fix: harden SQLite lock handling and WAL retention by Atlas-Key · Pull Request #335 · edheltzel/Recall

Atlas-Key · 2026-10-17T11:47:02Z

What Changed

Set a 5-second busy timeout before WAL setup across CLI/MCP and hook connections, and configure writers to retain at most 64 MiB of WAL space after log resets.
Remove redundant published-message view recreation from schema initialization.
Add pragma and concurrent recovery regression coverage, extend the locked-extraction test timeout, and document WAL contention and retention behavior.

Risk Assessment

⚠️ Medium: The runtime change is bounded with no substantiated correctness regression, but duplicated policy and invalid test evidence warrant follow-up.

Testing

Installed locked dependencies, built CLI/MCP runtime, ran focused SQLite checks, and drove eight isolated product scenarios successfully, including 20 crash-recovery rounds. Saved transcripts and database-state evidence, then cleaned disposable data and build output. Extraction used a local protocol fixture; the WAL comparison used an unlimited-policy control on the same product.

Live validation: ✅ go - 8 of 8 scenarios driven live against the product

Scenario	Result	Live	Evidence
Initialize Recall, add a decision, and find it after reopening	✅ pass	live	Product transcript; Persisted state
Write through MCP while a reader pins its snapshot, then reclaim oversized WAL after reader release	✅ pass	live	MCP WAL comparison: target shrank to 67,108,864 bytes; unlimited control retained 73,937,552 bytes; all 43 records persisted
Add memory while another process briefly locks the database	✅ pass	live	Product transcript: CLI addition completed after the peer released its startup lock
Retry PreCompact after a database timeout without losing or duplicating messages	✅ pass	live	Product transcript; Persisted state: timeout preserved the watermark, retry saved one message, replay added none
Recover a killed WAL writer and retain committed data plus every concurrent CLI addition	✅ pass	live	Product transcript; bun test tests/db/recovery-concurrency.test.ts: 20 rounds of six concurrent CLI additions, with integrity and FTS checks
Run lifecycle hooks and replay extraction with one extracted decision, preserved transcript, recorded progress, and no remaining locks	✅ pass	live	Product transcript; Persisted state; Runtime SQLite settings
Start a session and recall durable memory while sweeping expired breadcrumbs	✅ pass	live	Product transcript; Persisted state: recalled decision displayed and expired breadcrumb removed
Consolidate old decisions while preserving their source lineage	✅ pass	live	Product transcript; Persisted state: one derived summary references three demoted source decisions

Evidence: Product transcript

Live Recall SQLite validation. Runtime data is disposable and isolated under the worktree.

SCENARIO: CLI initializes memory, persists additions, and finds them after reopening
{
  "HOME": "~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/normal/home",
  "RECALL_HOME": "~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/normal/home/.agents/Recall",
  "RECALL_DB_PATH": "~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/normal/recall.db"
}
$ bun dist/index.js init
{
  "exit": 0,
  "stdout": "✓ Created database at ~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/normal/recall.db\n",
  "stderr": ""
}
$ bun dist/index.js add decision SQLITE_GATE durable memory survives reopen --project sqlite-gate
{
  "exit": 0,
  "stdout": "✓ Added decision #1 [sqlite-gate] (medium)\n",
  "stderr": ""
}
$ bun dist/index.js search SQLITE_GATE --project sqlite-gate
{
  "exit": 0,
  "stdout": "Found 1 result(s):\n\n[decisions#1] [sqlite-gate] 2026-10-17 11:42:09\n  SQLITE_GATE durable memory survives reopen\n\n",
  "stderr": ""
}
{
  "persisted": [
    {
      "decision": "SQLITE_GATE durable memory survives reopen",
      "provenance": "user_authored"
    }
  ]
}

