fix: 暴露内存回收真实结果

2026-07-21 09:20:02 +02:00
parent 5e62ff4657
commit 01ca1d3dad
4 changed files with 46 additions and 8 deletions
@@ -83,6 +83,10 @@ NC01 的内存压力 cgroup 回收规则：
  - owning YAML 声明目标 cgroup、回收量、`minimumInactiveAnonBytes` 和 swap 保留线；
  - plan 与执行前复核 `inactive_anon` 不低于 YAML 下限；
  - 最坏情况下的换出量不会突破 YAML 声明的 `minimumSwapFreeBytesAfterReclaim`；
+- cgroup `reclaimBytes` 只表示内核请求上界，不是预期收益：
+  - plan 必须显示 `estimateKind=kernel-request-upper-bound`；
+  - job 终态必须分别披露 cgroup 实际下降、`MemAvailable` 变化、`targetMet` 和剩余缺口；
+  - 内核请求成功但目标未达到时，结果必须为 `memory-target-not-met`，不得按请求量宣称已回收；
 - 按 `plan`、`run --confirm`、`status`、`memory` 的顺序复核实际结果。

 默认 `/tmp` GC 只包含 allowlisted 诊断目录和已知低风险路径。非 allowlist 的 stale `/tmp` 一级子项必须显式 `--include-stale-tmp` 才能进入候选；扫描按 `--limit` 有界枚举候选，执行时仍只允许删除 `/tmp` 直接子项，并避开 X11/ICE/font socket、systemd private、tmux、ssh、vscode 等系统/session 前缀。该入口不能递归扩大成通用 `/tmp` 清空器，也不能为了估算全量临时目录而长时间阻塞。
@@ -1526,15 +1526,22 @@ def summarize(candidates, returned, disk=None):
    }

