fix(postgresql): address review correctness gaps

Author: weicao

addons/postgresql/config/pg12-config-constraint.cue

@@ -271,7 +271,7 @@
 	extra_float_digits?: int & >=-15 & <=3
 
 	// Forces use of parallel query facilities.
-	force_parallel_mode?: #PgBool
+	force_parallel_mode?: string & =~"(?i)^(off|on|regress)$"
 
 	// Sets the FROM-list size beyond which subqueries are not collapsed.
 	from_collapse_limit?: int & >=1 & <=2147483647

addons/postgresql/config/pg12-config.tpl

@@ -106,7 +106,7 @@ enable_sort = 'True'
 enable_tidscan = 'True'
 escape_string_warning = 'True'
 extra_float_digits = '1'
-force_parallel_mode = '0'
+force_parallel_mode = 'off'
 from_collapse_limit = '8'
 # fsync=off
 # full_page_writes=off

addons/postgresql/config/pg14-config-constraint.cue

@@ -295,7 +295,7 @@
 	extra_float_digits?: int & >=-15 & <=3
 
 	// Forces use of parallel query facilities.
-	force_parallel_mode?: #PgBool
+	force_parallel_mode?: string & =~"(?i)^(off|on|regress)$"
 
 	// Sets the FROM-list size beyond which subqueries are not collapsed.
 	from_collapse_limit?: int & >=1 & <=2147483647

addons/postgresql/config/pg14-config.tpl

@@ -115,7 +115,7 @@ enable_sort = 'True'
 enable_tidscan = 'True'
 escape_string_warning = 'True'
 extra_float_digits = '1'
-force_parallel_mode = '0'
+force_parallel_mode = 'off'
 from_collapse_limit = '8'
 # fsync=off # patroni for Extreme Performance
 # full_page_writes=off # patroni for Extreme Performance

addons/postgresql/config/pg15-config-constraint.cue

@@ -295,7 +295,7 @@
 	extra_float_digits?: int & >=-15 & <=3
 
 	// Forces use of parallel query facilities.
-	force_parallel_mode?: #PgBool
+	force_parallel_mode?: string & =~"(?i)^(off|on|regress)$"
 
 	// Sets the FROM-list size beyond which subqueries are not collapsed.
 	from_collapse_limit?: int & >=1 & <=2147483647

addons/postgresql/config/pg15-config.tpl

@@ -115,7 +115,7 @@ enable_sort = 'True'
 enable_tidscan = 'True'
 escape_string_warning = 'True'
 extra_float_digits = '1'
-force_parallel_mode = '0'
+force_parallel_mode = 'off'
 from_collapse_limit = '8'
 # fsync=off # patroni for Extreme Performance
 # full_page_writes=off # patroni for Extreme Performance

addons/postgresql/dataprotection/pgdumpall-restore.sh

@@ -33,7 +33,15 @@ if ! datasafed pull -d zstd-fastest "${DP_BACKUP_NAME}.sql.zst" - \
   exit 1
 fi
 cat "${errlog}" >&2
-if grep "^ERROR:" "${errlog}" | grep -vE "already exists" | grep -q .; then
+if ! awk '
+  /^ERROR:/ {
+    if ($0 !~ /^ERROR:[[:space:]]+role "(postgres|kbadmin|kbdataprotection|kbprobe|kbmonitoring|kbreplicator)" already exists[[:space:]]*$/ &&
+        $0 !~ /^ERROR:[[:space:]]+database "postgres" already exists[[:space:]]*$/) {
+      invalid = 1
+    }
+  }
+  END { exit invalid }
+' "${errlog}"; then
   echo "ERROR: pgdumpall restore hit non-conflict SQL errors (see above); data may be partially restored" >&2
   exit 1
 fi

addons/postgresql/dataprotection/postgresql-fetch-wal-log.sh

@@ -15,7 +15,7 @@ function fetch-wal-log(){
     exit_fetch_wal=0 && mkdir -p $wal_destination_dir
     for dir_name in $(datasafed list /) ; do
       if [[ $exit_fetch_wal -eq 1 ]]; then
-         exit 0
+         break
       fi
 
       # check if the latest_wal_log after the start_wal_log
@@ -54,4 +54,4 @@ function fetch-wal-log(){
          fi
       done
     done
-}
+}

addons/postgresql/dataprotection/wal-g-archive-restore.sh

@@ -1,5 +1,6 @@
 #!/bin/bash
 set -e
+set -o pipefail
 backup_base_path="$(dirname $DP_BACKUP_BASE_PATH)/wal-g/"
 
 function config_wal_g_for_fetch_wal_log() {
@@ -86,19 +87,42 @@ if [[ "${IS_REPLICA}" == "true" ]]; then
     exit 1
 fi
 