SCENARIO: CLI and PreCompact wait for database startup locks; timed-out PreCompact retries without losing messages
$ bun dist/index.js init
{
  "exit": 0,
  "stdout": "✓ Created database at ~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/startup-locks/recall.db\n",
  "stderr": ""
}
$ bun dist/index.js add breadcrumb STARTUP_LOCK memory saved after peer releases
{
  "peer": "LOCKED 1792237329648\nRELEASING 1792237330549"
}
{
  "exit": 0,
  "stdout": "✓ Added breadcrumb #1 [project]\n",
  "stderr": ""
}
$ bun hooks/RecallPreCompact.ts
{
  "exit": 0,
  "stdout": "",
  "stderr": ""
}
{
  "peer": "LOCKED 1792237330698\nRELEASING 1792237345702"
}
$ bun hooks/RecallPreCompact.ts
{
  "exit": 0,
  "stdout": "",
  "stderr": ""
}
$ bun hooks/RecallPreCompact.ts
{
  "exit": 0,
  "stdout": "",
  "stderr": ""
}
{
  "hookLog": "[2026-10-17T11:42:17.527Z] FLUSH_ERROR: ~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/startup-locks/locked-session.jsonl: database is locked\n[2026-10-17T11:42:25.754Z] PRECOMPACT_FIRED: trigger=manual conv=~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/startup-locks/locked-session.jsonl\n[2026-10-17T11:42:25.765Z] FLUSHED: ~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/startup-locks/locked-session.jsonl +1 messages (200 bytes, watermark 0 → 200)\n[2026-10-17T11:42:25.792Z] PRECOMPACT_FIRED: trigger=manual conv=~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/startup-locks/locked-session.jsonl\n[2026-10-17T11:42:25.793Z] NO_NEW_DATA: ~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/startup-locks/locked-session.jsonl (size=200, watermark=200)\n",
  "persisted": [
    {
      "session_id": "locked-session",
      "content": "LOCK_TIMEOUT this transcript must survive the failed compact flush and retry.",
      "provenance": "verbatim"
    }
  ]
}

SCENARIO: Concurrent CLI writers recover a crashed WAL without losing committed data or publishing uncommitted data
$ bun dist/index.js init
{
  "exit": 0,
  "stdout": "✓ Created database at ~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/crash-recovery-1792237345799/recall.db\n",
  "stderr": ""
}
$ bun dist/index.js add breadcrumb RECOVERY_WRITER_0
$ bun dist/index.js add breadcrumb RECOVERY_WRITER_1
$ bun dist/index.js add breadcrumb RECOVERY_WRITER_2
$ bun dist/index.js add breadcrumb RECOVERY_WRITER_3
$ bun dist/index.js add breadcrumb RECOVERY_WRITER_4
$ bun dist/index.js add breadcrumb RECOVERY_WRITER_5
{
  "exit": 0,
  "stdout": "✓ Added breadcrumb #2 [project]\n",
  "stderr": ""
}
{
  "exit": 0,
  "stdout": "✓ Added breadcrumb #3 [project]\n",
  "stderr": ""
}
{
  "exit": 0,
  "stdout": "✓ Added breadcrumb #4 [project]\n",
  "stderr": ""
}
{
  "exit": 0,
  "stdout": "✓ Added breadcrumb #5 [project]\n",
  "stderr": ""
}
{
  "exit": 0,
  "stdout": "✓ Added breadcrumb #6 [project]\n",
  "stderr": ""
}
{
  "exit": 0,
  "stdout": "✓ Added breadcrumb #7 [project]\n",
  "stderr": ""
}
{
  "persisted": [
    {
      "content": "CRASH_COMMITTED memory"
    },
    {
      "content": "RECOVERY_WRITER_0"
    },
    {
      "content": "RECOVERY_WRITER_1"
    },
    {
      "content": "RECOVERY_WRITER_2"
    },
    {
      "content": "RECOVERY_WRITER_3"
    },
    {
      "content": "RECOVERY_WRITER_4"
    },
    {
      "content": "RECOVERY_WRITER_5"
    }
  ],
  "integrity": [
    {
      "integrity_check": "ok"
    }
  ]
}
{
  "results": [
    {
      "name": "CLI initializes memory, persists additions, and finds them after reopening",
      "result": "pass",
      "live": true,
      "detail": [
        {
          "decision": "SQLITE_GATE durable memory survives reopen",
          "provenance": "user_authored"
        }
      ]
    },
    {
      "name": "CLI and PreCompact wait for database startup locks; timed-out PreCompact retries without losing messages",
      "result": "pass",
      "live": true,
      "detail": {
        "startupWriteFinished": 1792237330684,
        "peerReleased": 1792237330549,
        "persisted": [
          {
            "session_id": "locked-session",
            "content": "LOCK_TIMEOUT this transcript must survive the failed compact flush and retry.",
            "provenance": "verbatim"
          }
        ],
        "hookLog": "[2026-10-17T11:42:17.527Z] FLUSH_ERROR: ~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/startup-locks/locked-session.jsonl: database is locked\n[2026-10-17T11:42:25.754Z] PRECOMPACT_FIRED: trigger=manual conv=~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/startup-locks/locked-session.jsonl\n[2026-10-17T11:42:25.765Z] FLUSHED: ~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/startup-locks/locked-session.jsonl +1 messages (200 bytes, watermark 0 → 200)\n[2026-10-17T11:42:25.792Z] PRECOMPACT_FIRED: trigger=manual conv=~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/startup-locks/locked-session.jsonl\n[2026-10-17T11:42:25.793Z] NO_NEW_DATA: ~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/startup-locks/locked-session.jsonl (size=200, watermark=200)\n"
      }
    },
    {
      "name": "Concurrent CLI writers recover a crashed WAL without losing committed data or publishing uncommitted data",
      "result": "pass",
      "live": true,
      "detail": {
        "persisted": [
          {
            "content": "CRASH_COMMITTED memory"
          },
          {
            "content": "RECOVERY_WRITER_0"
          },
          {
            "content": "RECOVERY_WRITER_1"
          },
          {
            "content": "RECOVERY_WRITER_2"
          },
          {
            "content": "RECOVERY_WRITER_3"
          },
          {
            "content": "RECOVERY_WRITER_4"
          },
          {
            "content": "RECOVERY_WRITER_5"
          }
        ],
        "integrity": [
          {
            "integrity_check": "ok"
          }
        ]
      }
    }
  ]
}