 def summarize_memory_candidates(candidates, returned):
-    total_memory = sum(safe_int(item.get("expectedMemoryRssBytes")) for item in candidates)
-    returned_memory = sum(safe_int(item.get("expectedMemoryRssBytes")) for item in returned)
+    total_memory = sum(safe_int(item.get("expectedMemoryRssBytes")) for item in candidates if item.get("kind") != "cgroup-memory-reclaim")
+    returned_memory = sum(safe_int(item.get("expectedMemoryRssBytes")) for item in returned if item.get("kind") != "cgroup-memory-reclaim")
+    requested_memory = sum(safe_int(item.get("reclaimBytes")) for item in candidates if item.get("kind") == "cgroup-memory-reclaim")
+    returned_requested_memory = sum(safe_int(item.get("reclaimBytes")) for item in returned if item.get("kind") == "cgroup-memory-reclaim")
    by_kind = {}
    for item in candidates:
        kind = item.get("kind") or "unknown"
-        bucket = by_kind.setdefault(kind, {"count": 0, "expectedMemoryRssBytes": 0})
+        bucket = by_kind.setdefault(kind, {"count": 0})
        bucket["count"] += 1
-        bucket["expectedMemoryRssBytes"] += safe_int(item.get("expectedMemoryRssBytes"))
-        bucket["expectedMemoryRss"] = fmt_bytes(bucket["expectedMemoryRssBytes"])
+        if kind == "cgroup-memory-reclaim":
+            bucket["requestedMemoryReclaimBytes"] = safe_int(bucket.get("requestedMemoryReclaimBytes")) + safe_int(item.get("reclaimBytes"))
+            bucket["requestedMemoryReclaim"] = fmt_bytes(bucket["requestedMemoryReclaimBytes"])
+            bucket["estimateKind"] = "kernel-request-upper-bound"
+        else:
+            bucket["expectedMemoryRssBytes"] = safe_int(bucket.get("expectedMemoryRssBytes")) + safe_int(item.get("expectedMemoryRssBytes"))
+            bucket["expectedMemoryRss"] = fmt_bytes(bucket["expectedMemoryRssBytes"])
    return {
        "scope": "memory-pressure-only",
        "candidateCount": len(candidates),
@@ -1543,6 +1550,11 @@ def summarize_memory_candidates(candidates, returned):
        "expectedMemoryRss": fmt_bytes(total_memory),
        "returnedExpectedMemoryRssBytes": returned_memory,
        "returnedExpectedMemoryRss": fmt_bytes(returned_memory),
+        "requestedMemoryReclaimBytes": requested_memory,
+        "requestedMemoryReclaim": fmt_bytes(requested_memory),
+        "returnedRequestedMemoryReclaimBytes": returned_requested_memory,
+        "returnedRequestedMemoryReclaim": fmt_bytes(returned_requested_memory),
+        "cgroupEstimateKind": "kernel-request-upper-bound" if requested_memory > 0 else None,
        "estimatedDiskBytes": 0,
        "outcome": "memory-reclaim-candidates" if candidates else "no-memory-reclaim",
        "byKind": by_kind,
@@ -2612,6 +2624,15 @@ def finish_memory_pressure_job(job_id, candidates, observed_at, worker_pid, work
            - safe_int((memory_before.get("memory") or {}).get("availableBytes"))
        ) if memory_before.get("ok") and memory_after.get("ok") else None,
    })
+    target_available = safe_int((MEMORY_CONFIG.get("cgroupReclaim") or {}).get("targetMemAvailableBytes"))
+    after_available = safe_int((memory_after.get("memory") or {}).get("availableBytes"))
+    target_met = target_available > 0 and after_available > target_available
+    summary.update({
+        "targetMemAvailableBytes": target_available,
+        "targetMet": target_met,
+        "targetGapBytes": max(0, target_available - after_available + 1) if target_available > 0 else None,
+        "outcome": "memory-target-met" if target_met else "memory-target-not-met",
+    })
    payload = {
        "ok": len(failed) == 0,
        "action": "gc remote run",
@@ -455,6 +455,9 @@ def collect_cgroup_reclaim_candidates():
            "memoryCurrentBytes": current,
            "memorySwapCurrentBytes": read_cgroup_integer(path, "memory.swap.current"),
            "memoryStat": memory_stat,
+            "requestedMemoryReclaimBytes": reclaim_bytes,
+            "requestedMemoryReclaim": fmt_bytes(reclaim_bytes),
+            "estimateKind": "kernel-request-upper-bound",
            "expectedMemoryRssBytes": reclaim_bytes,
            "expectedMemoryRss": fmt_bytes(reclaim_bytes),
            "estimatedDiskBytes": 0,
@@ -728,9 +731,16 @@ def execute_cgroup_reclaim_candidate(candidate):
    cgroup_delta = max(0, cgroup_before - cgroup_after) if cgroup_before is not None and cgroup_after is not None else 0
    host_delta = max(0, after_available - before_available) if after_available is not None else 0
    succeeded = write_error is None or cgroup_delta > 0 or host_delta > 0
+    partial = max(cgroup_delta, host_delta) < safe_int(configured.get("reclaimBytes"))
+    if write_error and succeeded:
+        outcome = "partial-reclaim-after-kernel-error"
+    elif succeeded and partial:
+        outcome = "partial-reclaim"
+    else:
+        outcome = "reclaim-requested" if succeeded else "reclaim-request-failed"
    return {
        "status": "succeeded" if succeeded else "failed",
-        "outcome": "partial-reclaim-after-kernel-error" if write_error and succeeded else ("reclaim-requested" if succeeded else "reclaim-request-failed"),
+        "outcome": outcome,
        "reclaimedMemoryBytes": max(cgroup_delta, host_delta),
        "hostMemAvailableDeltaBytes": (after_available - before_available) if after_available is not None else None,
        "cgroupMemoryDeltaBytes": cgroup_delta,
@@ -400,8 +400,6 @@ function compactMemoryPressureCandidate(value: unknown): Record<string, unknown>
    kind: candidate.kind,
    reason: candidate.reason,
    risk: candidate.risk,
-    expectedMemoryRssBytes: candidate.expectedMemoryRssBytes,
-    expectedMemoryRss: candidate.expectedMemoryRss,
    action: { op: action.op, allowlist: action.allowlist },
  };
  if (candidate.kind === "cgroup-memory-reclaim") {
@@ -410,6 +408,9 @@ function compactMemoryPressureCandidate(value: unknown): Record<string, unknown>
      configId: candidate.configId,
      path: candidate.path,
      reclaimBytes: candidate.reclaimBytes,
+      requestedMemoryReclaimBytes: candidate.requestedMemoryReclaimBytes,
+      requestedMemoryReclaim: candidate.requestedMemoryReclaim,
+      estimateKind: candidate.estimateKind,
      swappiness: candidate.swappiness,
      minimumInactiveAnonBytes: candidate.minimumInactiveAnonBytes,
      targetMemAvailableBytes: candidate.targetMemAvailableBytes,
@@ -428,6 +429,8 @@ function compactMemoryPressureCandidate(value: unknown): Record<string, unknown>
      && positiveInteger(identity.startTicks) !== null);
  return {
    ...base,
+    expectedMemoryRssBytes: candidate.expectedMemoryRssBytes,
+    expectedMemoryRss: candidate.expectedMemoryRss,
    rootPid: candidate.rootPid,
    sid: candidate.sid,
    rootStartTicks: candidate.rootStartTicks,