20260716 增加verification-only缓存正确性复核

This commit is contained in:
2026-07-16 19:08:16 +08:00
parent 7aac4855bd
commit 59623b449b
14 changed files with 2269 additions and 19 deletions

View File

@@ -0,0 +1,50 @@
# verification-only 使用说明
`verification-only` 用于复核启用验证缓存的最新成功 normal run。它冻结当前仓库视图,关闭 PP、ROA、child certificate cache、prefetch、VCIR 持久化和网络访问,在源 CIR 的同一验证时间重新执行完整验证,并比较 CCR 的 MFT、VRP、VAP、TA、router key 五类 state digest。
## 前提
- 源 run 必须是当前 state DB 对应的最新成功 normal run。
- 源 run 必须包含 `input.cir`、`result.ccr`、`run-summary.json` 和 `validation-contract.json`。
- state root 必须包含 `db/work-db`、`db/repo-bytes.db` 和 `meta/current-state-binding.json`。
- 源 normal run 必须启用至少一种验证缓存;仅启用 transport prefetch 不算验证缓存。
- verification-only 与 normal soak 共用 `state/locks/state-execution.lock`,不能绕过 wrapper 并发执行。
## 执行
在 portable package 根目录执行:
```bash
./scripts/verification/run_verification_only.sh \
--source-run-dir /data/ours-rp/runs/run_0123 \
--source-state-root /data/ours-rp/state \
--out-dir /data/ours-rp/verification-runs/run_0123
```
验证时间自动从源 `input.cir` 读取,不接受命令行覆盖。成功后 scratch work-db 默认删除;排障时可加 `--keep-scratch`。
## 输出
- `input.cir`、`result.ccr`、`report.json`
- `vrps.csv`、`vaps.csv`
- `stage-timing.json`、`validation-events.jsonl`
- `validation-contract.json`、`verification-meta.json`
- `compare/ccr-state-compare.json`、`compare/ccr-state-compare.md`
- `stdout.log`、`stderr.log`、`process-time.txt`、`wrapper-timing.json`
通过条件是 compare JSON 中 `stateDigestMatch=true`。`producedAt` 不参与比较。
## 隔离保证
- 不分配 normal run id,不更新 lifecycle、periodic snapshot 计数和 retention。
- 源 work-db 只用于创建 checkpoint,后续写入只发生在独立 scratch work-db。
- 源 repo-bytes.db 以只读模式打开。
- 输出目录必须不存在,防止覆盖旧验证证据。
## 常见失败
- `current state binding`:所选 run 已不是当前 state 对应的最新成功 run,需选择 binding 指向的 run。
- `differs from current state binding`:源 CIR、CCR 或 contract 已变化,不能继续复核。
- `blob is missing`:repository view 引用的原始 bytes 不完整,验证不会启动。
- `did not enable a validation cache`:源 run 不是缓存正确性对比的有效基准。
- `CCR state digest mismatch`:查看 `compare/` 下 JSON/Markdown,按不匹配状态集合继续排查。