SCENARIO: Stop, PreCompact, in-session progress, SessionStart, and consolidation preserve memory with hardened SQLite settings
$ bun dist/index.js init
{
  "exit": 0,
  "stdout": "✓ Created database at ~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/lifecycle/recall.db\n",
  "stderr": ""
}
$ bun hooks/RecallExtract.ts
{
  "exit": 0,
  "stdout": "",
  "stderr": ""
}
$ bun hooks/RecallExtract.ts --reextract .sqlite-gate-runtime/lifecycle/home/.c

... [2042 bytes truncated] ...

\n",
  "stderr": ""
}
$ bun dist/index.js add decision CONSOLIDATE_GATE old low-importance decision 1 --project project
{
  "exit": 0,
  "stdout": "✓ Added decision #3 [project] (medium)\n",
  "stderr": ""
}
$ bun dist/index.js add decision CONSOLIDATE_GATE old low-importance decision 2 --project project
{
  "exit": 0,
  "stdout": "✓ Added decision #4 [project] (medium)\n",
  "stderr": ""
}
$ bun dist/index.js consolidate --execute --table decisions --min-cluster-size 2
{
  "exit": 0,
  "stdout": "[EXECUTE — consolidating records]\n\nRecommended: run 'recall export --backup' before consolidating.\n\nAge cutoff: 180d  ·  importance threshold: <5  ·  window: 30d  ·  min cluster: 2\n\ndecisions [project] window 2019-12-10: 3 record(s) → demote importance to 1 (rep #2; ids: 2, 3, 4)\n\nTotal: 1 cluster(s), 3 record(s) to consolidate\n\nConsolidated: wrote 1 derived summary(ies), demoted 3 source record(s).\n",
  "stderr": "[FabricExtract] Trying ollama extraction...\n[FabricExtract] Ollama extraction successful\nRECALL_CONSOLIDATE_PROGRESS {\"written\":1,\"demoted\":3}\n"
}
{
  "externalModelBoundary": "disposable localhost Ollama protocol fixture; no model-quality claim",
  "requests": 3,
  "tracker": [
    {
      "conversation_path": "~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/lifecycle/home/.claude/projects/-Users-ed--no-mistakes-worktrees-fd32172f1538-01M52TP6ZQ2ARNP0T3RN3BBG9G--sqlite-gate-runtime-lifecycle-project/lifecycle-session.jsonl",
      "extracted_at": "2026-10-17 11:42:26",
      "error": null
    },
    {
      "conversation_path": "~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/lifecycle/old-session.jsonl",
      "extracted_at": "2026-01-01T00:00:00Z",
      "error": null
    }
  ],
  "progress": [
    {
      "session_id": "lifecycle-session",
      "turns_seen": 1
    }
  ],
  "derived": [
    {
      "title": "Consolidated decisions — project (2019-12-10)",
      "provenance": "derived",
      "source_ids": "[{\"table\":\"decisions\",\"id\":2},{\"table\":\"decisions\",\"id\":3},{\"table\":\"decisions\",\"id\":4}]"
    }
  ],
  "hookLog": "[2026-10-17T11:42:26.228Z] TRACKER_MIGRATE: imported 1 rows from JSON tracker into SQLite\n[2026-10-17T11:42:26.231Z] SPAWNED: Self-extract PID 78227 for ~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/lifecycle/home/.claude/projects/-Users-ed--no-mistakes-worktrees-fd32172f1538-01M52TP6ZQ2ARNP0T3RN3BBG9G--sqlite-gate-runtime-lifecycle-project/lifecycle-session.jsonl\n[2026-10-17T11:42:26.244Z] BACKGROUND: Starting extraction for ~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/lifecycle/home/.claude/projects/-Users-ed--no-mistakes-worktrees-fd32172f1538-01M52TP6ZQ2ARNP0T3RN3BBG9G--sqlite-gate-runtime-lifecycle-project/lifecycle-session.jsonl\n[2026-10-17T11:42:26.282Z] QUALITY GATE PASSED: extraction contains required sections\n[2026-10-17T11:42:26.282Z] SQLITE_WRITE: sessions=1 decisions=1 learnings=1 breadcrumbs=3 errors=1 loa=1\n[2026-10-17T11:42:26.282Z] SCRUB: redacted 0 secret kind(s) before SQLite write\n[2026-10-17T11:42:26.285Z] SUCCESS: All memory files updated for session=project\n[2026-10-17T11:42:26.285Z] BACKGROUND: Extraction complete\n[2026-10-17T11:42:26.307Z] REEXTRACT: Forcing re-extraction of ~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/lifecycle/home/.claude/projects/-Users-ed--no-mistakes-worktrees-fd32172f1538-01M52TP6ZQ2ARNP0T3RN3BBG9G--sqlite-gate-runtime-lifecycle-project/lifecycle-session.jsonl\n[2026-10-17T11:42:26.343Z] QUALITY GATE PASSED: extraction contains required sections\n[2026-10-17T11:42:26.343Z] SQLITE_WRITE: sessions=1 decisions=0 learnings=0 breadcrumbs=0 errors=1 loa=1\n[2026-10-17T11:42:26.343Z] SCRUB: redacted 0 secret kind(s) before SQLite write\n[2026-10-17T11:42:26.346Z] SUCCESS: All memory files updated for session=project\n[2026-10-17T11:42:26.346Z] REEXTRACT: Complete\n"
