_DIR}/ocf-shellfuncs # Parameter defaults OCF_RESKEY_image_default="default" OCF_RESKEY_pod_manifest_default="/etc/kubernetes/static-pod-resources/etcd-certs/configmaps/external-etcd-pod/pod.yaml" OCF_RESKEY_name_default="etcd" OCF_RESKEY_nic_default="br-ex" OCF_RESKEY_authfile_default="/var/lib/kubelet/config.json" OCF_RESKEY_allow_pull_default="1" OCF_RESKEY_reuse_default="0" : ${OCF_RESKEY_image=${OCF_RESKEY_image_default}} : ${OCF_RESKEY_pod_manifest=${OCF_RESKEY_pod_manifest_default}} : ${OCF_RESKEY_name=${OCF_RESKEY_name_default}} : ${OCF_RESKEY_nic=${OCF_RESKEY_nic_default}} : ${OCF_RESKEY_authfile=${OCF_RESKEY_authfile_default}} : ${OCF_RESKEY_allow_pull=${OCF_RESKEY_allow_pull_default}} : ${OCF_RESKEY_reuse=${OCF_RESKEY_reuse_default}} ####################################################################### meta_data() { cat < 1.0 The podman-etcd HA resource agent creates and launches a etcd podman container based off a supplied podman image. Containers managed by this agent are both created and removed upon the agent's start and stop actions. Podman etcd container resource agent. The Pod manifest with the configuration for Etcd. Etcd pod manifest The podman image to base this container off of. podman image The name to give the created container. By default this will be that resource's instance name. podman container name A mapping of node names to IPs. This takes the form of: n1:ip1;n2:ip2 where the etcd container on n1 would have IP ip1 Container node name to IP mapping Network interface to lookup interface for host. Network interface Path of the authentication file. The file is created by podman login. Path of the authentication file Allow the image to be pulled from the configured podman registry when the image does not exist locally. NOTE, this can drastically increase the time required to start the container if the image repository is pulled over the network. Allow pulling non-local images Add options to be appended to the 'podman run' command which is used when creating the container during the start action. This option allows users to do things such as setting a custom entry point and injecting environment variables into the newly created container. Note the '-d' option is supplied regardless of this value to force containers to run in the background. NOTE: Do not explicitly specify the --name argument in the run_opts. This agent will set --name using either the resource's instance or the name provided in the 'name' argument of this agent. run options Specify a command to launch within the container once it has initialized. run command Options to be added to the 'run_cmd'. run command options A comma separated list of directories that the container is expecting to use. The agent will ensure they exist by running 'mkdir -p' Required mount points Specify the full path of a command to launch within the container to check the health of the container. This command must return 0 to indicate that the container is healthy. A non-zero return code will indicate that the container has failed and should be recovered. Note: Using this method for monitoring processes inside a container is not recommended, as containerd tries to track processes running inside the container and does not deal well with many short-lived processes being spawned. Ensure that your container monitors its own processes and terminates on fatal error rather than invoking a command from the outside. monitor command Kill a container immediately rather than waiting for it to gracefully shutdown force kill Allow the container to be reused once it is stopped. By default, containers get removed once they are stopped. Enable this option to have the particular one persist when this happens. reuse container Use transient drop-in files to add extra dependencies to the systemd scopes associated to the container. During reboot, this prevents systemd to stop the container before pacemaker. drop-in dependency END } ####################################################################### REQUIRE_IMAGE_PULL=0 podman_usage() { cat <&1) rc=$? # 125: no container with name or ID ${CONTAINER} found # 126: container state improper (not running) # 127: any other error # 255: podman 2+: container not running case "$rc" in 125|126|255) rc=$OCF_NOT_RUNNING ;; 0) ocf_log debug "monitor cmd passed: exit code = $rc" ;; *) ocf_exit_reason "monitor cmd failed (rc=$rc), output: $out" rc=$OCF_ERR_GENERIC ;; esac return $rc } container_exists() { local rc local out out=$(podman exec ${CONTAINER} $OCF_RESKEY_monitor_cmd 2>&1) rc=$? # 125: no container with name or ID ${CONTAINER} found if [ $rc -ne 125 ]; then return 0 fi return 1 } remove_container() { local rc local execids if ocf_is_true "$OCF_RESKEY_reuse"; then # never remove the container if we have reuse enabled. return 0 fi container_exists if [ $? -ne 0 ]; then # don't attempt to remove a container that doesn't exist return 0 fi ocf_log notice "Cleaning up inactive container, ${CONTAINER}." ocf_run podman rm -v $CONTAINER rc=$? if [ $rc -ne 0 ]; then if [ $rc -eq 2 ]; then if podman inspect --format '{{.State.Status}}' $CONTAINER | grep -wq "stopping"; then ocf_log err "Inactive container ${CONTAINER} is stuck in 'stopping' state. Force-remove it." ocf_run podman rm -f $CONTAINER rc=$? fi fi # due to a podman bug (rhbz#1841485), sometimes a stopped # container can still be associated with Exec sessions, in # which case the "podman rm" has to be forced execids=$(podman inspect $CONTAINER --format '{{len .ExecIDs}}') if [ "$execids" -ne "0" ]; then ocf_log warn "Inactive container ${CONTAINER} has lingering exec sessions. Force-remove it." ocf_run podman rm -f $CONTAINER rc=$? fi fi return $rc } attribute_node_ip() { local action="$1" local attribute="node_ip" local value if ! value=$(ip -brief addr show "$OCF_RESKEY_nic" | awk '{gsub("/.