20260526 增加持续soak监控与本地回放工具

scripts/soak/portable-soak.env.example

@@ -2,6 +2,7 @@
 # 复制为 .env 后可以在远端直接调整；所有路径默认相对 package 根目录。
 
 # 最大运行轮次。重复执行 run_soak.sh 时会从已有最后一轮之后继续编号。
+# 正整数表示固定运行轮次；负数（例如 -1）表示持续运行不自动停止；0 非法。
 MAX_RUNS=3
 
 # 两轮之间等待秒数。做连续无等待验收时设置为 0。
@@ -15,7 +16,7 @@ RIRS=afrinic,apnic,arin,lacnic,ripe
 # 运行根目录。默认使用 package 根目录；如需把产物写到独立数据盘，可改成绝对路径。
 RUN_ROOT="${PACKAGE_ROOT}"
 
-# 保留最近多少轮 run 目录。旧 run 会由 rpki_daemon 自身或后续脚本策略清理。
+# 保留最近多少轮 run 目录。持续运行模式建议设置为 5 或按磁盘容量评估。
 RETAIN_RUNS=10
 
 # 是否输出 compact report JSON。1 表示启用，0 表示关闭。

scripts/soak/run_soak.sh

@@ -77,6 +77,11 @@ validate_non_negative_int() {
   [[ "$value" =~ ^[0-9]+$ ]] || die "$name must be an integer: $value"
 }
 
+validate_max_runs() {
+  [[ "$MAX_RUNS" =~ ^-?[0-9]+$ ]] || die "MAX_RUNS must be an integer: $MAX_RUNS"
+  [[ "$MAX_RUNS" != "0" ]] || die "MAX_RUNS must be non-zero; use a positive value for fixed runs or -1 for continuous mode"
+}
+
 validate_rsync_scope() {
   case "$RSYNC_SCOPE" in
     publication-point|module-root)
@@ -486,20 +491,30 @@ PY
 
 apply_outer_retention() {
   local dirs=()
+  local retain_limit="$RETAIN_RUNS"
+  local keep_run="${1:-}"
   local run_dir
   shopt -s nullglob
   for run_dir in "$RUNS_ROOT"/run_[0-9][0-9][0-9][0-9]; do
     [[ -d "$run_dir" ]] && dirs+=("$run_dir")
   done
   shopt -u nullglob
-  if (( ${#dirs[@]} <= RETAIN_RUNS )); then
+  if (( ${#dirs[@]} <= retain_limit )); then
     return 0
   fi
   mapfile -t dirs < <(printf '%s\n' "${dirs[@]}" | sort)
-  local remove_count=$(( ${#dirs[@]} - RETAIN_RUNS ))
-  local index
-  for (( index = 0; index < remove_count; index++ )); do
-    rm -rf "${dirs[$index]}"
+  local remove_count=$(( ${#dirs[@]} - retain_limit ))
+  local removed_count=0
+  local candidate
+  for candidate in "${dirs[@]}"; do
+    if [[ -n "$keep_run" && "$(basename "$candidate")" == "$keep_run" ]]; then
+      continue
+    fi
+    rm -rf "$candidate"
+    removed_count=$((removed_count + 1))
+    if (( removed_count >= remove_count )); then
+      break
+    fi
   done
 }
 
@@ -519,6 +534,7 @@ run_one_round() {
   local summary_state
 
   mkdir -p "$run_dir" "$daemon_state_root"
+  apply_outer_retention "$run_id"
   started_at="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
   write_run_meta "$run_dir/run-meta.json" "running" "$run_index" "$run_id" "$sync_mode" \
     "$snapshot_reason" "$previous_run_id" "$previous_success_value" "$started_at" "" \
@@ -573,7 +589,7 @@ main() {
   require_command python3
   require_command date
   require_command find
-  validate_positive_int "MAX_RUNS" "$MAX_RUNS"
+  validate_max_runs
   validate_non_negative_int "INTERVAL_SECS" "$INTERVAL_SECS"
   validate_positive_int "RETAIN_RUNS" "$RETAIN_RUNS"
   validate_rsync_scope
@@ -599,12 +615,20 @@ main() {
 
   local max_index
   local next_index
+  local run_forever=0
+  local stop_index=0
   max_index="$(max_existing_run_index)"
   next_index=$((max_index + 1))
-  local stop_index=$((max_index + MAX_RUNS))
+  if (( MAX_RUNS < 0 )); then
+    run_forever=1
+    echo "run_soak mode=continuous max_existing_run_index=$max_index next_run=$(printf 'run_%04d' "$next_index")"
+  else
+    stop_index=$((max_index + MAX_RUNS))
+    echo "run_soak mode=fixed max_existing_run_index=$max_index next_run=$(printf 'run_%04d' "$next_index") stop_run=$(printf 'run_%04d' "$stop_index")"
+  fi
   local any_failed=0
 
-  while (( next_index <= stop_index )); do
+  while (( run_forever == 1 || next_index <= stop_index )); do
     INVALID_DB_PATH=""
     INVALID_STATE_PATH=""
     INVALID_TMP_PATH=""
@@ -649,7 +673,7 @@ main() {
       echo "completed run $(printf 'run_%04d' "$next_index") status=failed" >&2
       any_failed=1
     fi
-    if (( next_index < stop_index && INTERVAL_SECS > 0 )); then
+    if (( (run_forever == 1 || next_index < stop_index) && INTERVAL_SECS > 0 )); then
       sleep "$INTERVAL_SECS"
     fi
     next_index=$((next_index + 1))