}
{
  "results": [
    {
      "name": "CLI initializes memory, persists additions, and finds them after reopening",
      "result": "pass",
      "live": true,
      "detail": [
        {
          "decision": "SQLITE_GATE durable memory survives reopen",
          "provenance": "user_authored"
        }
      ]
    },
    {
      "name": "CLI and PreCompact wait for database startup locks; timed-out PreCompact retries without losing messages",
      "result": "pass",
      "live": true,
      "detail": {
        "startupWriteFinished": 1792237330684,
        "peerReleased": 1792237330549,
        "persisted": [
          {
            "session_id": "locked-session",
            "content": "LOCK_TIMEOUT this transcript must survive the failed compact flush and retry.",
            "provenance": "verbatim"
          }
        ],
        "hookLog": "[2026-10-17T11:42:17.527Z] FLUSH_ERROR: ~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/startup-locks/locked-session.jsonl: database is locked\n[2026-10-17T11:42:25.754Z] PRECOMPACT_FIRED: trigger=manual conv=~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/startup-locks/locked-session.jsonl\n[2026-10-17T11:42:25.765Z] FLUSHED: ~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/startup-locks/locked-session.jsonl +1 messages (200 bytes, watermark 0 → 200)\n[2026-10-17T11:42:25.792Z] PRECOMPACT_FIRED: trigger=manual conv=~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/startup-locks/locked-session.jsonl\n[2026-10-17T11:42:25.793Z] NO_NEW_DATA: ~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/startup-locks/locked-session.jsonl (size=200, watermark=200)\n"
      }
    },
    {
      "name": "Concurrent CLI writers recover a crashed WAL without losing committed data or publishing uncommitted data",
      "result": "pass",
      "live": true,
      "detail": {
        "persisted": [
          {
            "content": "CRASH_COMMITTED memory"
          },
          {
            "content": "RECOVERY_WRITER_0"
          },
          {
            "content": "RECOVERY_WRITER_1"
          },
          {
            "content": "RECOVERY_WRITER_2"
          },
          {
            "content": "RECOVERY_WRITER_3"
          },
          {
            "content": "RECOVERY_WRITER_4"
          },
          {
            "content": "RECOVERY_WRITER_5"
          }
        ],
        "integrity": [
          {
            "integrity_check": "ok"
          }
        ]
      }
    },
    {
      "name": "Stop, PreCompact, in-session progress, SessionStart, and consolidation preserve memory with hardened SQLite settings",
      "result": "pass",
      "live": true,
      "detail": {
        "modelBoundary": "local protocol fixture",
        "requests": 3,
        "tracker": [
          {
            "conversation_path": "~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/lifecycle/home/.claude/projects/-Users-ed--no-mistakes-worktrees-fd32172f1538-01M52TP6ZQ2ARNP0T3RN3BBG9G--sqlite-gate-runtime-lifecycle-project/lifecycle-session.jsonl",
            "extracted_at": "2026-10-17 11:42:26",
            "error": null
          },
          {
            "conversation_path": "~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/lifecycle/old-session.jsonl",
            "extracted_at": "2026-01-01T00:00:00Z",
            "error": null
          }
        ],
        "progress": [
          {
            "session_id": "lifecycle-session",
            "turns_seen": 1
          }
        ],
        "derived": [
          {
            "title": "Consolidated decisions — project (2019-12-10)",
            "provenance": "derived",
            "source_ids": "[{\"table\":\"decisions\",\"id\":2},{\"table\":\"decisions\",\"id\":3},{\"table\":\"decisions\",\"id\":4}]"
          }
        ]
      }
    }
  ]
}