*", "", $3); print $3}'); then rc=$? ocf_log error "could not get node ip: error $rc" return "$rc" fi case "$action" in get) echo "$value" ;; update) if ! crm_attribute --type nodes --node "$NODENAME" --name "$attribute" --update "$value"; then rc="$?" ocf_log error "could not set $attribute to $value: error $rc" return "$rc" fi ;; clear) crm_attribute --name "$attribute" --delete ;; *) ocf_log error "unsupported $action for $attribute" return $OCF_ERR_GENERIC ;; esac } attribute_node_ip_peer() { local peer_name peer_name=$(get_peer_node_name) crm_attribute --query --name "node_ip" --node "$peer_name" | awk -F"value=" '{print $2}' } get_env_from_manifest() { local env_var_name="$1" local env_var_value if [ ! -f "$OCF_RESKEY_pod_manifest" ]; then ocf_log error "external etcd pod manifest ($OCF_RESKEY_pod_manifest) not found" return "$OCF_ERR_GENERIC" fi if ! env_var_value=$(jq -r ".spec.containers[].env[] | select( .name == \"$env_var_name\" ).value" "$OCF_RESKEY_pod_manifest"); then return "$?" fi ocf_log debug "ETCD pod environment variable $env_var_name: $env_var_value" echo "$env_var_value" } prepare_env() { local name ip standalone_node NODEIP="$(attribute_node_ip get)" if is_force_new_cluster; then ALL_ETCD_ENDPOINTS="https://$NODEIP:2379" ETCD_INITIAL_CLUSTER_STATE="new" ETCD_INITIAL_CLUSTER="$NODENAME=https://$NODEIP:2380" else ETCD_INITIAL_CLUSTER_STATE="existing" for node in $(echo "$OCF_RESKEY_node_ip_map" | sed "s/\s//g;s/;/ /g"); do name=$(echo "$node" | awk -F":" '{print $1}') ip=$(echo "$node" | awk -F":" '{print $2}') if [ -z "$name" ] || [ -z "$ip" ]; then ocf_exit_reason "name or ip missing for 1 or more nodes" exit $OCF_ERR_CONFIGURED fi [ -z "$ALL_ETCD_ENDPOINTS" ] && ALL_ETCD_ENDPOINTS="https://$ip:2379" || ALL_ETCD_ENDPOINTS="$ALL_ETCD_ENDPOINTS,https://$ip:2379" [ -z "$ETCD_INITIAL_CLUSTER" ] && ETCD_INITIAL_CLUSTER="$name=https://$ip:2380" || ETCD_INITIAL_CLUSTER="$ETCD_INITIAL_CLUSTER,$name=https://$ip:2380" done fi ETCDCTL_API=$(get_env_from_manifest "ETCDCTL_API") ETCD_CIPHER_SUITES=$(get_env_from_manifest "ETCD_CIPHER_SUITES") ETCD_DATA_DIR=$(get_env_from_manifest "ETCD_DATA_DIR") ETCD_ELECTION_TIMEOUT=$(get_env_from_manifest "ETCD_ELECTION_TIMEOUT") ETCD_ENABLE_PPROF=$(get_env_from_manifest "ETCD_ENABLE_PPROF") ETCD_EXPERIMENTAL_MAX_LEARNERS=$(get_env_from_manifest "ETCD_EXPERIMENTAL_MAX_LEARNERS") ETCD_EXPERIMENTAL_WARNING_APPLY_DURATION=$(get_env_from_manifest "ETCD_EXPERIMENTAL_WARNING_APPLY_DURATION") ETCD_EXPERIMENTAL_WATCH_PROGRESS_NOTIFY_INTERVAL=$(get_env_from_manifest "ETCD_EXPERIMENTAL_WATCH_PROGRESS_NOTIFY_INTERVAL") ETCD_HEARTBEAT_INTERVAL=$(get_env_from_manifest "ETCD_HEARTBEAT_INTERVAL") ETCD_QUOTA_BACKEND_BYTES=$(get_env_from_manifest "ETCD_QUOTA_BACKEND_BYTES") ETCD_SOCKET_REUSE_ADDRESS=$(get_env_from_manifest "ETCD_SOCKET_REUSE_ADDRESS") SERVER_CACERT=$(get_env_from_manifest "ETCDCTL_CACERT") ETCD_PEER_CERT=$(get_env_from_manifest "ETCDCTL_CERT") ETCD_PEER_KEY=$(get_env_from_manifest "ETCDCTL_KEY") if is_learner; then LISTEN_CLIENT_URLS="$NODEIP" LISTEN_PEER_URLS="$NODEIP" LISTEN_METRICS_URLS="$NODEIP" else LISTEN_CLIENT_URLS="0.0.0.0" LISTEN_PEER_URLS="0.0.0.0" LISTEN_METRICS_URLS="0.0.0.0" fi } archive_data_folder() { # TODO: use etcd snapshots local dest_dir_name="members-snapshot-$(date +%Y%M%d%H%M%S)" local data_dir="/var/lib/etcd/member" if [ ! -d $data_dir ]; then ocf_log info "no data dir to backup" return $OCF_SUCCESS fi ocf_log info "backing up $data_dir under $HA_RSCTMP/$dest_dir_name" mv $data_dir $HA_RSCTMP/$dest_dir_name sync } etcd_pod_container_exists() { local count_matches # Check whether the etcd pod exists on the same node (header line included) count_matches=$(crictl pods --label app=etcd -q | xargs -I {} crictl ps --pod {} -o json | jq -r '.containers[].metadata | select ( .name == "etcd" ).name' | wc -l) if [ "$count_matches" -eq 1 ]; then # etcd pod found return 0 fi # etcd pod not found return 1 } attribute_node_cluster_id() { local action="$1" local value if ! value=$(jq -r ".clusterId" /var/lib/etcd/revision.json); then rc=$? ocf_log error "could not get cluster-id: $rc" return "$rc" fi case "$action" in get) echo "$value" ;; update) ocf_log info "$NODENAME: update cluster-id: $value" if ! crm_attribute --type nodes --node "$NODENAME" --name "cluster_id" --update "$value"; then rc=$? ocf_log err "could not update cluster-id: error code $rc" return "$rc" fi ;; *) ocf_log error "unsupported $action for attribute_node_cluster_id" return $OCF_ERR_GENERIC ;; esac } attribute_node_cluster_id_peer() { local nodename=$(get_peer_node_name) crm_attribute --query --type nodes --node "$nodename" --name "cluster_id" | awk -F"value=" '{print $2}' } attribute_node_revision() { local action="$1" local value local attribute="revision" if ! value=$(jq -r ".maxRaftIndex" /var/lib/etcd/revision.json); then rc=$? ocf_log error "could not get $attribute: $rc" return "$rc" fi case "$action" in get) echo "$value" ;; update) ocf_log info "$NODENAME: update $attribute: $value" if ! crm_attribute --type nodes --node "$NODENAME" --name "$attribute" --update "$value"; then rc=$? ocf_log err "could not update etcd $revision: error code $rc" return "$rc" fi ;; *) ocf_log error "unsupported $action for attribute_node_revision" return "$OCF_ERR_GENERIC" ;; esac } attribute_node_revision_peer() { local nodename nodename=$(get_peer_node_name) crm_attribute --query --type nodes --node "$nodename" --name "revision" | awk -F"value=" '{print $2}' } attribute_node_member_id() { local action="$1" local attribute="member_id" if ! container_exists; then # we need a running container to execute etcdctl. return 0 fi case "$action" in get) # When we need this value at the agent startup we don't have a etcd # container running, so we always get this value from CIB crm_attribute --query --type nodes --node "$NODENAME" --name "$attribute" | awk -F"value=" '{print $2}' ;; update) local member_list_json member_list_json=$(get_member_list_json) ocf_log info "member list: $member_list_json" if [ -z "$member_list_json" ] ; then ocf_log error "could not get $attribute: could not get member list JSON" return "$rc" fi local value if ! value=$(echo -n "$member_list_json" | jq -r ".header.member_id"); then rc=$? ocf_log error "could not get $attribute from member list JSON: error $rc" return "$rc" fi # JSON member_id is decimal, while etcdctl command needs the hex version value=$(printf "%x" "$value") if ! crm_attribute --type nodes --node "$NODENAME" --name "$attribute" --update "$value"; then rc=$? ocf_log err "could not update etcd $attribute: error code $rc" return "$rc" fi ;; clear) crm_attribute --type nodes --node "$NODENAME" --name "$attribute" --delete ;; *) ocf_log error "unsupported $action for attribute_node_member_id" return "$OCF_ERR_GENERIC" ;; esac } add_member_as_learner() { local rc local member_name=$1 local member_ip=$2 ocf_log info "add $member_name ($member_ip) to the member list as learner" out=$(podman exec ${CONTAINER} etcdctl --endpoints="https://$(attribute_node_ip get):2379" member add $member_name --peer-urls=https://$member_ip:2380 --learner) rc=$? if [ $rc -ne 0 ]; then ocf_log error "could not add $member_name as learner: error code $rc" return $rc fi ocf_log info "$out" attribute_learner_node update "$member_name" return $? } set_force_new_cluster() { local rc crm_attribute --lifetime reboot --node "$NODENAME" --name "force_new_cluster" --update "$NODENAME" rc=$? if [ $rc -ne 0 ]; then ocf_log error "could not set force_new_cluster attribute to $NODENAME" fi return $rc } get_force_new_cluster() { crm_attribute --lifetime reboot --query --name "force_new_cluster" | awk -F"value=" '{print $2}' } clear_force_new_cluster() { local force_new_cluster_node=$(get_force_new_cluster) if [ -z "$force_new_cluster_node" ]; then ocf_log info "$NODENAME: force_new_cluster attribute not set" return $OCF_SUCCESS fi # only the holder of "force_new_cluster" attribute can delete it if [ "$NODENAME" = "$force_new_cluster_node" ]; then crm_attribute --lifetime reboot --name "force_new_cluster" --delete rc=$? if [ $rc -ne 0 ]; then ocf_log error "could not clear force_new_cluster attribute: error $rc" else ocf_log info "$NODENAME: force_new_cluster attribute cleared" fi return $rc else ocf_log info "$NODENAME does not hold force_new_cluster ($force_new_cluster_node has it)" return $OCF_SUCCESS fi } is_force_new_cluster() { # Return 0 if 'force_new_cluster' is set and the value matches the current node name, 1 otherwise. local value=$(get_force_new_cluster) if [ -z "$value" ]; then ocf_log debug "force_new_cluster attribute is not set" return 1 fi if [ "$value" = "$NODENAME" ]; then ocf_log debug "$NODENAME has force_new_cluster set" return 0 fi ocf_log info "force_new_cluster attribute set on peer node $value" return 1 } is_standalone() { local nodename=${1:-$NODENAME} local standalone_node=$(get_standalone_node) if [ -z "$standalone_node" ]; then ocf_log debug "no node running standalone" return 1 fi if [ "$nodename" = "$standalone_node" ]; then ocf_log debug "$nodename is set as standalone" return 0 fi ocf_log debug "$nodename is set as learner" return 1 } set_standalone_node() { local rc ocf_log info "add $NODENAME as standalone" crm_attribute --name "standalone_node" --update $NODENAME rc=$? if [ $rc -ne 0 ]; then ocf_log error "could not set standalone_node attribute to $NODENAME" fi return $rc } get_standalone_node() { crm_attribute --query --name "standalone_node" | awk -F"value=" '{print $2}' } clear_standalone_node() { crm_attribute --name "standalone_node" --delete } clear_standalone_and_learner_if_not_learners() { local rc local member_list_json="$1" number_of_members=$(echo -n $member_list_json | jq -r ".members[].ID" | wc -l) if [ $number_of_members -ne 2 ]; then ocf_log info "could not clear standalone_node, nor learner_node properties: found $number_of_members members, need 2" return $OCF_SUCCESS fi id=$(echo -n $member_list_json | jq -r ".members[] | select( .isLearner==true ).ID") rc=$? if [ $rc -ne 0 ]; then ocf_log error "could not get isLearner field from member list: error code $rc" return $rc fi if [ -z "$id" ]; then