Add PostgreSQL HA OCF agent modular baseline

2026-09-12 01:28:33 -04:00
commit 4a358ae0f6
30 changed files with 7119 additions and 0 deletions
@@ -0,0 +1,202 @@
+# Sourced by managed-container; uses the shared OCF and parameter environment.
+
+node_to_addr()
+{
+    local target="${1:-}"
+
+    [ -n "${target}" ] || return 1
+
+    corosync-cmapctl 2>/dev/null |
+    awk -v target="${target}" '
+        /nodelist\.node\.[0-9]+\.name / {
+            key=$1
+            sub(/\.name$/, "", key)
+            names[key]=$NF
+        }
+
+        /nodelist\.node\.[0-9]+\.ring0_addr / {
+            key=$1
+            sub(/\.ring0_addr$/, "", key)
+            addrs[key]=$NF
+        }
+
+        END {
+            for (key in names) {
+                if (names[key] == target && addrs[key] != "") {
+                    print addrs[key]
+                    exit 0
+                }
+            }
+            exit 1
+        }
+    '
+}
+
+managed_demote()
+{
+    local monitor_rc
+    local target_node target_addr
+    local local_node slot_name target_port
+    local role receiver_status i
+
+    managed_monitor
+    monitor_rc=$?
+
+    case "${monitor_rc}" in
+        "${OCF_RUNNING_PROMOTED}")
+            ;;
+        "${OCF_SUCCESS}")
+            ocf_log info \
+                "PostgreSQL is already unpromoted"
+            return "${OCF_SUCCESS}"
+            ;;
+        "${OCF_NOT_RUNNING}")
+            ocf_exit_reason \
+                "Cannot complete demotion because PostgreSQL is stopped"
+            return "${OCF_ERR_GENERIC}"
+            ;;
+        *)
+            ocf_exit_reason \
+                "Cannot demote unhealthy PostgreSQL instance (monitor rc=${monitor_rc})"
+            return "${monitor_rc}"
+            ;;
+    esac
+
+    target_node="$(demote_target)" || {
+        ocf_exit_reason \
+            "No fresh valid Pacemaker pre-demote target is available"
+        return "${OCF_ERR_GENERIC}"
+    }
+
+    target_addr="$(node_to_addr "${target_node}")" || {
+        ocf_exit_reason \
+            "Unable to resolve Pacemaker promotion target ${target_node} through Corosync"
+        return "${OCF_ERR_GENERIC}"
+    }
+
+    target_port="${REJOIN_SOURCE_PORT:-5432}"
+    local_node="$(hostname)"
+    slot_name="ha_$(printf '%s' "${local_node}" |
+        tr '[:upper:]-' '[:lower:]_')"
+
+    ocf_log info \
+        "Preparing controlled demotion of ${local_node}; successor=${target_node} (${target_addr})"
+
+    if ! candidate_is_standby "${target_addr}"; then
+        ocf_exit_reason \
+            "Promotion target ${target_node} is not a reachable PostgreSQL standby"
+        return "${OCF_ERR_GENERIC}"
+    fi
+
+    if ! candidate_caught_up "${target_node}"; then
+        ocf_exit_reason \
+            "Promotion target ${target_node} is not streaming and fully replayed to the current flushed WAL position"
+        return "${OCF_ERR_GENERIC}"
+    fi
+
+    if ! ensure_remote_slot "${target_addr}" "${slot_name}"; then
+        ocf_exit_reason \
+            "Unable to ensure physical replication slot ${slot_name} on ${target_node}"
+        return "${OCF_ERR_GENERIC}"
+    fi
+
+    ocf_log info \
+        "Demotion safety checks passed; stopping PostgreSQL primary"
+
+    if ! pg_command '
+        BINDIR="$(pg_config --bindir)" || exit 1
+        exec "$BINDIR/pg_ctl" \
+            -D "$PGDATA" \
+            stop -m fast -w
+    '
+    then
+        ocf_exit_reason \
+            "pg_ctl stop failed during controlled demotion"
+        return "${OCF_ERR_GENERIC}"
+    fi
+
+    if pg_running; then
+        ocf_exit_reason \
+            "PostgreSQL remained running after pg_ctl stop"
+        return "${OCF_ERR_GENERIC}"
+    fi
+
+    if ! service_running || ! container_running; then
+        ocf_exit_reason \
+            "Maintenance service/container stopped unexpectedly during demotion"
+        return "${OCF_ERR_GENERIC}"
+    fi
+
+    ocf_log info \
+        "Configuring ${local_node} as cascading standby of ${target_node}"
+
+    if ! as_service_user \
+        podman exec \
+        --user "${PG_OS_USER}" \
+        --env "RA_PRIMARY_HOST=${target_addr}" \
+        --env "RA_PRIMARY_PORT=${target_port}" \
+        --env "RA_REPLICATION_USER=${REPLICATION_USER}" \
+        --env "RA_REPLICATION_SLOT=${slot_name}" \
+        --env "RA_APPLICATION_NAME=${local_node}" \
+        "${CONTAINER_NAME}" \
+        sh -c '
+            set -e
+
+            sed -i \
+                -e "/^[[:space:]]*primary_conninfo[[:space:]]*=/d" \
+                -e "/^[[:space:]]*primary_slot_name[[:space:]]*=/d" \
+                "$PGDATA/postgresql.auto.conf"
+
+            printf "%s
+" \
+                "primary_conninfo = '\''host=${RA_PRIMARY_HOST} port=${RA_PRIMARY_PORT} user=${RA_REPLICATION_USER} application_name=${RA_APPLICATION_NAME}'\''" \
+                "primary_slot_name = '\''${RA_REPLICATION_SLOT}'\''" \
+                >> "$PGDATA/postgresql.auto.conf"
+
+            touch "$PGDATA/standby.signal"
+
+            chmod 600 \
+                "$PGDATA/postgresql.auto.conf" \
+                "$PGDATA/standby.signal"
+        '
+    then
+        ocf_exit_reason \
+            "Failed to configure PostgreSQL as standby of ${target_node}"
+        return "${OCF_ERR_GENERIC}"
+    fi
+
+    ocf_log info \
+        "Starting PostgreSQL as cascading standby of ${target_node}"
+
+    if ! pg_command '
+        BINDIR="$(pg_config --bindir)" || exit 1
+        exec "$BINDIR/pg_ctl" \
+            -D "$PGDATA" \
+            -l "$PGDATA/postgresql.log" \
+            start -w
+    '
+    then
+        ocf_exit_reason \
+            "PostgreSQL failed to start as cascading standby"
+        return "${OCF_ERR_GENERIC}"
+    fi
+
+    for i in $(seq 1 30); do
+        role="$(pg_role 2>/dev/null || true)"
+        receiver_status="$(pg_wal_receiver_status 2>/dev/null || true)"
+
+        if [ "${role}" = "t" ] &&
+           [ "${receiver_status}" = "streaming" ]
+        then
+            ocf_log info \
+                "Controlled demotion complete: ${local_node} is streaming from ${target_node}"
+            return "${OCF_SUCCESS}"
+        fi
+
+        sleep 1
+    done
+
+    ocf_exit_reason \
+        "PostgreSQL started after demotion but did not become a streaming standby"
+    return "${OCF_ERR_GENERIC}"
+}