Evidence: MCP WAL comparison

{
  "result": "pass",
  "live": true,
  "control": {
    "policy": "unlimited control",
    "pinnedBytes": 73900472,
    "resetBytes": 73937552,
    "pinnedSnapshotRows": 1,
    "durableRows": 43,
    "added": 40,
    "integrity": {
      "integrity_check": "ok"
    },
    "stderr": "RECALL MCP server running\n"
  },
  "target": {
    "policy": "target",
    "pinnedBytes": 73900472,
    "resetBytes": 67108864,
    "pinnedSnapshotRows": 1,
    "durableRows": 43,
    "added": 40,
    "integrity": {
      "integrity_check": "ok"
    },
    "stderr": "RECALL MCP server running\n"
  },
  "note": "Control uses the same real product with only journal_size_limit overridden to -1; it is not a base-commit run."
}

Evidence: Persisted state

{
  "normal": {
    "decisions": [
      {
        "decision": "SQLITE_GATE durable memory survives reopen",
        "project": "sqlite-gate",
        "provenance": "user_authored"
      }
    ]
  },
  "startup-locks": {
    "retried_messages": [
      {
        "session_id": "locked-session",
        "content": "LOCK_TIMEOUT this transcript must survive the failed compact flush and retry.",
        "provenance": "verbatim"
      }
    ]
  },
  "lifecycle": {
    "extracted_decisions": [
      {
        "decision": "HOOK_MEMORY use the shared database for durable extracted decisions",
        "provenance": "extracted"
      }
    ],
    "tracker": [
      {
        "conversation_path": "~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/lifecycle/old-session.jsonl",
        "extracted_at": "2026-01-01T00:00:00Z",
        "error": null
      },
      {
        "conversation_path": "~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/.sqlite-gate-runtime/lifecycle/home/.claude/projects/-Users-ed--no-mistakes-worktrees-fd32172f1538-01M52TP6ZQ2ARNP0T3RN3BBG9G--sqlite-gate-runtime-lifecycle-project/lifecycle-session.jsonl",
        "extracted_at": "2026-10-17 11:42:26",
        "error": null
      }
    ],
    "remaining_locks": [
      {
        "n": 0
      }
    ],
    "messages": [
      {
        "session_id": "lifecycle-session",
        "role": "user",
        "provenance": "verbatim"
      },
      {
        "session_id": "lifecycle-session",
        "role": "assistant",
        "provenance": "verbatim"
      }
    ],
    "progress": [
      {
        "session_id": "lifecycle-session",
        "turns_seen": 1
      }
    ],
    "expired_breadcrumbs": [
      {
        "n": 0
      }
    ],
    "consolidated_sources": [
      {
        "decision": "CONSOLIDATE_GATE old low-importance decision 0",
        "importance": 1
      },
      {
        "decision": "CONSOLIDATE_GATE old low-importance decision 1",
        "importance": 1
      },
      {
        "decision": "CONSOLIDATE_GATE old low-importance decision 2",
        "importance": 1
      }
    ],
    "consolidated_summary": [
      {
        "title": "Consolidated decisions \u2014 project (2019-12-10)",
        "provenance": "derived",
        "source_ids": "[{\"table\":\"decisions\",\"id\":2},{\"table\":\"decisions\",\"id\":3},{\"table\":\"decisions\",\"id\":4}]"
      }
    ]
  }
}