clear_standalone_node rc=$? if [ $rc -ne 0 ]; then ocf_og error "could not clear standalone_node attribute: error code $rc" return $rc fi fi if [ -z "$id" ]; then attribute_learner_node clear rc=$? if [ $rc -ne 0 ]; then ocf_og error "could not clear learner_node attribute: error code $rc" return $rc fi fi return $rc } attribute_learner_node() { local action="$1" local value="$2" local attribute="learner_node" case "$action" in get) crm_attribute --query --name "$attribute" | awk -F"value=" '{print $2}' ;; update) if ! crm_attribute --name "$attribute" --update "$value"; then rc="$?" ocf_log error "could not set $attribute to $value: error $rc" return "$rc" fi ;; clear) crm_attribute --name "$attribute" --delete ;; *) ocf_log error "unsupported $action for $attribute" return $OCF_ERR_GENERIC ;; esac } is_learner() { if [ "$NODENAME" == "$(attribute_learner_node get)" ]; then return 0 fi return 1 } get_peer_node_name() { crm_node -l | awk '{print $2}' | grep -v $NODENAME } get_all_etcd_endpoints() { for node in $(echo "$OCF_RESKEY_node_ip_map" | sed "s/\s//g;s/;/ /g"); do name=$(echo "$node" | awk -F":" '{print $1}') ip=$(echo "$node" | awk -F":" '{print $2}') if [ -z "$name" ] || [ -z "$ip" ]; then ocf_exit_reason "name or ip missing for 1 or more nodes" exit $OCF_ERR_CONFIGURED fi [ -z "$ALL_ETCD_ENDPOINTS" ] && ALL_ETCD_ENDPOINTS="https://$ip:2379" || ALL_ETCD_ENDPOINTS="$ALL_ETCD_ENDPOINTS,https://$ip:2379" done echo $ALL_ETCD_ENDPOINTS } get_endpoint_status_json() { # Get the status of all endpoints local all_etcd_endpoints=$(get_all_etcd_endpoints) podman exec ${CONTAINER} etcdctl endpoint status --endpoints=$all_etcd_endpoints -w json } get_member_list_json() { # Get the list of members visible to the current node local this_node_endpoint="https://$(attribute_node_ip get):2379" podman exec ${CONTAINER} etcdctl member list --endpoints=$this_node_endpoint -w json } check_peers() { # Check peers endpoint status and locally accessible member list local member_list_json if ! container_exists; then # we need a running container to execute etcdctl. return $OCF_SUCCESS fi member_list_json=$(get_member_list_json) rc=$? ocf_log debug "member list: $member_list_json" if [ $rc -ne 0 ]; then ocf_log info "podman failed to get member list: error code $rc" endpoint_status_json=$(get_endpoint_status_json) ocf_log info "endpoint status: $endpoint_status_json" count_endpoints=$(echo -n $endpoint_status_json | jq -r ".[].Endpoint" | wc -l) if [ $count_endpoints -eq 1 ]; then ocf_log info "one endpoint only: checking status errors" endpoint_status_errors=$(echo -n $endpoint_status_json | jq -r ".[0].Status.errors") if echo "$endpoint_status_errors" | grep -q "no leader"; then set_force_new_cluster set_standalone_node ocf_exit_reason "$NODENAME must force a new cluster" return $OCF_ERR_GENERIC fi if [ "$endpoint_status_errors" != "null" ]; then ocf_log error "unmanaged endpoint status error: $endpoint_status_errors" fi fi return $OCF_SUCCESS fi # Example of .members[] instance fields in member list json format: # NOTE that "name" is present in voting members only, while "isLearner" in learner members only # and the value is always true (not a string) in that case. # { # "ID": , # "name": "", # "peerURLs": [ # "https://:2380" # ], # "clientURLs": [ # "https://:2379" # ] # } for node in $(echo "$OCF_RESKEY_node_ip_map" | sed "s/\s//g;s/;/ /g"); do name=$(echo "$node" | awk -F":" '{print $1}') # do not check itself if [ "$name" == "$NODENAME" ]; then continue fi # Check by IP instead of Name since "learner" members appear only in peerURLs, not by Name. ip=$(echo "$node" | awk -F":" '{print $2}') id=$(echo -n $member_list_json | jq -r ".members[] | select( .peerURLs | map(test(\"$ip\")) | any).ID") if [ -z "$id" ]; then ocf_log info "$name is not in the members list" add_member_as_learner $name $ip set_standalone_node else ocf_log debug "$name is in the members list by IP: $ip" clear_standalone_and_learner_if_not_learners "$member_list_json" fi done return $OCF_SUCCESS } podman_simple_status() { local rc # simple status is implemented via podman exec # everything besides success is considered "not running" monitor_cmd_exec rc=$? if [ $rc -ne $OCF_SUCCESS ]; then rc=$OCF_NOT_RUNNING; fi return $rc } podman_monitor() { # We rely on running podman exec to monitor the container # state because that command seems to be less prone to # performance issue under IO load. # # For probes to work, we expect cmd_exec to be able to report # when a container is not running. Here, we're not interested # in distinguishing whether it's stopped or non existing # (there's function container_exists for that) monitor_cmd_exec rc=$? if [ $rc -ne 0 ]; then return $rc fi if is_learner; then ocf_log info "$NODENAME is learner. Cannot get member id" return "$OCF_SUCCESS" fi # Failing to cache data and check member list should not cause the # monitor operation to fail. # TODO: move this inside check_peers where we already query member list json attribute_node_member_id update if ! check_peers; then return $OCF_ERR_GENERIC fi # node revision comes from the disk, so if it is not available is a fatal failure attribute_node_revision update return $? } podman_create_mounts() { oldIFS="$IFS" IFS="," for directory in $OCF_RESKEY_mount_points; do mkdir -p "$directory" done