-# Primary restore: Restore from successful backup (.old directory exists)
-if [[ -d "${DATA_DIR}.old" ]]; then
-    echo "Restoring data from ${DATA_DIR}.old..."
-    mkdir -p "${DATA_DIR}"
-    mv -f ${DATA_DIR}.old/* ${DATA_DIR}/
-    # Durability barrier BEFORE removing the signal: the signal is the retry
-    # marker for this hook, so it must be the FINAL commit action. If sync
-    # (or the mv) fails, the signal survives and the bootstrap re-runs the
-    # hook instead of losing the restore marker with a half-durable DATA_DIR.
+HANDOFF_MARKER=".kb_walg_handoff"
+
+function finalize_handoff() {
+    local source_dir="${DATA_DIR}.old"
+    local marker_path="${DATA_DIR}/${HANDOFF_MARKER}"
+
+    if [[ -d "${source_dir}" ]]; then
+        if [[ -f "${marker_path}" ]]; then
+            if [[ "$(cat "${marker_path}")" != "normal" ]] \
+                || [[ -n "$(find "${source_dir}" -mindepth 1 -maxdepth 1 -print -quit)" ]]; then
+                echo "ERROR: WAL-G archive restore handoff state is inconsistent" >&2
+                exit 1
+            fi
+            rmdir "${source_dir}"
+        else
+            if [[ -d "${DATA_DIR}" ]] \
+                && [[ -n "$(find "${DATA_DIR}" -mindepth 1 -maxdepth 1 -print -quit)" ]]; then
+                echo "ERROR: PostgreSQL data directory is not empty before WAL-G archive handoff" >&2
+                exit 1
+            fi
+            printf 'normal\n' > "${source_dir}/${HANDOFF_MARKER}"
+            rm -rf "${DATA_DIR}"
+            mv "${source_dir}" "${DATA_DIR}"
+        fi
+    elif [[ ! -f "${marker_path}" ]] || [[ "$(cat "${marker_path}")" != "normal" ]]; then
+        echo "ERROR: WAL-G archive restore handoff source is missing" >&2
+        exit 1
+    fi
+
     sync
-    rm -f ${RESTORE_SCRIPT_DIR}/kb_restore.signal
-    echo "Data restore completed successfully"
-fi
+    rm -f "${RESTORE_SCRIPT_DIR}/kb_restore.signal"
+}
+
+echo "Restoring data from ${DATA_DIR}.old..."
+finalize_handoff
+echo "Data restore completed successfully"
 EOF
 
 # Replace variables in the generated script

addons/postgresql/dataprotection/wal-g-backup.sh

@@ -9,10 +9,9 @@ export DATASAFED_BACKEND_BASE_PATH=${backup_base_path}
 export WALG_DELTA_MAX_STEPS=0
 # 20Gi for bundle file
 export WALG_TAR_SIZE_THRESHOLD=21474836480
-# The cmpd pins Service serviceName=postgresql with roleSelector=primary; the
-# suffix must be that fixed serviceName, not the user-chosen component name.
-# pg_switch_wal()/archive checks below must run on the primary.
-PSQL="psql -h ${CLUSTER_COMPONENT_NAME}-postgresql -U ${DP_DB_USER} -d postgres"
+# Use the ActionSet-provided target endpoint instead of reconstructing a
+# controller-owned Service name.
+PSQL="psql -h ${DP_DB_HOST} -U ${DP_DB_USER} -p ${DP_DB_PORT} -d postgres"
 
 # if the script exits with a non-zero exit code, touch a file to indicate that the backup failed,
 # the sync progress container will check this file and exit if it exists
@@ -90,11 +89,10 @@ fi
 
 # 1. do full backup
 writeSentinelInBaseBackupPath "${backup_base_path}" "wal-g-backup-repo.path"
-PGHOST=${DP_DB_HOST} PGUSER=${DP_DB_USER} PGPORT=5432 wal-g backup-push ${DATA_DIR} 2>&1 | tee result.txt
+PGHOST=${DP_DB_HOST} PGUSER=${DP_DB_USER} PGPORT=${DP_DB_PORT} wal-g backup-push ${DATA_DIR} 2>&1 | tee result.txt
 
-set +e
 echo "switch wal log"
-PSQL="psql -h ${CLUSTER_COMPONENT_NAME}-postgresql -U ${DP_DB_USER} -d postgres"
+PSQL="psql -h ${DP_DB_HOST} -U ${DP_DB_USER} -p ${DP_DB_PORT} -d postgres"
 ${PSQL} -c "select pg_switch_wal();"
 
 # 2. get backup name of the wal-g
@@ -117,4 +115,4 @@ START_TIME=$(echo $result_json | jq -r ".StartTime")
 TOTAL_SIZE=$(echo $result_json | jq -r ".CompressedSize")
 
 # 5. update backup status
-echo "{\"totalSize\":\"$TOTAL_SIZE\",\"extras\":[{\"wal-g-backup-name\":\"${backupName}\"}],\"timeRange\":{\"start\":\"${START_TIME}\",\"end\":\"${STOP_TIME}\"}}" >"${DP_BACKUP_INFO_FILE}"
+echo "{\"totalSize\":\"$TOTAL_SIZE\",\"extras\":[{\"wal-g-backup-name\":\"${backupName}\"}],\"timeRange\":{\"start\":\"${START_TIME}\",\"end\":\"${STOP_TIME}\"}}" >"${DP_BACKUP_INFO_FILE}"