Evidence: Runtime SQLite settings

{
  "runtime_writer_observations": {
    "at closeDb (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/dist/chunk-LISQL7NL.js:3313:8)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 2,
      "wal_autocheckpoint": 1000
    },
    "at flushConversation (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/hooks/RecallPreCompact.ts:393:10)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 1,
      "wal_autocheckpoint": 1000
    },
    "at migrateTrackerJson (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/hooks/lib/extraction-migration.ts:99:8)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 1,
      "wal_autocheckpoint": 1000
    },
    "at getExtractionRecord (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/hooks/lib/extraction-tracker.ts:33:8)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 1,
      "wal_autocheckpoint": 1000
    },
    "at acquireSemaphore (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/hooks/lib/extraction-semaphore.ts:146:8)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 1,
      "wal_autocheckpoint": 1000
    },
    "at ensureDbWritable (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/hooks/lib/sqlite-writers.ts:457:10)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 1,
      "wal_autocheckpoint": 1000
    },
    "at writeExtractionSession (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/hooks/lib/sqlite-writers.ts:134:8)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 1,
      "wal_autocheckpoint": 1000
    },
    "at writeDecisionsBatch (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/hooks/lib/sqlite-writers.ts:202:8)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 1,
      "wal_autocheckpoint": 1000
    },
    "at writeLearningsBatch (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/hooks/lib/sqlite-writers.ts:281:8)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 1,
      "wal_autocheckpoint": 1000
    },
    "at writeBreadcrumbsBatch (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/hooks/lib/sqlite-writers.ts:335:8)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 1,
      "wal_autocheckpoint": 1000
    },
    "at writeExtractionErrors (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/hooks/lib/sqlite-writers.ts:441:8)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 1,
      "wal_autocheckpoint": 1000
    },
    "at writeLoaEntryFromExtraction (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/hooks/lib/sqlite-writers.ts:391:8)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 1,
      "wal_autocheckpoint": 1000
    },
    "at markAsExtracted (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/hooks/lib/extraction-tracker.ts:44:8)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 1,
      "wal_autocheckpoint": 1000
    },
    "at childReleaseLock (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/hooks/lib/extraction-lock.ts:80:8)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 1,
      "wal_autocheckpoint": 1000
    },
    "at releaseSemaphore (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/hooks/lib/extraction-semaphore.ts:159:8)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 1,
      "wal_autocheckpoint": 1000
    },
    "at clearTrackerRow (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/hooks/RecallExtract.ts:864:8)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 1,
      "wal_autocheckpoint": 1000
    },
    "at childAcquireLock (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/hooks/lib/extraction-lock.ts:60:8)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 1,
      "wal_autocheckpoint": 1000
    },
    "at ensureSession (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/hooks/lib/session-progress.ts:74:8)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 1,
      "wal_autocheckpoint": 1000
    },
    "at incrementTurns (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/hooks/lib/session-progress.ts:86:8)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 1,
      "wal_autocheckpoint": 1000
    },
    "at getSessionProgress (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/hooks/lib/session-progress.ts:52:8)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 1,
      "wal_autocheckpoint": 1000
    },
    "at sweepExpiredBreadcrumbs (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/dist/index.js:2101:10)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 2,
      "wal_autocheckpoint": 1000
    },
    "at applyConsolidation (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/hooks/lib/consolidate-core.ts:211:8)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": 67108864,
      "synchronous": 1,
      "wal_autocheckpoint": 1000
    }
  },
  "excluded_from_writer_assertions": {
    "at ingestHostTranscriptBatch (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/dist/index.js:3548:14)": {
      "busy_timeout": 0,
      "journal_mode": "memory",
      "journal_size_limit": -1,
      "synchronous": 0
    },
    "already_closed_connections": "A second close cannot expose PRAGMA values; successful first-close observations are retained.",
    "at tableColumns (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/dist/index.js:2112:8)": {
      "busy_timeout": 0,
      "journal_mode": "wal",
      "journal_size_limit": -1,
      "synchronous": 2
    },
    "at queryDb (~/.no-mistakes/worktrees/fd32172f1538/01M52TP6ZQ2ARNP0T3RN3BBG9G/dist/index.js:2054:8)": {
      "busy_timeout": 5000,
      "journal_mode": "wal",
      "journal_size_limit": -1,
      "synchronous": 2
    }
  },
  "note": "Values read from actual SQLite connections at close during live CLI and hook execution; they supplement persisted-state and CLI evidence."
}

- Outcome: ⚠️ 1 info across 2 runs (19m28s)

Pipeline

Updates from git push no-mistakes

⏭️ **intent** - skipped

✅ No issues found.