View File

@@ -0,0 +1,102 @@
#!/usr/bin/env bash
set -euo pipefail
usage() {
cat <<'USAGE'
Usage:
run_verification_only.sh \
--source-run-dir <run_dir> \
--source-state-root <state_dir> \
--out-dir <verification_dir> \
[--rpki-bin <path>] [--keep-scratch]
validation time is always read from the source CIR. The wrapper serializes access
with normal soak runs and never writes into the normal run sequence.
USAGE
}
SOURCE_RUN_DIR=""
SOURCE_STATE_ROOT=""
OUT_DIR=""
RPKI_BIN=""
KEEP_SCRATCH=0
while [[ $# -gt 0 ]]; do
case "$1" in
--source-run-dir) SOURCE_RUN_DIR="${2:?missing value}"; shift 2 ;;
--source-state-root) SOURCE_STATE_ROOT="${2:?missing value}"; shift 2 ;;
--out-dir) OUT_DIR="${2:?missing value}"; shift 2 ;;
--rpki-bin) RPKI_BIN="${2:?missing value}"; shift 2 ;;
--keep-scratch) KEEP_SCRATCH=1; shift ;;
-h|--help) usage; exit 0 ;;
*) echo "unknown argument: $1" >&2; usage >&2; exit 2 ;;
esac
done
[[ -n "$SOURCE_RUN_DIR" ]] || { echo "--source-run-dir is required" >&2; exit 2; }
[[ -n "$SOURCE_STATE_ROOT" ]] || { echo "--source-state-root is required" >&2; exit 2; }
[[ -n "$OUT_DIR" ]] || { echo "--out-dir is required" >&2; exit 2; }
if [[ -z "$RPKI_BIN" ]]; then
RPKI_BIN="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)/target/release/rpki"
fi
[[ -x "$RPKI_BIN" ]] || { echo "rpki binary is not executable: $RPKI_BIN" >&2; exit 2; }
command -v flock >/dev/null || { echo "flock is required" >&2; exit 2; }
command -v python3 >/dev/null || { echo "python3 is required" >&2; exit 2; }
now_ms() {
python3 -c 'import time; print(time.time_ns() // 1_000_000)'
}
mkdir -p "$SOURCE_STATE_ROOT/locks" "$(dirname "$OUT_DIR")"
[[ ! -e "$OUT_DIR" ]] || { echo "verification output already exists: $OUT_DIR" >&2; exit 2; }
LOCK_PATH="$SOURCE_STATE_ROOT/locks/state-execution.lock"
STDOUT_TMP="${OUT_DIR}.stdout.tmp"
STDERR_TMP="${OUT_DIR}.stderr.tmp"
PROCESS_TIME_TMP="${OUT_DIR}.process-time.tmp"
WAIT_STARTED_MS="$(now_ms)"
exec {LOCK_FD}>"$LOCK_PATH"
flock "$LOCK_FD"
LOCK_ACQUIRED_MS="$(now_ms)"
ARGS=(
--verification-only
--verification-source-run "$SOURCE_RUN_DIR"
--verification-source-state "$SOURCE_STATE_ROOT"
--verification-out "$OUT_DIR"
)
if [[ "$KEEP_SCRATCH" -eq 1 ]]; then
ARGS+=(--keep-verification-scratch)
fi
set +e
if [[ -x /usr/bin/time ]]; then
RPKI_STATE_EXECUTION_LOCK_HELD=1 /usr/bin/time -v -o "$PROCESS_TIME_TMP" -- \
"$RPKI_BIN" "${ARGS[@]}" >"$STDOUT_TMP" 2>"$STDERR_TMP"
else
RPKI_STATE_EXECUTION_LOCK_HELD=1 "$RPKI_BIN" "${ARGS[@]}" >"$STDOUT_TMP" 2>"$STDERR_TMP"
fi
EXIT_CODE=$?
set -e
LOCK_RELEASED_MS="$(now_ms)"
flock -u "$LOCK_FD"
exec {LOCK_FD}>&-
mkdir -p "$OUT_DIR"
mv "$STDOUT_TMP" "$OUT_DIR/stdout.log"
mv "$STDERR_TMP" "$OUT_DIR/stderr.log"
if [[ -f "$PROCESS_TIME_TMP" ]]; then
mv "$PROCESS_TIME_TMP" "$OUT_DIR/process-time.txt"
fi
python3 - "$OUT_DIR/wrapper-timing.json" "$WAIT_STARTED_MS" "$LOCK_ACQUIRED_MS" "$LOCK_RELEASED_MS" "$EXIT_CODE" <<'PY'
import json
import sys
from pathlib import Path
path = Path(sys.argv[1])
wait_started_ms, lock_acquired_ms, lock_released_ms, exit_code = map(int, sys.argv[2:])
path.write_text(json.dumps({
"lockWaitMs": lock_acquired_ms - wait_started_ms,
"lockHeldMs": lock_released_ms - lock_acquired_ms,
"exitCode": exit_code,
}, indent=2, sort_keys=True) + "\n", encoding="utf-8")
PY
exit "$EXIT_CODE"