IFS="$oldIFS" } podman_container_id() { # Retrieve the container ID by doing a "podman ps" rather than # a "podman inspect", because the latter has performance issues # under IO load. # We could have run "podman start $CONTAINER" to get the ID back # but if the container is stopped, the command will return a # name instead of a container ID. This would break us. podman ps --no-trunc --format '{{.ID}} {{.Names}}' | grep -F -w -m1 "$CONTAINER" | cut -d' ' -f1 } create_transient_drop_in_dependency() { local cid=$1 local rc=$OCF_SUCCESS if [ -z "$cid" ]; then ocf_exit_reason "Container ID not found for \"$CONTAINER\". Not creating drop-in dependency" return $OCF_ERR_GENERIC fi ocf_log info "Creating drop-in dependency for \"$CONTAINER\" ($cid)" for scope in "libpod-$cid.scope.d" "libpod-conmon-$cid.scope.d"; do if [ $rc -eq $OCF_SUCCESS ] && [ ! -d /run/systemd/transient/"$scope" ]; then mkdir -p /run/systemd/transient/"$scope" && \ printf "[Unit]\nBefore=pacemaker.service" > /run/systemd/transient/"$scope"/dep.conf && \ chmod ago+r /run/systemd/transient/"$scope" /run/systemd/transient/"$scope"/dep.conf rc=$? fi done if [ $rc -ne $OCF_SUCCESS ]; then ocf_log error "Could not create drop-in dependency for \"$CONTAINER\" ($cid)" else systemctl daemon-reload rc=$? if [ $rc -ne $OCF_SUCCESS ]; then ocf_log error "Could not refresh service definition after creating drop-in for \"$CONTAINER\"" fi fi return $rc } run_new_container() { local opts=$1 local image=$2 local cmd=$3 local rc ocf_log info "running container $CONTAINER for the first time" out=$(podman run $opts $image $cmd 2>&1) rc=$? if [ -n "$out" ]; then out="$(echo "$out" | tr -s ' \t\r\n' ' ')" if [ $rc -eq 0 ]; then ocf_log info "$out" else ocf_log err "$out" fi fi if [ $rc -eq 125 ]; then # If an internal podman error occurred, it might be because # the internal storage layer still references an old container # with the same name, even though podman itself thinks there # is no such container. If so, purge the storage layer to try # to clean the corruption and try again. if echo "$out" | grep -q "unknown.*flag"; then ocf_exit_reason "$out" return $rc fi ocf_log warn "Internal podman error while creating new container $CONTAINER. Retrying." ocf_run podman rm --storage $CONTAINER ocf_run podman run $opts $image $cmd rc=$? elif [ $rc -eq 127 ]; then # rhbz#1972209: podman 3.0.x seems to be hit by a race # where the cgroup is not yet set up properly when the OCI # runtime configures the container. If that happens, recreate # the container as long as we get the same error code or # until start timeout preempts us. while [ $rc -eq 127 ] && (echo "$out" | grep -q "cgroup.*scope not found") ; do ocf_log warn "Internal podman error while assigning cgroup. Retrying." # Arbitrary sleep to prevent consuming all CPU while looping sleep 1 podman rm -f "$CONTAINER" out=$(podman run $opts $image $cmd 2>&1) rc=$? done # Log the created container ID if it succeeded if [ $rc -eq 0 ]; then ocf_log info "$out" fi fi return $rc } compare_revision() { # Compare local revision (from disk) against peer revision (from CIB). # returns "older", "equal" or "newer" local revision local peer_node_name local peer_revision revision=$(attribute_node_revision get) peer_revision=$(attribute_node_revision_peer) if [ "$revision" == "" ] || [ "$revision" == "null" ] || [ "$peer_revision" == "" ] || [ "$peer_revision" == "null" ]; then ocf_log error "could not compare revisions: $NODENAME local revision: $revision, peer revision: $peer_revision" return "$OCF_ERR_GENERIC" fi if [ "$revision" -gt "$peer_revision" ]; then ocf_log info "$NODENAME revision: $revision is newer than peer revision: $peer_revision" echo "newer" elif [ "$revision" -eq "$peer_revision" ]; then ocf_log info "$NODENAME revision: $revision is equal to peer revision: $peer_revision" echo "equal" else ocf_log info "$NODENAME revision: $revision is older than peer revision: $peer_revision" echo "older" fi return "$OCF_SUCCESS" } +ensure_pod_manifest_exists() +{ + local wait_timeout_sec=$((10 * 60)) + local poll_interval_sec=5 + local poll_retries=$((wait_timeout_sec/poll_interval_sec)) + + for try in $(seq "$poll_retries"); do + if [ -f "$OCF_RESKEY_pod_manifest" ]; then + ocf_log info "pod manifest ($OCF_RESKEY_pod_manifest) found" + break + fi + ocf_log debug "pod manifest ($OCF_RESKEY_pod_manifest) does not exist yet: retry in $poll_interval_sec seconds." + sleep "$poll_interval_sec" + done + + if [ ! -f "$OCF_RESKEY_pod_manifest" ]; then + ocf_log error "pod manifest ($OCF_RESKEY_pod_manifest) still missing after $wait_timeout_sec seconds." + return "$OCF_ERR_CONFIGURED" + fi + + return "$OCF_SUCCESS" +} + podman_start() { local cid local rc local etcd_pod_wait_timeout_sec=$((10 * 60)) local etcd_pod_poll_interval_sec=10 local etcd_pod_poll_retries=$((etcd_pod_wait_timeout_sec/etcd_pod_poll_interval_sec)) local pod_was_running=false ocf_log notice "podman-etcd start" attribute_node_ip update attribute_node_cluster_id update attribute_node_revision update # ensure the etcd pod is not running before starting the container ocf_log info "ensure etcd pod is not running (retries: $etcd_pod_poll_retries, interval: $etcd_pod_poll_interval_sec)" for try in $(seq $etcd_pod_poll_retries); do if ! etcd_pod_container_exists; then break fi ocf_log info "etcd pod running: retry in $etcd_pod_poll_interval_sec seconds." pod_was_running=true sleep $etcd_pod_poll_interval_sec done if etcd_pod_container_exists; then ocf_exit_reason "etcd pod is still running after $etcd_pod_wait_timeout_sec seconds." return $OCF_ERR_GENERIC fi + if ! ensure_pod_manifest_exists; then + ocf_exit_reason "could not find etcd pod manifest ($OCF_RESKEY_pod_manifest)" + return "$OCF_ERR_GENERIC" + fi + # force-new-cluster property is a runtime-scoped flag that instructs the agent to force a new cluster-of-1. # Since this attribute is configured with a reboot-lifetime, it is automatically cleared when the machine reboots. # If the agent detects during its start that this property is set, it indicates that the flag was explicitly set # during the current node boot session, implying a deliberate request to recover the cluster. if ocf_is_true "$pod_was_running"; then ocf_log info "static pod was running: start normally" else if is_force_new_cluster; then ocf_log notice "$NODENAME marked to force-new-cluster" else # When the local agent starts, we can infer the cluster state by counting # how many agents are starting or already active: # - 1 active agent: it's the peer (we are just starting) # - 0 active agents, 1 starting: we are starting; the peer is not starting # - 0 active agents, 2 starting: both agents are starting simultaneously - local active_resource_counts=$(echo "$OCF_RESKEY_CRM_meta_notify_active_resource" | wc -w) - case "$active_resource_counts" in + local active_resources_count + active_resources_count=$(echo "$OCF_RESKEY_CRM_meta_notify_active_resource" | wc -w) + case "$active_resources_count" in 1) if [ "$(attribute_learner_node get)" == "$(get_peer_node_name)" ]; then ocf_log info "peer active but in learner mode: start normally" else - ocf_log info "peer active already: joining as learner" + ocf_log info "peer is active standalone: joining as learner" JOIN_AS_LEARNER=true fi ;; 0) + # we need to compare the revisions in any of the following branches + # so call the function only once here + if ! revision_compare_result=$(compare_revision); then + ocf_log error "could not compare revisions: error $?" + return "$OCF_ERR_GENERIC" + fi + # count how many agents are starting now - local start_resources_count=$(echo "$OCF_RESKEY_CRM_meta_notify_start_resource" | wc -w) + local start_resources_count + start_resources_count=$(echo "$OCF_RESKEY_CRM_meta_notify_start_resource" | wc -w) + case "$start_resources_count" in 1) - ocf_log debug "peer not starting: compare revisions" - if ! result=$(compare_revision); then - ocf_log error "could not compare revisions: error $?" - return "$OCF_ERR_GENERIC" - elif [ "$result" != "older" ]; then + ocf_log debug "peer not starting: ensure we can start a new cluster" + if [ "$revision_compare_result" != "older" ]; then # If our revision is the same as or newer than the peer's last saved # revision, and the peer agent isn't currently starting, we can # restore e-quorum by forcing a new cluster. set_force_new_cluster else - ocf_log error "local revision is older and peer is not starting: can not start" + ocf_log error "local revision is older and peer is not starting: cannot start" + ocf_exit_reason "local revision is older and peer is not starting: cannot start" return "$OCF_ERR_GENERIC" fi ;; 2) - ocf_log info "peer starting: compare revisions" - if ! result=$(compare_revision); then - ocf_log error "could not compare revisions: error $?" - return "$OCF_ERR_GENERIC" - elif [ "$result" == "newer" ]; then + ocf_log info "peer starting" + if [ "$revision_compare_result" == "newer" ]; then set_force_new_cluster - elif [ "$result" == "older" ]; then + elif [ "$revision_compare_result" == "older" ]; then ocf_log info "$NODENAME shall join as learner" JOIN_AS_LEARNER=true else if [ "$(attribute_node_cluster_id get)" == "$(attribute_node_cluster_id_peer)" ]; then - ocf_log info "same revision and cluster id: start normal" + ocf_log info "same cluster_id and revision: start normal" else ocf_exit_reason "same revision but different cluster id" return "$OCF_ERR_GENERIC" fi fi ;; *) ocf_log error "Unexpected start resource count: $start_resources_count" + podman_notify + return "$OCF_ERR_GENERIC" ;; esac ;; *) ocf_log error "Unexpected active resource count: $active_resources_count" + podman_notify + return "$OCF_ERR_GENERIC" ;; esac fi fi podman_create_mounts local run_opts="-d --name=${CONTAINER}" # check to see if the container has already started podman_simple_status if [ $? -eq $OCF_SUCCESS ]; then return "$OCF_SUCCESS" fi if ocf_is_true "$JOIN_AS_LEARNER"; then local wait_timeout_sec=$((10*60)) local poll_interval_sec=5 local retries=$(( wait_timeout_sec / poll_interval_sec )) ocf_log info "ensure the leader node added $NODENAME as learner member before continuing (timeout: $wait_timeout_sec seconds)" for try in $(seq $retries); do learner_node=$(attribute_learner_node get) if [ "$NODENAME" != "$learner_node" ]; then ocf_log info "$learner_node is not in the member list yet. Retry in $poll_interval_sec seconds." sleep $poll_interval_sec continue fi ocf_log info "learner node $learner_node in the member list" break done if [ "$NODENAME" != "$(attribute_learner_node get)" ]; then ocf_log error "wait for $NODENAME to be in the member list timed out" return "$OCF_ERR_GENERIC" fi archive_data_folder fi prepare_env # add etcd-specific opts run_opts="$run_opts \ --network=host \ -v /etc/kubernetes/static-pod-resources/etcd-certs:/etc/kubernetes/static-pod-certs \ -v /var/lib/etcd:/var/lib/etcd \ --env ALL_ETCD_ENDPOINTS=$ALL_ETCD_ENDPOINTS \ --env ETCD_CIPHER_SUITES=$ETCD_CIPHER_SUITES \ --env ETCD_DATA_DIR=$ETCD_DATA_DIR \ --env ETCD_ELECTION_TIMEOUT=$ETCD_ELECTION_TIMEOUT \ --env ETCD_ENABLE_PPROF=$ETCD_ENABLE_PPROF \ --env ETCD_EXPERIMENTAL_MAX_LEARNERS=$ETCD_EXPERIMENTAL_MAX_LEARNERS \ --env ETCD_EXPERIMENTAL_WARNING_APPLY_DURATION=$ETCD_EXPERIMENTAL_WARNING_APPLY_DURATION \ --env ETCD_EXPERIMENTAL_WATCH_PROGRESS_NOTIFY_INTERVAL=$ETCD_EXPERIMENTAL_WATCH_PROGRESS_NOTIFY_INTERVAL \ --env ETCD_HEARTBEAT_INTERVAL=$ETCD_HEARTBEAT_INTERVAL \ --env ETCD_INITIAL_CLUSTER=$ETCD_INITIAL_CLUSTER \ --env ETCD_INITIAL_CLUSTER_STATE=$ETCD_INITIAL_CLUSTER_STATE \ --env ETCD_NAME=$NODENAME \ --env ETCD_QUOTA_BACKEND_BYTES=$ETCD_QUOTA_BACKEND_BYTES \ --env ETCD_SOCKET_REUSE_ADDRESS=$ETCD_SOCKET_REUSE_ADDRESS \ --env ETCDCTL_API=$ETCDCTL_API \ --env ETCDCTL_CACERT=$SERVER_CACERT \ --env ETCDCTL_CERT=$ETCD_PEER_CERT \ --env ETCDCTL_KEY=$ETCD_PEER_KEY \ --authfile=$OCF_RESKEY_authfile \ --security-opt label=disable" if [ -n "$OCF_RESKEY_run_opts" ]; then run_opts="$run_opts $OCF_RESKEY_run_opts" fi OCF_RESKEY_run_cmd="$OCF_RESKEY_run_cmd --logger=zap \ --log-level=info \ --experimental-initial-corrupt-check=true \ --snapshot-count=10000 \ --initial-advertise-peer-urls=https://${NODEIP}:2380 \ --cert-file=/etc/kubernetes/static-pod-certs/secrets/etcd-all-certs/etcd-serving-${NODENAME}.crt \ --key-file=/etc/kubernetes/static-pod-certs/secrets/etcd-all-certs/etcd-serving-${NODENAME}.key \ --trusted-ca-file=$SERVER_CACERT \ --client-cert-auth=true \ --peer-cert-file=$ETCD_PEER_CERT \ --peer-key-file=$ETCD_PEER_KEY \ --peer-trusted-ca-file=$SERVER_CACERT \ --peer-client-cert-auth=true \ --advertise-client-urls=https://${NODEIP}:2379 \ --listen-client-urls=https://${LISTEN_CLIENT_URLS}:2379,unixs://${NODEIP}:0 \ --listen-peer-urls=https://${LISTEN_PEER_URLS}:2380 \ --metrics=extensive \ --listen-metrics-urls=https://${LISTEN_METRICS_URLS}:9978" if [ -n "$OCF_RESKEY_run_cmd_opts" ]; then OCF_RESKEY_run_cmd="$OCF_RESKEY_run_cmd $OCF_RESKEY_run_cmd_opts" fi if is_force_new_cluster; then OCF_RESKEY_run_cmd="$OCF_RESKEY_run_cmd --force-new-cluster" fi if [ "$OCF_RESKEY_image" == "$OCF_RESKEY_image_default" ]; then # no container image provided via input parameters. Read it from the pod manifest. OCF_RESKEY_image=$(jq -r '.spec.containers[] | select( .name=="etcd").image' "$OCF_RESKEY_pod_manifest") ocf_log info "using container image ($OCF_RESKEY_image) from Pod manifest ($OCF_RESKEY_pod_manifest)" else # use the container image provided as input parameter ocf_log info "using container image ($OCF_RESKEY_image) via input parameters" fi if [ $REQUIRE_IMAGE_PULL -eq 1 ]; then ocf_log notice "Beginning pull of image, ${OCF_RESKEY_image}" podman pull --authfile=$OCF_RESKEY_authfile "${OCF_RESKEY_image}" if [ $? -ne 0 ]; then ocf_exit_reason "failed to pull image ${OCF_RESKEY_image}" return $OCF_ERR_GENERIC fi else ocf_log notice "Pull image not required, ${OCF_RESKEY_image}" fi if ocf_is_true "$OCF_RESKEY_reuse" && container_exists; then ocf_log info "starting existing container $CONTAINER." ocf_run podman start $CONTAINER else # make sure any previous container matching our container name is cleaned up first. # we already know at this point it wouldn't be running remove_container run_new_container "$run_opts" $OCF_RESKEY_image "$OCF_RESKEY_run_cmd" if [ $? -eq 125 ]; then return $OCF_ERR_GENERIC fi fi rc=$? # if the container was stopped or didn't exist before, systemd # removed the libpod* scopes. So always try to recreate the drop-ins if [ $rc -eq 0 ] && ocf_is_true "$OCF_RESKEY_drop_in_dependency"; then cid=$(podman_container_id) create_transient_drop_in_dependency "$cid" rc=$? fi if [ $rc -ne 0 ]; then ocf_exit_reason "podman failed to launch container (rc: $rc)" return $OCF_ERR_GENERIC fi # wait for monitor to pass before declaring that the container is started while true; do podman_simple_status if [ $? -ne $OCF_SUCCESS ]; then ocf_exit_reason "Newly created podman container exited after start" ocf_run podman logs --tail 20 ${CONTAINER} return $OCF_ERR_GENERIC fi monitor_cmd_exec if [ $? -eq $OCF_SUCCESS ]; then ocf_log notice "Container $CONTAINER started successfully" if is_force_new_cluster; then clear_force_new_cluster local peer_node_name local peer_node_ip peer_node_name="$(get_peer_node_name)" peer_node_ip="$(attribute_node_ip_peer)" if [ -n "$peer_node_name" ] && [ -n "$peer_node_ip" ]; then add_member_as_learner $peer_node_name $peer_node_ip else ocf_log error "could not add peer as learner (peer node name: ${peer_node_name:-unknown}, peer ip: ${peer_node_ip:-unknown})" fi fi return $OCF_SUCCESS fi ocf_exit_reason "waiting on monitor_cmd to pass after start" sleep 1 done } podman_stop() { local timeout=60 local rc podman_simple_status if [ $? -eq $OCF_NOT_RUNNING ]; then remove_container ocf_log info "could not leave members list: etcd container not running" return $OCF_SUCCESS fi attribute_node_revision update attribute_node_cluster_id update if ! member_id=$(attribute_node_member_id get); then ocf_log err "error leaving members list: could not get member-id" else # TODO: is it worth/possible to check the current status instead than relying on cached attributes? if is_standalone; then ocf_log info "last member. Not leaving the member list" else ocf_log info "leaving members list as member with ID $member_id" endpoint="https://$(attribute_node_ip get):2379" ocf_run podman exec $CONTAINER etcdctl member remove $member_id --endpoints=$endpoint rc=$? if [ $? -ne 0 ]; then ocf_log err "error leaving members list: error code $rc" fi fi fi attribute_node_member_id clear if [ -n "$OCF_RESKEY_CRM_meta_timeout" ]; then timeout=$((($OCF_RESKEY_CRM_meta_timeout/1000) -10 )) if [ $timeout -lt 10 ]; then timeout=10 fi fi if ocf_is_true "$OCF_RESKEY_force_kill"; then ocf_run podman kill $CONTAINER rc=$? else ocf_log debug "waiting $timeout second[s] before killing container" ocf_run podman stop -t=$timeout $CONTAINER rc=$? # on stop, systemd will automatically delete any transient # drop-in conf that has been created earlier fi if [ $rc -ne 0 ]; then # If the stop failed, it could be because the controlling conmon # process died unexpectedly. If so, a generic error code is returned # but the associated container exit code is -1. If that's the case, # assume there's no failure and continue with the rm as usual. if [ $rc -eq 125 ] && \ podman inspect --format '{{.State.Status}}:{{.State.ExitCode}}' $CONTAINER | grep -Eq '^(exited|stopped):-1$'; then ocf_log err "Container ${CONTAINER} had an unexpected stop outcome. Trying to remove it anyway." else ocf_exit_reason "Failed to stop container, ${CONTAINER}, based on image, ${OCF_RESKEY_image}." return $OCF_ERR_GENERIC fi fi remove_container if [ $? -ne 0 ]; then ocf_exit_reason "Failed to remove stopped container, ${CONTAINER}, based on image, ${OCF_RESKEY_image}." return $OCF_ERR_GENERIC fi return $OCF_SUCCESS } image_exists() { if [ "$OCF_RESKEY_image" == "$OCF_RESKEY_image_default" ]; then # the actual container image was not defined yet. Nor by # the user via OCF_RESKEY, nor by reading the Pod manifest return 0 fi podman image exists "${OCF_RESKEY_image}" if [ $? -eq 0 ]; then # image found return 0 fi if ocf_is_true "$OCF_RESKEY_allow_pull"; then REQUIRE_IMAGE_PULL=1 ocf_log notice "Image (${OCF_RESKEY_image}) does not exist locally but will be pulled during start" return 0 fi # image not found. return 1 } podman_validate() { check_binary curl check_binary crictl check_binary oc check_binary podman check_binary jq if [ -z "$OCF_RESKEY_node_ip_map" ]; then ocf_exit_reason "'node_ip_map' option is required" exit $OCF_ERR_CONFIGURED fi if [ -z "$OCF_RESKEY_pod_manifest" ]; then ocf_exit_reason "'pod_manifest' option is required" exit $OCF_ERR_CONFIGURED fi if [ -z "$OCF_RESKEY_image" ]; then ocf_exit_reason "'image' option is required" exit $OCF_ERR_CONFIGURED fi image_exists if [ $? -ne 0 ]; then ocf_exit_reason "base image, ${OCF_RESKEY_image}, could not be found." exit $OCF_ERR_CONFIGURED fi return $OCF_SUCCESS } podman_notify() { ocf_log info "notify: type=${OCF_RESKEY_CRM_meta_notify_type}, operation=${OCF_RESKEY_CRM_meta_notify_operation}, nodes { active=[${OCF_RESKEY_CRM_meta_notify_active_uname}], start=[${OCF_RESKEY_CRM_meta_notify_start_uname}], stop=[${OCF_RESKEY_CRM_meta_notify_stop_uname}] }, resources { active=[${OCF_RESKEY_CRM_meta_notify_active_resource}], start =[${OCF_RESKEY_CRM_meta_notify_start_resource}], stop=[${OCF_RESKEY_CRM_meta_notify_stop_resource}] }" } # TODO : # When a user starts plural clones in a node in globally-unique, a user cannot appoint plural name parameters. # When a user appoints reuse, the resource agent cannot connect plural clones with a container. if ocf_is_true "$OCF_RESKEY_CRM_meta_globally_unique"; then if [ -n "$OCF_RESKEY_name" ]; then if [ -n "$OCF_RESKEY_CRM_meta_clone_node_max" ] && [ "$OCF_RESKEY_CRM_meta_clone_node_max" -ne 1 ] then ocf_exit_reason "Cannot make plural clones from the same name parameter." exit $OCF_ERR_CONFIGURED fi if [ -n "$OCF_RESKEY_CRM_meta_master_node_max" ] && [ "$OCF_RESKEY_CRM_meta_master_node_max" -ne 1 ] then ocf_exit_reason "Cannot make plural master from the same name parameter." exit $OCF_ERR_CONFIGURED fi fi : ${OCF_RESKEY_name=`echo ${OCF_RESOURCE_INSTANCE} | tr ':' '-'`} else : ${OCF_RESKEY_name=${OCF_RESOURCE_INSTANCE}} fi CONTAINER=$OCF_RESKEY_name # Note: we currently monitor podman containers by with the "podman exec" # command, so make sure that invocation is always valid by enforcing the # exec command to be non-empty : ${OCF_RESKEY_monitor_cmd:=/bin/true} # When OCF_RESKEY_drop_in_dependency is not populated, we # look at another file-based way of enabling the option. # Otherwise, consider it disabled. if [ -z "$OCF_RESKEY_drop_in_dependency" ]; then if [ -f "/etc/sysconfig/podman_drop_in" ] || \ [ -f "/etc/default/podman_drop_in" ]; then OCF_RESKEY_drop_in_dependency=yes fi fi case $__OCF_ACTION in meta-data) meta_data exit $OCF_SUCCESS;; usage|help) podman_usage exit $OCF_SUCCESS ;; esac NODENAME=$(ocf_local_nodename) JOIN_AS_LEARNER=false case $__OCF_ACTION in start) podman_validate || exit $? podman_start;; stop) podman_stop;; monitor) podman_monitor;; notify) podman_notify;; validate-all) podman_validate;; *) podman_usage exit $OCF_ERR_UNIMPLEMENTED ;; esac rc=$? ocf_log debug "${OCF_RESOURCE_INSTANCE} $__OCF_ACTION : $rc" exit $rc