✅ **Rebase** - passed

✅ No issues found.

⚠️ **Review** - 2 warnings

⚠️ src/db/connection.ts:30 - Simplification: harden: bound WAL growth — centralize journal_size_limit across all Recall DB writers #192 explicitly requests a shared pragma helper, but the new journal-limit policy has ten separate hook copies alongside this constant. No requirement needs these parallel definitions. Remove the copies by defining the common setup under hooks/lib and consuming it from both CLI and hooks. Other sites: hooks/RecallExtract.ts:862, hooks/RecallPreCompact.ts:346, hooks/lib/consolidate-core.ts:108, hooks/lib/extraction-lock.ts:18, hooks/lib/extraction-migration.ts:12, hooks/lib/extraction-semaphore.ts:15, hooks/lib/extraction-tracker.ts:11, hooks/lib/session-progress.ts:18, hooks/lib/session-start-context.ts:93, hooks/lib/sqlite-writers.ts:59.
⚠️ tests/hooks/extract-core.test.ts:488 - The edited suite retains two source-content-only tests: these assertions inspect RecallExtract.ts text, and tests/hooks/extract-core.test.ts:499 inspects extract-core.ts text. Neither executes the claimed behavior; commented or unreachable calls can satisfy the positive assertions. Under the explicit test-quality rule for directly encountered tests, remove both tests or replace them with isolated execution that asserts actual files and side effects.

⚠️ **Test** - 1 info

🚨 src/lib/memory.ts:295 - Concurrent crash recovery intermittently rejects a real recall add breadcrumb command with SQLITE_SCHEMA: vtable constructor failed: breadcrumbs_fts. One of seven attempts without instrumentation persisted only five of six requested breadcrumbs. Concurrent recovery remains unreliable; whether this predates the change was not established.
🚨 live validation verdict: no-go (8 of 8 scenarios were driven live against the product); failed: Recover a killed WAL writer and accept all six concurrent CLI additions
Live validation: ❌ no-go - 8 of 8 scenarios driven live against the product

Scenario	Result	Live	Evidence
Initialize Recall, add a decision, and find it after reopening	✅ pass	live	cli-and-lock-evidence.log
Write through MCP while a reader pins the WAL; preserve its snapshot and reclaim excess WAL after release	✅ pass	live	mcp-wal-results.json
Add memory while another process holds a startup lock; save it after the lock releases	✅ pass	live	cli-and-lock-evidence.log
Retry PreCompact after a lock timeout; persist one message without prematurely advancing the watermark	✅ pass	live	cli-and-lock-evidence.log
End and re-extract a session; migrate its tracker, preserve extracted memory, and release extraction locks	✅ pass	live	lifecycle-product-transcript.log; hook-runtime-pragmas.json
Submit a prompt and resume the session; retain progress, remove expired memory, and render captured context	✅ pass	live	lifecycle-product-transcript.log; hook-runtime-pragmas.json
Consolidate three old decisions into one derived summary while retaining and demoting their sources	✅ pass	live	lifecycle-product-transcript.log
Recover a killed WAL writer and accept all six concurrent CLI additions	❌ fail	live	recovery-failure.log

bun install --frozen-lockfile
./node_modules/.bin/tsup src/index.ts --format esm --external bun:sqlite
./node_modules/.bin/tsup src/mcp-server.ts --format esm --external bun:sqlite
bun test tests/db/pragmas.test.ts tests/hooks/sqlite-writers-concurrency.test.ts with isolated HOME and Recall paths
bun test tests/hooks/extract-core.test.ts --test-name-pattern 'exclusively locked' with isolated HOME and Recall paths
bun ~/.no-mistakes/evidence/01M52TP6ZQ2ARNP0T3RN3BBG9G/validate-sqlite.ts
bun ~/.no-mistakes/evidence/01M52TP6ZQ2ARNP0T3RN3BBG9G/validate-sqlite.ts --hooks-only after correcting fixture setup
bun ~/.no-mistakes/evidence/01M52TP6ZQ2ARNP0T3RN3BBG9G/validate-sqlite.ts --only='Concurrent CLI' for seven uninstrumented recovery attempts
bun ~/.no-mistakes/evidence/01M52TP6ZQ2ARNP0T3RN3BBG9G/validate-mcp-wal.ts
SQLite persisted-state and integrity readbacks; removed disposable databases, build output, and locally installed test dependencies

🔧 Fix applied.
1 info still open:

ℹ️ tests/db/recovery-concurrency.test.ts - new test file written by agent: tests/db/recovery-concurrency.test.ts
Live validation: ✅ go - 8 of 8 scenarios driven live against the product

Scenario	Result	Live	Evidence
Initialize Recall, add a decision, and find it after reopening	✅ pass	live	Product transcript; Persisted state
Write through MCP while a reader pins its snapshot, then reclaim oversized WAL after reader release	✅ pass	live	MCP WAL comparison: target shrank to 67,108,864 bytes; unlimited control retained 73,937,552 bytes; all 43 records persisted
Add memory while another process briefly locks the database	✅ pass	live	Product transcript: CLI addition completed after the peer released its startup lock
Retry PreCompact after a database timeout without losing or duplicating messages	✅ pass	live	Product transcript; Persisted state: timeout preserved the watermark, retry saved one message, replay added none
Recover a killed WAL writer and retain committed data plus every concurrent CLI addition	✅ pass	live	Product transcript; bun test tests/db/recovery-concurrency.test.ts: 20 rounds of six concurrent CLI additions, with integrity and FTS checks
Run lifecycle hooks and replay extraction with one extracted decision, preserved transcript, recorded progress, and no remaining locks	✅ pass	live	Product transcript; Persisted state; Runtime SQLite settings
Start a session and recall durable memory while sweeping expired breadcrumbs	✅ pass	live	Product transcript; Persisted state: recalled decision displayed and expired breadcrumb removed
Consolidate old decisions while preserving their source lineage	✅ pass	live	Product transcript; Persisted state: one derived summary references three demoted source decisions

bun install --frozen-lockfile --ignore-scripts
bun run --bun tsup src/index.ts src/mcp-server.ts --format esm --clean --external bun:sqlite
bun test tests/db/pragmas.test.ts tests/db/recovery-concurrency.test.ts tests/hooks/sqlite-writers-concurrency.test.ts with isolated HOME, runtime paths, and database
bun test tests/hooks/extract-core.test.ts --test-name-pattern 'exclusively locked'
bun ~/.no-mistakes/evidence/01M52TP6ZQ2ARNP0T3RN3BBG9G/round-2/validate-sqlite.ts
bun ~/.no-mistakes/evidence/01M52TP6ZQ2ARNP0T3RN3BBG9G/round-2/validate-mcp-wal.ts
Checked live connection PRAGMAs and queried persisted SQLite state; saved evidence before removing disposable databases and build output

✅ **Document** - passed

✅ No issues found.

✅ **Lint** - passed

✅ No issues found.

✅ **Push** - passed

✅ No issues found.

edheltzel

64MiB should be large enough but we might consider bumping this to 128 or 256MiB per session run.

…#212, #192)

…AL cap (#192)

…rash recovery

…entation

* fix(db): busy_timeout before WAL and cap journal size on every writer (#212, #192) * docs(db): journal_size_limit truncates after checkpoint, not a hard WAL cap (#192) * no-mistakes(test): Prevent concurrent SQLite schema failures during crash recovery * no-mistakes(document): Clarify SQLite locking and WAL retention documentation --------- Co-authored-by: Atlas-Key <296298943+Atlas-Key@users.noreply.github.com>

Atlas-Key requested a review from edheltzel as a code owner October 17, 2026 11:47

edheltzel approved these changes Oct 17, 2026

View reviewed changes

Atlas-Key added 4 commits October 17, 2026 10:12

fix(db): busy_timeout before WAL and cap journal size on every writer (…

3e3e4a0

…#212, #192)

docs(db): journal_size_limit truncates after checkpoint, not a hard W…

f920177

…AL cap (#192)

no-mistakes(test): Prevent concurrent SQLite schema failures during c…

b4c3ef0

…rash recovery

no-mistakes(document): Clarify SQLite locking and WAL retention docum…

c82471e

…entation

edheltzel force-pushed the harden/sqlite-pragmas branch from 711054e to c82471e Compare October 17, 2026 14:12

edheltzel merged commit a130387 into main Oct 17, 2026
3 checks passed

edheltzel deleted the harden/sqlite-pragmas branch October 17, 2026 15:16

Navigation Menu

Repository navigation

fix: harden SQLite lock handling and WAL retention - #335

fix: harden SQLite lock handling and WAL retention#335
edheltzel merged 4 commits into
mainfrom
harden/sqlite-pragmas

Atlas-Key commented Oct 17, 2026

Uh oh!

edheltzel left a comment

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

Atlas-Key commented Oct 17, 2026

What Changed

Risk Assessment

Testing

Pipeline

Uh oh!

edheltzel left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants