feat(ql3): add cluster log retention authority

docs/QINGLONG_3_0_ARCHITECTURE_RFC.md

@@ -11,6 +11,17 @@
 
 最新增量证据（2026-08-12）：
 
+- D-291/ADR-0379（进行中，PostgreSQL authority 阶段完成）
+  Cluster Run Attempt 日志 retention 第一阶段已冻结并实现多副本权威边界，且没有新增 package：共享 claim contract 位于既有
+  Runtime Core Run log-retention 目录，PostgreSQL v54 增加 durable control 与 immutable tombstone、terminal remote Worker
+  candidate index 和 `run_attempt_log_retention` capability。候选只允许 runtime-owned、Run/Attempt 双终态、非 lost、canonical
+  `remote_worker/wlog-*` 且超过 cutoff；每批最多 16 条，通过短 `READ COMMITTED` 事务和 `FOR UPDATE ... SKIP LOCKED` 获取
+  owner/token/version/expiry fence。S3 网络调用明确位于数据库事务之外；finalize 在第二个短事务中重验完整 fence 与 durable
+  Run/Attempt identity，写 exact retirement record 后删除 control。retry/manual 持久化失败分类并提供最长 24 小时 backoff；lease
+  过期可由其他副本安全接管。`ql3_runtime` 可完整维护 control，但 tombstone 只有 SELECT/INSERT，其他角色保持零权限；repository
+  已提供 digest/identity 失败关闭的 retention state read，为 Cluster 410 wiring 提供权威来源。migration/schema/readiness/repository
+  定向 63/63 通过。ADR 暂保持 Proposed；下一阶段继续完成 validated S3 HEAD、ETag/VersionId 条件删除、bounded lifecycle、MinIO
+  failure matrix 与 PostgreSQL HA failover，全部完整门通过后才接受。
 - D-290/ADR-0378（已接受）
   Local Run Attempt 日志 retention 已形成真实纵向切片：Runtime Core 增加精确 identity、canonical SHA-256 的 immutable
   retirement record、容量压力策略、有界 page/delete budget 与 durable cursor；日志读取在存储前检查 tombstone，并在 missing 后二次

docs/adr/ADR-0379-cluster-run-attempt-log-retention.md

@@ -0,0 +1,74 @@
+# ADR-0379：Cluster Run Attempt 日志多副本保留与条件删除
+
+- 状态：Proposed（PostgreSQL authority 已实现，S3/lifecycle/HA 验收待完成）
+- 日期：2026-08-12
+- 关联 RFC：QL-RFC-0001 D-291
+- 前置决策：ADR-0026、ADR-0027、ADR-0377、ADR-0378
+
+## 上下文
+
+Local retention 已证明 durable tombstone、删除后收敛和读取 410 的通用语义，但 Cluster 不能复用 Local cursor + unlink 模型。多个控制面副本可能同时发现同一 remote Worker 日志；对象删除需要跨 S3 网络，PostgreSQL 事务不能覆盖该网络调用；删除响应也可能丢失。若没有 durable ownership fence，副本会重复删除或把另一个副本的结果写成自己的证据；若删除前写 tombstone，则可能把仍存在的对象声明为 retired。
+
+Cluster 还必须保持与低配 Local 部署的物理隔离：本能力不得让 Edge/Standalone closure 引入 PostgreSQL、AWS SDK、额外连接或后台任务。
+
+## 决策
+
+### 1. 不新增 package
+
+共享多副本 claim contract 放在 `runtime-core` 的既有 Run log-retention 目录；PostgreSQL authority 放在 `cluster-postgres`；S3 条件删除和调度 lifecycle 放在 `cluster-control`。不创建单文件 retention 微包，也不把 Cluster 依赖加入 Local closure。
+
+### 2. PostgreSQL v54 是唯一 ownership authority
+
+`pg-0055-run-attempt-log-retention` 将 control-core 升至 v54，并声明 `run_attempt_log_retention` capability：
+
+- `run_attempt_log_retention_controls` 保存精确 Project/Run/Attempt/Artifact identity、eligibility、claim owner/token/version/expiry、retry time、failure count 与最后失败分类；
+- `run_attempt_log_artifact_tombstones` 保存 immutable `qinglong/run-attempt-log-retirement@v1` 标量证据与 canonical digest；
+- terminal remote Worker Attempt 增加局部候选索引；
+- `ql3_runtime` 对 control 拥有 `SELECT/INSERT/UPDATE/DELETE`，对 tombstone 仅拥有 `SELECT/INSERT`，其他运行角色继续零权限。
+
+候选必须同时满足 runtime-owned Run、Run/Attempt 均为非 `lost` 终态、`remote_worker` executor、canonical `wlog-*` identity、Run/Attempt 均超过 retention cutoff 且不存在 tombstone。claim 在短 `READ COMMITTED` 事务中以 `FOR UPDATE ... SKIP LOCKED` 获取；冲突更新再次校验 immutable identity、retry due/lease expiry 与单调 version。
+
+### 3. S3 调用不进入数据库事务
+
+每个副本先取得有界 durable claim，提交并释放 PostgreSQL client 后才执行 S3：
+
+1. validated HEAD 校验 content type、metadata identity、checksum、byte length，并取得 ETag 与可用的 VersionId；
+2. 对 versioned object 使用精确 VersionId，对未版本化对象使用 ETag `If-Match` 条件删除；
+3. 412/对象身份变化失败关闭并进入 retry/manual，不得删除新对象；
+4. 删除成功或 HEAD 已不存在后，开启第二个短 PostgreSQL 事务；
+5. 事务重验 owner/token/version/expiry、terminal Run/Attempt 与 immutable identity，插入 exact tombstone 后删除 control；
+6. 删除响应丢失时，lease 过期后的新 claim 以 HEAD absent 写入 `already_absent`，最终收敛。
+
+### 4. 有界调度与退避
+
+每轮 claim 不超过 16 条，lease 范围 5 秒至 5 分钟；retry delay 最大 24 小时。副本不得持有跨 sweep cursor，也不得为每个 Attempt 建 timer。调度复用 Cluster control application 既有 lifecycle，并受每轮 claim 数、删除数和 wall-clock budget 共同限制。`artifact_unavailable`、`artifact_integrity_mismatch`、`retirement_record_unavailable` 是持久化失败分类；达到策略阈值后转 manual，避免坏对象形成热循环。
+
+### 5. 读取收敛
+
+PostgreSQL claim repository 同时实现 retention state reader。Cluster 日志读取在 S3 前检查 tombstone，并在 S3 missing 后二次检查：已 retired 返回 410；没有 tombstone 的 missing 继续保持 503，而不是伪造 retention。tombstone identity 或 digest 漂移失败关闭。
+
+## 阶段验收
+
+第一阶段已完成 PostgreSQL authority：共享 claim contract、v54 migration、typed Drizzle/schema/readiness、最小权限、短事务 claim、完整 lease fence、retry/manual settlement、exact tombstone finalize/replay、tombstone state read。定向 63 项 migration/schema/readiness/repository 门通过；`runtime-core` 498 项全通过，`cluster-postgres` 302 项通过、1 项条件跳过，`cluster-control` 216 项通过、2 项条件跳过。
+
+阶段收口还通过 18 个 QL3 workspace package 的完整 build/test 门，以及后端兼容回归 1163 项通过、2 项条件跳过、0 失败/取消。v54 readiness 引入的两张表已同步进入 `cluster-control` 测试数据库的最小权限 fixture，避免旧 v53 fixture 把正常启动误判为 `runtime_role_invalid`。
+
+结构门保持 18 个 workspace package，`singleSourcePackages=[]`、`shallowSourcePackages=[]`。新 repository 只从明确的 Cluster runtime entrypoint 导出，不扩大 package root；PostgreSQL migration append-only `ordered_ledger` 的 reviewed hard cap 随 pg-0055 由 57 精确推进到 58，不把版本账本伪拆为子目录。
+
+ADR 保持 Proposed，只有以下剩余项全部完成后才转 Accepted：
+
+1. S3 validated HEAD + ETag/VersionId 条件删除和失败矩阵；
+2. Cluster service 的 bounded claim/delete/backoff/manual 策略；
+3. production composition、lifecycle drain 与读取 410；
+4. MinIO versioned/unversioned 集成、响应丢失重放；
+5. PostgreSQL 18 HA failover 中 lease takeover/tombstone 收敛；
+6. 完整 package/backend/boundary/Profile/image gates，证明 Local closure 无 PostgreSQL/AWS SDK 回归。
+
+## 被否决的替代方案
+
+1. **复用 Local durable cursor**：无法形成多副本 ownership，拒绝。
+2. **在 PostgreSQL 事务中调用 S3**：把网络停顿扩大为数据库锁和故障域，拒绝。
+3. **删除前写 tombstone**：可能把仍存在的对象声明为 retired，拒绝。
+4. **无条件 DeleteObject**：无法防止对象身份变化或运维侧替换，拒绝。
+5. **授予 tombstone UPDATE/DELETE**：破坏 immutable evidence，拒绝。
+6. **新增 Cluster retention 微包**：现有聚合边界足够，且会继续放大 package 碎片化，拒绝。

docs/adr/README.md

@@ -379,6 +379,10 @@
 | [ADR-0373](./ADR-0373-profile-reachable-runtime-javascript-projection.md)                                      | Profile 可达的 Runtime JavaScript 投影                                                                                     | Accepted                     |
 | [ADR-0374](./ADR-0374-shared-bounded-task-discovery-http-api.md)                                               | 共享、有界的 Task Discovery HTTP API                                                                                       | Accepted |
 | [ADR-0375](./ADR-0375-shared-current-task-point-read-api.md)                                                   | 共享的 current Task point-read API                                                                                         | Proposed（设计冻结，实现中） |
+| [ADR-0376](./ADR-0376-policy-and-digest-fenced-task-start.md)                                                  | Policy 与 digest fenced 的 Task Start                                                                                      | Accepted |
+| [ADR-0377](./ADR-0377-profile-aware-run-attempt-log-range-read.md)                                             | Profile-aware Run Attempt 日志 Range 读取                                                                                  | Accepted |
+| [ADR-0378](./ADR-0378-local-run-attempt-log-retention-and-tombstones.md)                                       | Local Run Attempt 日志有界保留与 durable tombstone                                                                         | Accepted |
+| [ADR-0379](./ADR-0379-cluster-run-attempt-log-retention.md)                                                    | Cluster Run Attempt 日志多副本保留与条件删除                                                                                | Proposed（PostgreSQL authority 已完成） |
 
 ## 规则
 

docs/ql3-package-boundaries.json

@@ -6,7 +6,7 @@
     "reviewedDenseDirectories": [
       {
         "kind": "ordered_ledger",
-        "maxDirectSourceFiles": 57,
+        "maxDirectSourceFiles": 58,
         "path": "packages/ql3-cluster-postgres/src/migrations",
         "rationale": "PostgreSQL migrations are an append-only version ledger whose ordering and discoverability are safer in one reviewed directory."
       },

packages/ql3-cluster-control/test/application.test.cjs

@@ -137,6 +137,8 @@ function runtimePrivileges() {
     tool_invocation_input_artifacts: [true, true, false, false],
     tool_invocation_preview_artifacts: [true, true, false, false],
     run_attempts: [true, true, true, false],
+    run_attempt_log_retention_controls: [true, true, true, true],
+    run_attempt_log_artifact_tombstones: [true, true, false, false],
     run_recovery_controls: [true, true, true, false],
     worker_sessions: [true, true, true, false],
     run_dispatch_leases: [true, true, true, false],

packages/ql3-cluster-control/test/bootstrap.test.cjs

@@ -51,6 +51,8 @@ function runtimePrivileges() {
     tool_invocation_input_artifacts: [true, true, false, false],
     tool_invocation_preview_artifacts: [true, true, false, false],
     run_attempts: [true, true, true, false],
+    run_attempt_log_retention_controls: [true, true, true, true],
+    run_attempt_log_artifact_tombstones: [true, true, false, false],
     run_recovery_controls: [true, true, true, false],
     worker_sessions: [true, true, true, false],
     run_dispatch_leases: [true, true, true, false],

packages/ql3-cluster-postgres/src/entrypoints/runtime.ts

@@ -54,6 +54,7 @@ export { PostgresToolResultKeyCatalogReader } from '../tool-execution/toolResult
 export { PostgresToolResultRekeyReader } from '../tool-execution/toolResultRekeyRepository';
 
 export * from '../run/runRepository';
+export * from '../run/runAttemptLogRetentionClaimRepository';
 export * from '../security/projectPolicyRepository';
 export * from '../security/apiCredentialRepository';
 export * from '../security/securityAuditRepository';

packages/ql3-cluster-postgres/src/migration/migrationManifest.ts

@@ -278,5 +278,10 @@ export const postgresqlMainMigrationManifest: MigrationStreamManifest =
         checksum:
           '5e3e6b222269f095e0d7a985fdeb0ea154510e59dfe15873192af8c8d603fca3',
       }),
+      Object.freeze({
+        id: 'pg-0055-run-attempt-log-retention',
+        checksum:
+          'c775c65ec03ae3a1606f899064d2d38fa63fd136ce52cbd1b1172c3a51e6bf30',
+      }),
     ]),
   });

packages/ql3-cluster-postgres/src/migrations/index.ts

@@ -57,6 +57,7 @@ import { pg0051AutomationManagementBoundaryMigration } from './pg-0051-automatio
 import { pg0052AutomationManagementIdentityKeysetLedgerMigration } from './pg-0052-automation-management-identity-keyset-ledger';
 import { pg0053PluginPackageWorkflowRunListIndexMigration } from './pg-0053-plugin-package-workflow-run-list-index';
 import { pg0054ApprovalManagementBoundaryMigration } from './pg-0054-approval-management-boundary';
+import { pg0055RunAttemptLogRetentionMigration } from './pg-0055-run-attempt-log-retention';
 
 export const postgresqlMainMigrationStream: MigrationStreamDefinition<PostgresMigrationContext> =
   Object.freeze({
@@ -119,5 +120,6 @@ export const postgresqlMainMigrationStream: MigrationStreamDefinition<PostgresMi
       pg0052AutomationManagementIdentityKeysetLedgerMigration,
       pg0053PluginPackageWorkflowRunListIndexMigration,
       pg0054ApprovalManagementBoundaryMigration,
+      pg0055RunAttemptLogRetentionMigration,
     ]),
   });

packages/ql3-cluster-postgres/src/migrations/pg-0055-run-attempt-log-retention.ts

@@ -0,0 +1,189 @@
+import { CAPABILITIES_V53 } from './pg-0054-approval-management-boundary';
+import { definePostgresSqlMigration } from './sqlMigration';
+
+export const CAPABILITIES_V54 = CAPABILITIES_V53.replace(
+  '"run_core":1,',
+  '"run_attempt_log_retention":1,"run_core":1,',
+);
+
+export const pg0055RunAttemptLogRetentionMigration =
+  definePostgresSqlMigration({
+    id: 'pg-0055-run-attempt-log-retention',
+    statements: [
+      `
+CREATE TABLE "ql3"."run_attempt_log_retention_controls" (
+  attempt_id varchar(36) PRIMARY KEY,
+  project_id varchar(128) NOT NULL,
+  run_id varchar(36) NOT NULL,
+  log_artifact_id varchar(36) NOT NULL,
+  executor_type varchar(32) NOT NULL,
+  finished_at_ms bigint NOT NULL,
+  eligible_at_ms bigint NOT NULL,
+  state varchar(16) NOT NULL,
+  claim_owner varchar(128),
+  claim_token varchar(64),
+  claim_version integer NOT NULL DEFAULT 1,
+  claim_expires_at_ms bigint,
+  next_claim_at_ms bigint,
+  failure_count integer NOT NULL DEFAULT 0,
+  last_failure_code varchar(64),
+  created_at_ms bigint NOT NULL,
+  updated_at_ms bigint NOT NULL,
+  CONSTRAINT ql3_run_log_retention_control_artifact_key
+    UNIQUE (log_artifact_id),
+  CONSTRAINT ql3_run_log_retention_control_identity_check CHECK (
+    char_length(project_id) BETWEEN 1 AND 128
+    AND char_length(run_id) BETWEEN 1 AND 36
+    AND char_length(attempt_id) BETWEEN 1 AND 36
+    AND log_artifact_id ~ '^wlog-[a-f0-9]{30}$'
+    AND executor_type = 'remote_worker'
+  ),
+  CONSTRAINT ql3_run_log_retention_control_time_check CHECK (
+    finished_at_ms >= 0
+    AND eligible_at_ms >= finished_at_ms
+    AND created_at_ms >= 0
+    AND updated_at_ms >= created_at_ms
+  ),
+  CONSTRAINT ql3_run_log_retention_control_state_check
+    CHECK (state IN ('claimed', 'retry', 'manual')),
+  CONSTRAINT ql3_run_log_retention_control_claim_owner_check CHECK (
+    claim_owner IS NULL
+    OR claim_owner ~ '^[A-Za-z0-9][A-Za-z0-9._:-]{0,127}$'
+  ),
+  CONSTRAINT ql3_run_log_retention_control_claim_token_check CHECK (
+    claim_token IS NULL
+    OR claim_token ~ '^[A-Za-z0-9][A-Za-z0-9._:-]{15,63}$'
+  ),
+  CONSTRAINT ql3_run_log_retention_control_claim_version_check
+    CHECK (claim_version BETWEEN 1 AND 2147483647),
+  CONSTRAINT ql3_run_log_retention_control_claim_expiry_check
+    CHECK (claim_expires_at_ms IS NULL OR claim_expires_at_ms >= 0),
+  CONSTRAINT ql3_run_log_retention_control_next_claim_check
+    CHECK (next_claim_at_ms IS NULL OR next_claim_at_ms >= 0),
+  CONSTRAINT ql3_run_log_retention_control_failure_count_check
+    CHECK (failure_count BETWEEN 0 AND 2147483647),
+  CONSTRAINT ql3_run_log_retention_control_failure_code_check CHECK (
+    last_failure_code IS NULL
+    OR last_failure_code IN (
+      'artifact_unavailable',
+      'artifact_integrity_mismatch',
+      'retirement_record_unavailable'
+    )
+  ),
+  CONSTRAINT ql3_run_log_retention_control_state_shape_check CHECK (
+    (
+      state = 'claimed'
+      AND claim_owner IS NOT NULL
+      AND claim_token IS NOT NULL
+      AND claim_expires_at_ms IS NOT NULL
+      AND next_claim_at_ms IS NULL
+    )
+    OR (
+      state = 'retry'
+      AND claim_owner IS NULL
+      AND claim_token IS NULL
+      AND claim_expires_at_ms IS NULL
+      AND next_claim_at_ms IS NOT NULL
+      AND last_failure_code IS NOT NULL
+    )
+    OR (
+      state = 'manual'
+      AND claim_owner IS NULL
+      AND claim_token IS NULL
+      AND claim_expires_at_ms IS NULL
+      AND next_claim_at_ms IS NULL
+      AND last_failure_code IS NOT NULL
+    )
+  ),
+  CONSTRAINT ql3_run_log_retention_control_attempt_fk
+    FOREIGN KEY (attempt_id) REFERENCES "ql3"."run_attempts" (id)
+    ON DELETE CASCADE,
+  CONSTRAINT ql3_run_log_retention_control_run_fk
+    FOREIGN KEY (run_id) REFERENCES "ql3"."runs" (id)
+    ON DELETE CASCADE
+)
+      `.trim(),
+      `
+CREATE TABLE "ql3"."run_attempt_log_artifact_tombstones" (
+  log_artifact_id varchar(36) PRIMARY KEY,
+  project_id varchar(128) NOT NULL,
+  run_id varchar(36) NOT NULL,
+  attempt_id varchar(36) NOT NULL,
+  executor_type varchar(32) NOT NULL,
+  finished_at_ms bigint NOT NULL,
+  eligible_at_ms bigint NOT NULL,
+  retired_at_ms bigint NOT NULL,
+  disposition varchar(16) NOT NULL,
+  byte_length bigint NOT NULL,
+  truncated varchar(16) NOT NULL,
+  maximum_bytes bigint,
+  truncation_observed_at_ms bigint,
+  record_digest char(64) NOT NULL,
+  CONSTRAINT ql3_run_log_tombstone_attempt_key UNIQUE (attempt_id),
+  CONSTRAINT ql3_run_log_tombstone_identity_check CHECK (
+    char_length(project_id) BETWEEN 1 AND 128
+    AND char_length(run_id) BETWEEN 1 AND 36
+    AND char_length(attempt_id) BETWEEN 1 AND 36
+    AND log_artifact_id ~ '^wlog-[a-f0-9]{30}$'
+    AND executor_type = 'remote_worker'
+  ),
+  CONSTRAINT ql3_run_log_tombstone_time_check CHECK (
+    finished_at_ms >= 0
+    AND eligible_at_ms >= finished_at_ms
+    AND retired_at_ms >= eligible_at_ms
+  ),
+  CONSTRAINT ql3_run_log_tombstone_disposition_check CHECK (
+    disposition IN ('deleted', 'already_absent')
+    AND (disposition <> 'already_absent' OR byte_length = 0)
+  ),
+  CONSTRAINT ql3_run_log_tombstone_size_check
+    CHECK (byte_length BETWEEN 0 AND 1073741824),
+  CONSTRAINT ql3_run_log_tombstone_truncation_check CHECK (
+    (
+      truncated = 'unknown'
+      AND maximum_bytes IS NULL
+      AND truncation_observed_at_ms IS NULL
+    )
+    OR (
+      truncated IN ('true', 'false')
+      AND maximum_bytes >= 1
+      AND truncation_observed_at_ms >= 0
+    )
+  ),
+  CONSTRAINT ql3_run_log_tombstone_digest_check
+    CHECK (record_digest ~ '^[a-f0-9]{64}$'),
+  CONSTRAINT ql3_run_log_tombstone_attempt_fk
+    FOREIGN KEY (attempt_id) REFERENCES "ql3"."run_attempts" (id)
+    ON DELETE CASCADE,
+  CONSTRAINT ql3_run_log_tombstone_run_fk
+    FOREIGN KEY (run_id) REFERENCES "ql3"."runs" (id)
+    ON DELETE CASCADE
+)
+      `.trim(),
+      `CREATE INDEX ql3_run_log_retention_retry_idx ON "ql3"."run_attempt_log_retention_controls" (next_claim_at_ms, finished_at_ms, attempt_id) WHERE state = 'retry'`,
+      `CREATE INDEX ql3_run_log_retention_claim_expiry_idx ON "ql3"."run_attempt_log_retention_controls" (claim_expires_at_ms, finished_at_ms, attempt_id) WHERE state = 'claimed'`,
+      `CREATE INDEX ql3_run_log_tombstone_retired_idx ON "ql3"."run_attempt_log_artifact_tombstones" (retired_at_ms, attempt_id)`,
+      `CREATE INDEX ql3_run_log_retention_candidate_idx ON "ql3"."run_attempts" (finished_at_ms, id) WHERE executor_type = 'remote_worker' AND log_artifact_id IS NOT NULL AND status IN ('succeeded', 'failed', 'cancelled', 'timed_out')`,
+      `GRANT SELECT, INSERT, UPDATE, DELETE ON "ql3"."run_attempt_log_retention_controls" TO ql3_runtime`,
+      `GRANT SELECT, INSERT ON "ql3"."run_attempt_log_artifact_tombstones" TO ql3_runtime`,
+      `
+DO $ql3$
+BEGIN
+  UPDATE "ql3"."schema_capabilities"
+  SET contract_version = 54,
+      migration_id = 'pg-0055-run-attempt-log-retention',
+      capabilities = '${CAPABILITIES_V54}'::jsonb,
+      updated_at_ms = floor(extract(epoch FROM transaction_timestamp()) * 1000)::bigint
+  WHERE contract_name = 'control-core'
+    AND contract_version = 53
+    AND migration_id = 'pg-0054-approval-management-boundary'
+    AND capabilities = '${CAPABILITIES_V53}'::jsonb;
+  IF NOT FOUND THEN
+    RAISE EXCEPTION 'control-core capability is not at version 53'
+      USING ERRCODE = 'check_violation';
+  END IF;
+END
+$ql3$
+      `.trim(),
+    ],
+  });

packages/ql3-cluster-postgres/src/run/runAttemptLogRetentionClaimRepository.ts

@@ -0,0 +1,746 @@
+import { randomUUID } from 'node:crypto';
+
+import {
+  MAX_CLUSTER_RUN_ATTEMPT_LOG_RETENTION_CLAIMS,
+  MAX_CLUSTER_RUN_ATTEMPT_LOG_RETENTION_LEASE_MS,
+  MAX_CLUSTER_RUN_ATTEMPT_LOG_RETENTION_RETRY_DELAY_MS,
+  MIN_CLUSTER_RUN_ATTEMPT_LOG_RETENTION_LEASE_MS,
+  type ClusterRunAttemptLogRetentionClaim,
+  type ClusterRunAttemptLogRetentionClaimPage,
+  type ClusterRunAttemptLogRetentionClaimRepository,
+  type ClusterRunAttemptLogRetentionFailureCode,
+  type ClusterRunAttemptLogRetentionSettlement,
+} from '@qinglong/runtime-core/cluster-run-attempt-log-retention';
+import {
+  MAX_RUN_ATTEMPT_LOG_RETENTION_MS,
+  MIN_RUN_ATTEMPT_LOG_RETENTION_MS,
+  RunAttemptLogRetentionUnavailableError,
+  normalizeRunAttemptLogRetentionCandidate,
+  normalizeRunAttemptLogRetirementRecord,
+  type RunAttemptLogRetentionState,
+  type RunAttemptLogRetirementRecord,
+} from '@qinglong/runtime-core/run-attempt-log-retention';
+import type { RunAttemptLogReadIdentity } from '@qinglong/runtime-core/run-attempt-log-read';
+import type {
+  PostgresPool,
+  PostgresQueryable,
+} from '@qinglong/runtime-core';
+
+type Row = Record<string, unknown>;
+
+const CLAIM_SQL = `
+WITH observation AS (
+  SELECT floor(extract(epoch FROM statement_timestamp()) * 1000)::bigint
+    AS observed_at_ms
+), eligible AS (
+  SELECT attempt.id,
+         run.project_id,
+         attempt.run_id,
+         attempt.log_artifact_id,
+         attempt.executor_type,
+         attempt.finished_at_ms,
+         observation.observed_at_ms
+  FROM "ql3"."run_attempts" AS attempt
+  JOIN "ql3"."runs" AS run ON run.id = attempt.run_id
+  CROSS JOIN observation
+  LEFT JOIN "ql3"."run_attempt_log_retention_controls" AS control
+    ON control.attempt_id = attempt.id
+  LEFT JOIN "ql3"."run_attempt_log_artifact_tombstones" AS tombstone
+    ON tombstone.attempt_id = attempt.id
+      OR tombstone.log_artifact_id = attempt.log_artifact_id
+  WHERE run.execution_owner = 'runtime'
+    AND run.status IN ('succeeded', 'failed', 'cancelled', 'timed_out')
+    AND run.finished_at_ms IS NOT NULL
+    AND attempt.status IN ('succeeded', 'failed', 'cancelled', 'timed_out')
+    AND attempt.executor_type = 'remote_worker'
+    AND attempt.finished_at_ms IS NOT NULL
+    AND attempt.log_artifact_id ~ '^wlog-[a-f0-9]{30}$'
+    AND attempt.finished_at_ms <= observation.observed_at_ms - $1::bigint
+    AND run.finished_at_ms <= observation.observed_at_ms - $1::bigint
+    AND tombstone.attempt_id IS NULL
+    AND (
+      control.attempt_id IS NULL
+      OR (
+        control.claim_version < 2147483647
+        AND (
+          (control.state = 'retry'
+            AND control.next_claim_at_ms <= observation.observed_at_ms)
+          OR (control.state = 'claimed'
+            AND control.claim_expires_at_ms <= observation.observed_at_ms)
+        )
+      )
+    )
+  ORDER BY attempt.finished_at_ms, attempt.id
+  FOR UPDATE OF attempt SKIP LOCKED
+  LIMIT $2
+)
+INSERT INTO "ql3"."run_attempt_log_retention_controls" (
+  attempt_id, project_id, run_id, log_artifact_id, executor_type,
+  finished_at_ms, eligible_at_ms, state, claim_owner, claim_token,
+  claim_version, claim_expires_at_ms, failure_count, last_failure_code,
+  created_at_ms, updated_at_ms
+)
+SELECT id, project_id, run_id, log_artifact_id, executor_type,
+       finished_at_ms, finished_at_ms + $1::bigint, 'claimed', $3, $4,
+       1, observed_at_ms + $5::bigint, 0, NULL,
+       observed_at_ms, observed_at_ms
+FROM eligible
+ON CONFLICT (attempt_id) DO UPDATE
+SET state = 'claimed',
+    claim_owner = EXCLUDED.claim_owner,
+    claim_token = EXCLUDED.claim_token,
+    claim_version = "ql3"."run_attempt_log_retention_controls".claim_version + 1,
+    claim_expires_at_ms = EXCLUDED.claim_expires_at_ms,
+    next_claim_at_ms = NULL,
+    updated_at_ms = GREATEST(
+      "ql3"."run_attempt_log_retention_controls".updated_at_ms,
+      EXCLUDED.updated_at_ms
+    )
+WHERE "ql3"."run_attempt_log_retention_controls".project_id = EXCLUDED.project_id
+  AND "ql3"."run_attempt_log_retention_controls".run_id = EXCLUDED.run_id
+  AND "ql3"."run_attempt_log_retention_controls".log_artifact_id = EXCLUDED.log_artifact_id
+  AND "ql3"."run_attempt_log_retention_controls".executor_type = EXCLUDED.executor_type
+  AND "ql3"."run_attempt_log_retention_controls".finished_at_ms = EXCLUDED.finished_at_ms
+  AND "ql3"."run_attempt_log_retention_controls".eligible_at_ms = EXCLUDED.eligible_at_ms
+  AND "ql3"."run_attempt_log_retention_controls".claim_version < 2147483647
+  AND (
+    ("ql3"."run_attempt_log_retention_controls".state = 'retry'
+      AND "ql3"."run_attempt_log_retention_controls".next_claim_at_ms <= EXCLUDED.updated_at_ms)
+    OR ("ql3"."run_attempt_log_retention_controls".state = 'claimed'
+      AND "ql3"."run_attempt_log_retention_controls".claim_expires_at_ms <= EXCLUDED.updated_at_ms)
+  )
+RETURNING project_id AS "projectId", run_id AS "runId",
+  attempt_id AS "attemptId", log_artifact_id AS "logArtifactId",
+  executor_type AS "executorType", finished_at_ms AS "finishedAtMs",
+  eligible_at_ms AS "eligibleAtMs", updated_at_ms AS "observedAtMs",
+  claim_owner AS "claimOwner", claim_token AS "claimToken",
+  claim_version AS "claimVersion", claim_expires_at_ms AS "claimExpiresAtMs",
+  failure_count AS "failureCount"
+`.trim();
+
+const LOCK_CLAIM_SQL = `
+WITH observation AS (
+  SELECT floor(extract(epoch FROM statement_timestamp()) * 1000)::bigint
+    AS observed_at_ms
+)
+SELECT control.project_id AS "projectId", control.run_id AS "runId",
+  control.attempt_id AS "attemptId", control.log_artifact_id AS "logArtifactId",
+  control.executor_type AS "executorType", control.finished_at_ms AS "finishedAtMs",
+  control.eligible_at_ms AS "eligibleAtMs",
+  control.claim_owner AS "claimOwner", control.claim_token AS "claimToken",
+  control.claim_version AS "claimVersion",
+  control.claim_expires_at_ms AS "claimExpiresAtMs",
+  control.failure_count AS "failureCount",
+  observation.observed_at_ms AS "observedAtMs"
+FROM "ql3"."run_attempt_log_retention_controls" AS control
+JOIN "ql3"."run_attempts" AS attempt ON attempt.id = control.attempt_id
+JOIN "ql3"."runs" AS run ON run.id = control.run_id
+CROSS JOIN observation
+WHERE control.attempt_id = $1
+  AND control.state = 'claimed'
+  AND control.claim_owner = $2
+  AND control.claim_token = $3
+  AND control.claim_version = $4
+  AND control.claim_expires_at_ms = $5::bigint
+  AND control.claim_expires_at_ms > observation.observed_at_ms
+  AND attempt.run_id = control.run_id
+  AND attempt.log_artifact_id = control.log_artifact_id
+  AND attempt.executor_type = 'remote_worker'
+  AND attempt.finished_at_ms = control.finished_at_ms
+  AND attempt.status IN ('succeeded', 'failed', 'cancelled', 'timed_out')
+  AND run.project_id = control.project_id
+  AND run.execution_owner = 'runtime'
+  AND run.status IN ('succeeded', 'failed', 'cancelled', 'timed_out')
+FOR UPDATE OF control
+`.trim();
+
+const INSERT_TOMBSTONE_SQL = `
+INSERT INTO "ql3"."run_attempt_log_artifact_tombstones" (
+  log_artifact_id, project_id, run_id, attempt_id, executor_type,
+  finished_at_ms, eligible_at_ms, retired_at_ms, disposition,
+  byte_length, truncated, maximum_bytes, truncation_observed_at_ms,
+  record_digest
+) VALUES ($1, $2, $3, $4, $5, $6::bigint, $7::bigint, $8::bigint,
+  $9, $10::bigint, $11, $12::bigint, $13::bigint, $14)
+ON CONFLICT DO NOTHING
+RETURNING record_digest AS "recordDigest"
+`.trim();
+
+const READ_TOMBSTONE_SQL = `
+SELECT record_digest AS "recordDigest"
+FROM "ql3"."run_attempt_log_artifact_tombstones"
+WHERE attempt_id = $1 OR log_artifact_id = $2
+`.trim();
+
+const DELETE_CONTROL_SQL = `
+DELETE FROM "ql3"."run_attempt_log_retention_controls"
+WHERE attempt_id = $1 AND state = 'claimed'
+  AND claim_owner = $2 AND claim_token = $3 AND claim_version = $4
+  AND claim_expires_at_ms = $5::bigint
+`.trim();
+
+const SET_FAILURE_SQL = `
+WITH observation AS (
+  SELECT floor(extract(epoch FROM statement_timestamp()) * 1000)::bigint
+    AS observed_at_ms
+)
+UPDATE "ql3"."run_attempt_log_retention_controls" AS control
+SET state = $6,
+    claim_owner = NULL,
+    claim_token = NULL,
+    claim_expires_at_ms = NULL,
+    next_claim_at_ms = CASE WHEN $6 = 'retry'
+      THEN observation.observed_at_ms + $7::bigint ELSE NULL END,
+    failure_count = LEAST(control.failure_count + 1, 2147483647),
+    last_failure_code = $8,
+    updated_at_ms = GREATEST(control.updated_at_ms, observation.observed_at_ms)
+FROM observation
+WHERE control.attempt_id = $1
+  AND control.state = 'claimed'
+  AND control.claim_owner = $2
+  AND control.claim_token = $3
+  AND control.claim_version = $4
+  AND control.claim_expires_at_ms = $5::bigint
+  AND control.claim_expires_at_ms > observation.observed_at_ms
+RETURNING control.attempt_id AS "attemptId"
+`.trim();
+
+const READ_TOMBSTONE_STATE_SQL = `
+SELECT project_id AS "projectId", run_id AS "runId",
+  attempt_id AS "attemptId", log_artifact_id AS "logArtifactId",
+  executor_type AS "executorType", finished_at_ms AS "finishedAtMs",
+  eligible_at_ms AS "eligibleAtMs", retired_at_ms AS "retiredAtMs",
+  disposition, byte_length AS "byteLength", truncated,
+  maximum_bytes AS "maximumBytes",
+  truncation_observed_at_ms AS "truncationObservedAtMs",
+  record_digest AS "recordDigest"
+FROM "ql3"."run_attempt_log_artifact_tombstones"
+WHERE log_artifact_id = $1
+`.trim();
+
+const FAILURE_CODES = new Set<ClusterRunAttemptLogRetentionFailureCode>([
+  'artifact_unavailable',
+  'artifact_integrity_mismatch',
+  'retirement_record_unavailable',
+]);
+
+function integer(
+  name: string,
+  value: unknown,
+  minimum: number,
+  maximum = Number.MAX_SAFE_INTEGER,
+): number {
+  const converted =
+    typeof value === 'string' && /^\d+$/.test(value) ? Number(value) : value;
+  if (
+    typeof converted !== 'number' ||
+    !Number.isSafeInteger(converted) ||
+    converted < minimum ||
+    converted > maximum
+  ) {
+    throw new TypeError(`${name} is invalid`);
+  }
+  return converted;
+}
+
+function identifier(name: string, value: unknown, maximum = 128): string {
+  if (
+    typeof value !== 'string' ||
+    value.length > maximum ||
+    !/^[A-Za-z0-9][A-Za-z0-9._:-]*$/.test(value)
+  ) {
+    throw new TypeError(`${name} is invalid`);
+  }
+  return value;
+}
+
+function optionalInteger(
+  name: string,
+  value: unknown,
+  minimum: number,
+): number | undefined {
+  return value === null ? undefined : integer(name, value, minimum);
+}
+
+function readIdentity(
+  value: Readonly<RunAttemptLogReadIdentity>,
+): Readonly<RunAttemptLogReadIdentity> {
+  const normalized = Object.freeze({
+    projectId: identifier('Cluster log retention projectId', value?.projectId),
+    runId: identifier('Cluster log retention runId', value?.runId),
+    attemptId: identifier('Cluster log retention attemptId', value?.attemptId),
+    logArtifactId: identifier(
+      'Cluster log retention logArtifactId',
+      value?.logArtifactId,
+      36,
+    ),
+  });
+  if (
+    !value ||
+    typeof value !== 'object' ||
+    Array.isArray(value) ||
+    Object.keys(value).sort().join(',') !==
+      'attemptId,logArtifactId,projectId,runId' ||
+    !/^wlog-[a-f0-9]{30}$/.test(normalized.logArtifactId)
+  ) {
+    throw new TypeError('Cluster Run Attempt log retention identity is invalid');
+  }
+  return normalized;
+}
+
+function tombstoneFromRow(
+  row: Row,
+): Readonly<RunAttemptLogRetirementRecord> {
+  const truncated = row.truncated;
+  if (
+    truncated !== 'true' &&
+    truncated !== 'false' &&
+    truncated !== 'unknown'
+  ) {
+    throw new TypeError('PostgreSQL retention tombstone is invalid');
+  }
+  return normalizeRunAttemptLogRetirementRecord({
+    schema: 'qinglong/run-attempt-log-retirement@v1',
+    projectId: identifier('PostgreSQL tombstone projectId', row.projectId),
+    runId: identifier('PostgreSQL tombstone runId', row.runId),
+    attemptId: identifier('PostgreSQL tombstone attemptId', row.attemptId),
+    logArtifactId: identifier(
+      'PostgreSQL tombstone logArtifactId',
+      row.logArtifactId,
+      36,
+    ),
+    executorType: row.executorType as 'remote_worker',
+    finishedAtMs: integer(
+      'PostgreSQL tombstone finishedAtMs',
+      row.finishedAtMs,
+      0,
+    ),
+    eligibleAtMs: integer(
+      'PostgreSQL tombstone eligibleAtMs',
+      row.eligibleAtMs,
+      0,
+    ),
+    retiredAtMs: integer(
+      'PostgreSQL tombstone retiredAtMs',
+      row.retiredAtMs,
+      0,
+    ),
+    disposition: row.disposition as 'deleted' | 'already_absent',
+    byteLength: integer(
+      'PostgreSQL tombstone byteLength',
+      row.byteLength,
+      0,
+    ),
+    truncation:
+      truncated === 'unknown'
+        ? Object.freeze({ truncated: 'unknown' as const })
+        : Object.freeze({
+            truncated: truncated === 'true',
+            maximumBytes: optionalInteger(
+              'PostgreSQL tombstone maximumBytes',
+              row.maximumBytes,
+              1,
+            )!,
+            observedAtMs: optionalInteger(
+              'PostgreSQL tombstone truncationObservedAtMs',
+              row.truncationObservedAtMs,
+              0,
+            )!,
+          }),
+    recordDigest: identifier(
+      'PostgreSQL tombstone recordDigest',
+      row.recordDigest,
+      64,
+    ),
+  });
+}
+
+function failureCode(
+  value: unknown,
+): ClusterRunAttemptLogRetentionFailureCode {
+  if (!FAILURE_CODES.has(value as ClusterRunAttemptLogRetentionFailureCode)) {
+    throw new TypeError('Cluster Run Attempt log retention failure code is invalid');
+  }
+  return value as ClusterRunAttemptLogRetentionFailureCode;
+}
+
+function claimFromRow(row: Row): Readonly<ClusterRunAttemptLogRetentionClaim> {
+  const candidate = normalizeRunAttemptLogRetentionCandidate({
+    projectId: identifier('PostgreSQL retention projectId', row.projectId),
+    runId: identifier('PostgreSQL retention runId', row.runId),
+    attemptId: identifier('PostgreSQL retention attemptId', row.attemptId),
+    logArtifactId: identifier(
+      'PostgreSQL retention logArtifactId',
+      row.logArtifactId,
+    ),
+    executorType: row.executorType as 'remote_worker',
+    finishedAtMs: integer(
+      'PostgreSQL retention finishedAtMs',
+      row.finishedAtMs,
+      0,
+    ),
+  });
+  if (
+    candidate.executorType !== 'remote_worker' ||
+    !/^wlog-[a-f0-9]{30}$/.test(candidate.logArtifactId)
+  ) {
+    throw new TypeError('PostgreSQL retention candidate is invalid');
+  }
+  const observedAtMs = integer(
+    'PostgreSQL retention observedAtMs',
+    row.observedAtMs,
+    0,
+  );
+  const eligibleAtMs = integer(
+    'PostgreSQL retention eligibleAtMs',
+    row.eligibleAtMs,
+    candidate.finishedAtMs,
+  );
+  return Object.freeze({
+    candidate,
+    eligibleAtMs,
+    observedAtMs,
+    ownerId: identifier('PostgreSQL retention claimOwner', row.claimOwner),
+    token: identifier('PostgreSQL retention claimToken', row.claimToken, 64),
+    version: integer(
+      'PostgreSQL retention claimVersion',
+      row.claimVersion,
+      1,
+      2147483647,
+    ),
+    expiresAtMs: integer(
+      'PostgreSQL retention claimExpiresAtMs',
+      row.claimExpiresAtMs,
+      observedAtMs + 1,
+    ),
+    failureCount: integer(
+      'PostgreSQL retention failureCount',
+      row.failureCount,
+      0,
+      2147483647,
+    ),
+  });
+}
+
+function assertClaim(
+  value: Readonly<ClusterRunAttemptLogRetentionClaim>,
+): Readonly<ClusterRunAttemptLogRetentionClaim> {
+  const normalized = claimFromRow({
+    projectId: value?.candidate?.projectId,
+    runId: value?.candidate?.runId,
+    attemptId: value?.candidate?.attemptId,
+    logArtifactId: value?.candidate?.logArtifactId,
+    executorType: value?.candidate?.executorType,
+    finishedAtMs: value?.candidate?.finishedAtMs,
+    eligibleAtMs: value?.eligibleAtMs,
+    observedAtMs: value?.observedAtMs,
+    claimOwner: value?.ownerId,
+    claimToken: value?.token,
+    claimVersion: value?.version,
+    claimExpiresAtMs: value?.expiresAtMs,
+    failureCount: value?.failureCount,
+  });
+  if (normalized.expiresAtMs !== value.expiresAtMs) {
+    throw new TypeError('Cluster Run Attempt log retention claim is invalid');
+  }
+  return normalized;
+}
+
+function sameClaim(
+  actual: Readonly<ClusterRunAttemptLogRetentionClaim>,
+  expected: Readonly<ClusterRunAttemptLogRetentionClaim>,
+): boolean {
+  return (
+    actual.candidate.projectId === expected.candidate.projectId &&
+    actual.candidate.runId === expected.candidate.runId &&
+    actual.candidate.attemptId === expected.candidate.attemptId &&
+    actual.candidate.logArtifactId === expected.candidate.logArtifactId &&
+    actual.candidate.executorType === expected.candidate.executorType &&
+    actual.candidate.finishedAtMs === expected.candidate.finishedAtMs &&
+    actual.eligibleAtMs === expected.eligibleAtMs &&
+    actual.ownerId === expected.ownerId &&
+    actual.token === expected.token &&
+    actual.version === expected.version &&
+    actual.expiresAtMs === expected.expiresAtMs
+  );
+}
+
+function sameRecordClaim(
+  record: Readonly<RunAttemptLogRetirementRecord>,
+  claim: Readonly<ClusterRunAttemptLogRetentionClaim>,
+): boolean {
+  const candidate = claim.candidate;
+  return (
+    record.projectId === candidate.projectId &&
+    record.runId === candidate.runId &&
+    record.attemptId === candidate.attemptId &&
+    record.logArtifactId === candidate.logArtifactId &&
+    record.executorType === candidate.executorType &&
+    record.finishedAtMs === candidate.finishedAtMs &&
+    record.eligibleAtMs === claim.eligibleAtMs
+  );
+}
+
+function unavailable(error: unknown): RunAttemptLogRetentionUnavailableError {
+  return error instanceof RunAttemptLogRetentionUnavailableError
+    ? error
+    : new RunAttemptLogRetentionUnavailableError({ cause: error });
+}
+
+async function rollback(queryable: PostgresQueryable): Promise<void> {
+  try {
+    await queryable.query('ROLLBACK');
+  } catch {
+    // Preserve the originating authority failure.
+  }
+}
+
+export class PostgresRunAttemptLogRetentionClaimRepository
+  implements ClusterRunAttemptLogRetentionClaimRepository
+{
+  constructor(
+    private readonly pool: PostgresPool,
+    private readonly createToken: () => string = randomUUID,
+  ) {
+    if (
+      !pool ||
+      typeof pool.query !== 'function' ||
+      typeof pool.connect !== 'function' ||
+      typeof createToken !== 'function'
+    ) {
+      throw new TypeError(
+        'PostgreSQL Run Attempt log retention repository is invalid',
+      );
+    }
+  }
+
+  async inspect(
+    rawIdentity: Readonly<RunAttemptLogReadIdentity>,
+  ): Promise<RunAttemptLogRetentionState> {
+    const expected = readIdentity(rawIdentity);
+    try {
+      const result = await this.pool.query<Row>(READ_TOMBSTONE_STATE_SQL, [
+        expected.logArtifactId,
+      ]);
+      if (result.rows.length === 0) {
+        return Object.freeze({ status: 'active' as const });
+      }
+      if (result.rows.length !== 1) {
+        throw new TypeError('PostgreSQL retention tombstone is not unique');
+      }
+      const record = tombstoneFromRow(result.rows[0]!);
+      if (
+        record.projectId !== expected.projectId ||
+        record.runId !== expected.runId ||
+        record.attemptId !== expected.attemptId ||
+        record.logArtifactId !== expected.logArtifactId
+      ) {
+        throw new TypeError('PostgreSQL retention tombstone identity changed');
+      }
+      return Object.freeze({ status: 'retired' as const, record });
+    } catch (error) {
+      throw unavailable(error);
+    }
+  }
+
+  async claim(options: Readonly<{
+    ownerId: string;
+    retentionMs: number;
+    limit: number;
+    leaseMs: number;
+  }>): Promise<Readonly<ClusterRunAttemptLogRetentionClaimPage>> {
+    const ownerId = identifier(
+      'Cluster Run Attempt log retention ownerId',
+      options?.ownerId,
+    );
+    const retentionMs = integer(
+      'Cluster Run Attempt log retention duration',
+      options?.retentionMs,
+      MIN_RUN_ATTEMPT_LOG_RETENTION_MS,
+      MAX_RUN_ATTEMPT_LOG_RETENTION_MS,
+    );
+    const limit = integer(
+      'Cluster Run Attempt log retention claim limit',
+      options?.limit,
+      1,
+      MAX_CLUSTER_RUN_ATTEMPT_LOG_RETENTION_CLAIMS,
+    );
+    const leaseMs = integer(
+      'Cluster Run Attempt log retention lease',
+      options?.leaseMs,
+      MIN_CLUSTER_RUN_ATTEMPT_LOG_RETENTION_LEASE_MS,
+      MAX_CLUSTER_RUN_ATTEMPT_LOG_RETENTION_LEASE_MS,
+    );
+    const token = identifier(
+      'Cluster Run Attempt log retention generated token',
+      this.createToken(),
+      64,
+    );
+    if (token.length < 16) {
+      throw new TypeError(
+        'Cluster Run Attempt log retention generated token is invalid',
+      );
+    }
+    const client = await this.pool.connect().catch((error: unknown) => {
+      throw unavailable(error);
+    });
+    let transactionOpen = false;
+    try {
+      await client.query('BEGIN ISOLATION LEVEL READ COMMITTED');
+      transactionOpen = true;
+      await client.query("SET LOCAL statement_timeout = '5000ms'");
+      await client.query("SET LOCAL lock_timeout = '1000ms'");
+      const result = await client.query<Row>(CLAIM_SQL, [
+        retentionMs,
+        limit,
+        ownerId,
+        token,
+        leaseMs,
+      ]);
+      const claims = result.rows.map(claimFromRow);
+      if (claims.length > limit) {
+        throw new TypeError('PostgreSQL retention claim bound was violated');
+      }
+      await client.query('COMMIT');
+      transactionOpen = false;
+      return Object.freeze({
+        claims: Object.freeze(claims),
+        hasMore: claims.length === limit,
+      });
+    } catch (error) {
+      if (transactionOpen) await rollback(client);
+      throw unavailable(error);
+    } finally {
+      client.release();
+    }
+  }
+
+  async settle(
+    rawClaim: Readonly<ClusterRunAttemptLogRetentionClaim>,
+    settlement: Readonly<ClusterRunAttemptLogRetentionSettlement>,
+  ): Promise<'settled' | 'fenced'> {
+    const claim = assertClaim(rawClaim);
+    if (!settlement || typeof settlement !== 'object') {
+      throw new TypeError(
+        'Cluster Run Attempt log retention settlement is invalid',
+      );
+    }
+    if (settlement.status === 'retired') {
+      return this.recordRetirement(claim, settlement.record);
+    }
+    if (settlement.status !== 'retry' && settlement.status !== 'manual') {
+      throw new TypeError(
+        'Cluster Run Attempt log retention settlement is invalid',
+      );
+    }
+    const delayMs =
+      settlement.status === 'retry'
+        ? integer(
+            'Cluster Run Attempt log retention retry delay',
+            settlement.delayMs,
+            0,
+            MAX_CLUSTER_RUN_ATTEMPT_LOG_RETENTION_RETRY_DELAY_MS,
+          )
+        : 0;
+    const code = failureCode(settlement.failureCode);
+    try {
+      const result = await this.pool.query<Row>(SET_FAILURE_SQL, [
+        claim.candidate.attemptId,
+        claim.ownerId,
+        claim.token,
+        claim.version,
+        claim.expiresAtMs,
+        settlement.status,
+        delayMs,
+        code,
+      ]);
+      return result.rows.length === 1 ? 'settled' : 'fenced';
+    } catch (error) {
+      throw unavailable(error);
+    }
+  }
+
+  private async recordRetirement(
+    claim: Readonly<ClusterRunAttemptLogRetentionClaim>,
+    rawRecord: Readonly<RunAttemptLogRetirementRecord>,
+  ): Promise<'settled' | 'fenced'> {
+    const record = normalizeRunAttemptLogRetirementRecord(rawRecord);
+    if (!sameRecordClaim(record, claim)) {
+      throw new TypeError(
+        'Cluster Run Attempt log retirement record does not match its claim',
+      );
+    }
+    const client = await this.pool.connect().catch((error: unknown) => {
+      throw unavailable(error);
+    });
+    let transactionOpen = false;
+    try {
+      await client.query('BEGIN ISOLATION LEVEL READ COMMITTED');
+      transactionOpen = true;
+      await client.query("SET LOCAL statement_timeout = '5000ms'");
+      await client.query("SET LOCAL lock_timeout = '1000ms'");
+      const locked = await client.query<Row>(LOCK_CLAIM_SQL, [
+        claim.candidate.attemptId,
+        claim.ownerId,
+        claim.token,
+        claim.version,
+        claim.expiresAtMs,
+      ]);
+      if (locked.rows.length !== 1) {
+        await client.query('COMMIT');
+        transactionOpen = false;
+        return 'fenced';
+      }
+      if (!sameClaim(claimFromRow(locked.rows[0]!), claim)) {
+        throw new TypeError('PostgreSQL retention claim authority changed');
+      }
+      const truncation = record.truncation;
+      const inserted = await client.query<Row>(INSERT_TOMBSTONE_SQL, [
+        record.logArtifactId,
+        record.projectId,
+        record.runId,
+        record.attemptId,
+        record.executorType,
+        record.finishedAtMs,
+        record.eligibleAtMs,
+        record.retiredAtMs,
+        record.disposition,
+        record.byteLength,
+        String(truncation.truncated),
+        truncation.maximumBytes ?? null,
+        truncation.observedAtMs ?? null,
+        record.recordDigest,
+      ]);
+      if (inserted.rows.length === 0) {
+        const existing = await client.query<Row>(READ_TOMBSTONE_SQL, [
+          record.attemptId,
+          record.logArtifactId,
+        ]);
+        if (
+          existing.rows.length !== 1 ||
+          existing.rows[0]?.recordDigest !== record.recordDigest
+        ) {
+          throw new TypeError('PostgreSQL retention tombstone conflicts');
+        }
+      }
+      const removed = await client.query<Row>(DELETE_CONTROL_SQL, [
+        claim.candidate.attemptId,
+        claim.ownerId,
+        claim.token,
+        claim.version,
+        claim.expiresAtMs,
+      ]);
+      if (removed.rowCount !== 1) {
+        throw new TypeError('PostgreSQL retention claim fence was lost');
+      }
+      await client.query('COMMIT');
+      transactionOpen = false;
+      return 'settled';
+    } catch (error) {
+      if (transactionOpen) await rollback(client);
+      throw unavailable(error);
+    } finally {
+      client.release();
+    }
+  }
+}

packages/ql3-cluster-postgres/src/schema/schema.ts

@@ -4474,6 +4474,162 @@ export const runAttempts = ql3Schema.table(
     index('ql3_run_attempts_lease_idx')
       .on(table.leaseExpiresAtMs, table.id)
       .where(sql`${table.leaseExpiresAtMs} is not null`),
+    index('ql3_run_log_retention_candidate_idx')
+      .on(table.finishedAtMs, table.id)
+      .where(
+        sql`${table.executorType} = 'remote_worker' and ${table.logArtifactId} is not null and ${table.status} in ('succeeded', 'failed', 'cancelled', 'timed_out')`,
+      ),
+  ],
+);
+
+export const runAttemptLogRetentionControls = ql3Schema.table(
+  'run_attempt_log_retention_controls',
+  {
+    attemptId: varchar('attempt_id', { length: 36 }).primaryKey(),
+    projectId: varchar('project_id', { length: 128 }).notNull(),
+    runId: varchar('run_id', { length: 36 }).notNull(),
+    logArtifactId: varchar('log_artifact_id', { length: 36 }).notNull(),
+    executorType: varchar('executor_type', { length: 32 }).notNull(),
+    finishedAtMs: bigint('finished_at_ms', { mode: 'number' }).notNull(),
+    eligibleAtMs: bigint('eligible_at_ms', { mode: 'number' }).notNull(),
+    state: varchar('state', { length: 16 }).notNull(),
+    claimOwner: varchar('claim_owner', { length: 128 }),
+    claimToken: varchar('claim_token', { length: 64 }),
+    claimVersion: integer('claim_version').default(1).notNull(),
+    claimExpiresAtMs: bigint('claim_expires_at_ms', { mode: 'number' }),
+    nextClaimAtMs: bigint('next_claim_at_ms', { mode: 'number' }),
+    failureCount: integer('failure_count').default(0).notNull(),
+    lastFailureCode: varchar('last_failure_code', { length: 64 }),
+    createdAtMs: bigint('created_at_ms', { mode: 'number' }).notNull(),
+    updatedAtMs: bigint('updated_at_ms', { mode: 'number' }).notNull(),
+  },
+  (table) => [
+    uniqueIndex('ql3_run_log_retention_control_artifact_key').on(
+      table.logArtifactId,
+    ),
+    check(
+      'ql3_run_log_retention_control_identity_check',
+      sql`char_length(${table.projectId}) between 1 and 128 and char_length(${table.runId}) between 1 and 36 and char_length(${table.attemptId}) between 1 and 36 and ${table.logArtifactId} ~ '^wlog-[a-f0-9]{30}$' and ${table.executorType} = 'remote_worker'`,
+    ),
+    check(
+      'ql3_run_log_retention_control_time_check',
+      sql`${table.finishedAtMs} >= 0 and ${table.eligibleAtMs} >= ${table.finishedAtMs} and ${table.createdAtMs} >= 0 and ${table.updatedAtMs} >= ${table.createdAtMs}`,
+    ),
+    check(
+      'ql3_run_log_retention_control_state_check',
+      sql`${table.state} in ('claimed', 'retry', 'manual')`,
+    ),
+    check(
+      'ql3_run_log_retention_control_claim_owner_check',
+      sql`${table.claimOwner} is null or ${table.claimOwner} ~ '^[A-Za-z0-9][A-Za-z0-9._:-]{0,127}$'`,
+    ),
+    check(
+      'ql3_run_log_retention_control_claim_token_check',
+      sql`${table.claimToken} is null or ${table.claimToken} ~ '^[A-Za-z0-9][A-Za-z0-9._:-]{15,63}$'`,
+    ),
+    check(
+      'ql3_run_log_retention_control_claim_version_check',
+      sql`${table.claimVersion} between 1 and 2147483647`,
+    ),
+    check(
+      'ql3_run_log_retention_control_claim_expiry_check',
+      sql`${table.claimExpiresAtMs} is null or ${table.claimExpiresAtMs} >= 0`,
+    ),
+    check(
+      'ql3_run_log_retention_control_next_claim_check',
+      sql`${table.nextClaimAtMs} is null or ${table.nextClaimAtMs} >= 0`,
+    ),
+    check(
+      'ql3_run_log_retention_control_failure_count_check',
+      sql`${table.failureCount} between 0 and 2147483647`,
+    ),
+    check(
+      'ql3_run_log_retention_control_failure_code_check',
+      sql`${table.lastFailureCode} is null or ${table.lastFailureCode} in ('artifact_unavailable', 'artifact_integrity_mismatch', 'retirement_record_unavailable')`,
+    ),
+    check(
+      'ql3_run_log_retention_control_state_shape_check',
+      sql`(${table.state} = 'claimed' and ${table.claimOwner} is not null and ${table.claimToken} is not null and ${table.claimExpiresAtMs} is not null and ${table.nextClaimAtMs} is null) or (${table.state} = 'retry' and ${table.claimOwner} is null and ${table.claimToken} is null and ${table.claimExpiresAtMs} is null and ${table.nextClaimAtMs} is not null and ${table.lastFailureCode} is not null) or (${table.state} = 'manual' and ${table.claimOwner} is null and ${table.claimToken} is null and ${table.claimExpiresAtMs} is null and ${table.nextClaimAtMs} is null and ${table.lastFailureCode} is not null)`,
+    ),
+    foreignKey({
+      name: 'ql3_run_log_retention_control_attempt_fk',
+      columns: [table.attemptId],
+      foreignColumns: [runAttempts.id],
+    }).onDelete('cascade'),
+    foreignKey({
+      name: 'ql3_run_log_retention_control_run_fk',
+      columns: [table.runId],
+      foreignColumns: [runs.id],
+    }).onDelete('cascade'),
+    index('ql3_run_log_retention_retry_idx')
+      .on(table.nextClaimAtMs, table.finishedAtMs, table.attemptId)
+      .where(sql`${table.state} = 'retry'`),
+    index('ql3_run_log_retention_claim_expiry_idx')
+      .on(table.claimExpiresAtMs, table.finishedAtMs, table.attemptId)
+      .where(sql`${table.state} = 'claimed'`),
+  ],
+);
+
+export const runAttemptLogArtifactTombstones = ql3Schema.table(
+  'run_attempt_log_artifact_tombstones',
+  {
+    logArtifactId: varchar('log_artifact_id', { length: 36 }).primaryKey(),
+    projectId: varchar('project_id', { length: 128 }).notNull(),
+    runId: varchar('run_id', { length: 36 }).notNull(),
+    attemptId: varchar('attempt_id', { length: 36 }).notNull(),
+    executorType: varchar('executor_type', { length: 32 }).notNull(),
+    finishedAtMs: bigint('finished_at_ms', { mode: 'number' }).notNull(),
+    eligibleAtMs: bigint('eligible_at_ms', { mode: 'number' }).notNull(),
+    retiredAtMs: bigint('retired_at_ms', { mode: 'number' }).notNull(),
+    disposition: varchar('disposition', { length: 16 }).notNull(),
+    byteLength: bigint('byte_length', { mode: 'number' }).notNull(),
+    truncated: varchar('truncated', { length: 16 }).notNull(),
+    maximumBytes: bigint('maximum_bytes', { mode: 'number' }),
+    truncationObservedAtMs: bigint('truncation_observed_at_ms', {
+      mode: 'number',
+    }),
+    recordDigest: char('record_digest', { length: 64 }).notNull(),
+  },
+  (table) => [
+    uniqueIndex('ql3_run_log_tombstone_attempt_key').on(table.attemptId),
+    check(
+      'ql3_run_log_tombstone_identity_check',
+      sql`char_length(${table.projectId}) between 1 and 128 and char_length(${table.runId}) between 1 and 36 and char_length(${table.attemptId}) between 1 and 36 and ${table.logArtifactId} ~ '^wlog-[a-f0-9]{30}$' and ${table.executorType} = 'remote_worker'`,
+    ),
+    check(
+      'ql3_run_log_tombstone_time_check',
+      sql`${table.finishedAtMs} >= 0 and ${table.eligibleAtMs} >= ${table.finishedAtMs} and ${table.retiredAtMs} >= ${table.eligibleAtMs}`,
+    ),
+    check(
+      'ql3_run_log_tombstone_disposition_check',
+      sql`${table.disposition} in ('deleted', 'already_absent') and (${table.disposition} <> 'already_absent' or ${table.byteLength} = 0)`,
+    ),
+    check(
+      'ql3_run_log_tombstone_size_check',
+      sql`${table.byteLength} between 0 and 1073741824`,
+    ),
+    check(
+      'ql3_run_log_tombstone_truncation_check',
+      sql`(${table.truncated} = 'unknown' and ${table.maximumBytes} is null and ${table.truncationObservedAtMs} is null) or (${table.truncated} in ('true', 'false') and ${table.maximumBytes} >= 1 and ${table.truncationObservedAtMs} >= 0)`,
+    ),
+    check(
+      'ql3_run_log_tombstone_digest_check',
+      sql`${table.recordDigest} ~ '^[a-f0-9]{64}$'`,
+    ),
+    foreignKey({
+      name: 'ql3_run_log_tombstone_attempt_fk',
+      columns: [table.attemptId],
+      foreignColumns: [runAttempts.id],
+    }).onDelete('cascade'),
+    foreignKey({
+      name: 'ql3_run_log_tombstone_run_fk',
+      columns: [table.runId],
+      foreignColumns: [runs.id],
+    }).onDelete('cascade'),
+    index('ql3_run_log_tombstone_retired_idx').on(
+      table.retiredAtMs,
+      table.attemptId,
+    ),
   ],
 );
 
@@ -5700,6 +5856,8 @@ export const ql3PostgresTables = [
   toolExecutionResultRekeyHeads,
   toolResultKeyRetirementReceipts,
   runAttempts,
+  runAttemptLogRetentionControls,
+  runAttemptLogArtifactTombstones,
   workerSessions,
   runDispatchLeases,
   workerCredentials,

packages/ql3-cluster-postgres/src/schema/schemaContract.ts

@@ -15,12 +15,13 @@ export interface PostgresSchemaContractFunction {
 export interface PostgresSchemaContract {
   readonly schema: 'ql3';
   readonly contractName: 'control-core';
-  readonly contractVersion: 53;
+  readonly contractVersion: 54;
-  readonly migrationId: 'pg-0054-approval-management-boundary';
+  readonly migrationId: 'pg-0055-run-attempt-log-retention';
   readonly minimumServerMajor: 16;
   readonly maximumServerMajor: 18;
   readonly capabilities: Readonly<{
     run_core: 1;
+    run_attempt_log_retention: 1;
     run_dispatch_lease: 1;
     run_retry_policy: 1;
     project_policy: 1;
@@ -100,8 +101,8 @@ export const postgresqlControlSchemaContract: PostgresSchemaContract =
   Object.freeze({
     schema: 'ql3',
     contractName: 'control-core',
-    contractVersion: 53,
+    contractVersion: 54,
-    migrationId: 'pg-0054-approval-management-boundary',
+    migrationId: 'pg-0055-run-attempt-log-retention',
     minimumServerMajor: 16,
     maximumServerMajor: 18,
     capabilities: Object.freeze({
@@ -141,6 +142,7 @@ export const postgresqlControlSchemaContract: PostgresSchemaContract =
       project_policy: 1,
       project_tool_definition_snapshot: 1,
       run_core: 1,
+      run_attempt_log_retention: 1,
       run_dispatch_lease: 1,
       run_retry_policy: 1,
       security_audit: 1,
@@ -1166,6 +1168,41 @@ export const postgresqlControlSchemaContract: PostgresSchemaContract =
         'error_code',
         'error_summary',
       ]),
+      table('run_attempt_log_retention_controls', [
+        'attempt_id',
+        'project_id',
+        'run_id',
+        'log_artifact_id',
+        'executor_type',
+        'finished_at_ms',
+        'eligible_at_ms',
+        'state',
+        'claim_owner',
+        'claim_token',
+        'claim_version',
+        'claim_expires_at_ms',
+        'next_claim_at_ms',
+        'failure_count',
+        'last_failure_code',
+        'created_at_ms',
+        'updated_at_ms',
+      ]),
+      table('run_attempt_log_artifact_tombstones', [
+        'log_artifact_id',
+        'project_id',
+        'run_id',
+        'attempt_id',
+        'executor_type',
+        'finished_at_ms',
+        'eligible_at_ms',
+        'retired_at_ms',
+        'disposition',
+        'byte_length',
+        'truncated',
+        'maximum_bytes',
+        'truncation_observed_at_ms',
+        'record_digest',
+      ]),
       table('worker_sessions', [
         'worker_id',
         'session_id',
@@ -1594,6 +1631,14 @@ export const postgresqlControlSchemaContract: PostgresSchemaContract =
       'ql3_run_attempts_dispatch_candidates_idx',
       'ql3_run_attempts_recovery_idx',
       'ql3_run_attempts_lease_idx',
+      'run_attempt_log_retention_controls_pkey',
+      'ql3_run_log_retention_control_artifact_key',
+      'ql3_run_log_retention_retry_idx',
+      'ql3_run_log_retention_claim_expiry_idx',
+      'run_attempt_log_artifact_tombstones_pkey',
+      'ql3_run_log_tombstone_attempt_key',
+      'ql3_run_log_tombstone_retired_idx',
+      'ql3_run_log_retention_candidate_idx',
       'worker_sessions_pkey',
       'ql3_worker_sessions_available_idx',
       'run_dispatch_leases_pkey',
@@ -1913,6 +1958,23 @@ export const postgresqlControlSchemaContract: PostgresSchemaContract =
       'ql3_run_attempts_created_at_check',
       'ql3_run_attempts_started_at_check',
       'ql3_run_attempts_finished_at_check',
+      'ql3_run_log_retention_control_identity_check',
+      'ql3_run_log_retention_control_time_check',
+      'ql3_run_log_retention_control_state_check',
+      'ql3_run_log_retention_control_claim_owner_check',
+      'ql3_run_log_retention_control_claim_token_check',
+      'ql3_run_log_retention_control_claim_version_check',
+      'ql3_run_log_retention_control_claim_expiry_check',
+      'ql3_run_log_retention_control_next_claim_check',
+      'ql3_run_log_retention_control_failure_count_check',
+      'ql3_run_log_retention_control_failure_code_check',
+      'ql3_run_log_retention_control_state_shape_check',
+      'ql3_run_log_tombstone_identity_check',
+      'ql3_run_log_tombstone_time_check',
+      'ql3_run_log_tombstone_disposition_check',
+      'ql3_run_log_tombstone_size_check',
+      'ql3_run_log_tombstone_truncation_check',
+      'ql3_run_log_tombstone_digest_check',
       'ql3_worker_sessions_worker_id_check',
       'ql3_worker_sessions_session_id_check',
       'ql3_worker_sessions_generation_check',
@@ -2211,6 +2273,10 @@ export const postgresqlControlSchemaContract: PostgresSchemaContract =
       'ql3_result_retirement_catalog_fk',
       'ql3_run_attempts_run_fk',
       'ql3_run_attempts_step_run_fk',
+      'ql3_run_log_retention_control_attempt_fk',
+      'ql3_run_log_retention_control_run_fk',
+      'ql3_run_log_tombstone_attempt_fk',
+      'ql3_run_log_tombstone_run_fk',
       'ql3_run_dispatch_leases_attempt_fk',
       'ql3_run_dispatch_leases_run_fk',
       'ql3_run_dispatch_leases_worker_fk',

packages/ql3-cluster-postgres/src/schema/schemaReadiness.ts

@@ -539,6 +539,18 @@ const REQUIRED_RUNTIME_PRIVILEGES = Object.freeze({
     update: true,
     delete: false,
   }),
+  run_attempt_log_retention_controls: Object.freeze({
+    select: true,
+    insert: true,
+    update: true,
+    delete: true,
+  }),
+  run_attempt_log_artifact_tombstones: Object.freeze({
+    select: true,
+    insert: true,
+    update: false,
+    delete: false,
+  }),
   worker_sessions: Object.freeze({
     select: true,
     insert: true,
@@ -1034,6 +1046,18 @@ const REQUIRED_ADMIN_PRIVILEGES = Object.freeze({
     update: false,
     delete: false,
   }),
+  run_attempt_log_retention_controls: Object.freeze({
+    select: false,
+    insert: false,
+    update: false,
+    delete: false,
+  }),
+  run_attempt_log_artifact_tombstones: Object.freeze({
+    select: false,
+    insert: false,
+    update: false,
+    delete: false,
+  }),
   worker_sessions: Object.freeze({
     select: false,
     insert: false,

packages/ql3-cluster-postgres/test/postgresqlMigrationDefinitions.test.cjs

@@ -103,6 +103,7 @@ test('defines the immutable PostgreSQL capability and Run core stream', async ()
       'pg-0052-automation-management-identity-keyset-ledger',
       'pg-0053-plugin-package-workflow-run-list-index',
       'pg-0054-approval-management-boundary',
+      'pg-0055-run-attempt-log-retention',
     ],
   );
   for (const migration of postgresqlMainMigrationStream.migrations) {
@@ -508,6 +509,11 @@ test('freezes every published PostgreSQL migration checksum', () => {
       checksum:
         '5e3e6b222269f095e0d7a985fdeb0ea154510e59dfe15873192af8c8d603fca3',
     },
+    {
+      id: 'pg-0055-run-attempt-log-retention',
+      checksum:
+        'c775c65ec03ae3a1606f899064d2d38fa63fd136ce52cbd1b1172c3a51e6bf30',
+    },
   ];
   assert.deepEqual(
     postgresqlMainMigrationStream.migrations.map(({ id, checksum }) => ({
@@ -1875,3 +1881,26 @@ test('advances capability v53 with isolated human Approval management authority'
     /migration_id = 'pg-0053-plugin-package-workflow-run-list-index'/,
   );
 });
+
+test('advances capability v54 with durable Cluster log retention authority', async () => {
+  const migration = migrationById('pg-0055-run-attempt-log-retention');
+  const statements = [];
+  await migration.up({
+    async query(statement) {
+      statements.push(statement);
+      return { rows: [] };
+    },
+  });
+  const sql = statements.join('\n');
+  assert.match(sql, /run_attempt_log_retention_controls/);
+  assert.match(sql, /run_attempt_log_artifact_tombstones/);
+  assert.match(sql, /FOR UPDATE|SKIP LOCKED|claim_expires_at_ms/);
+  assert.match(sql, /TO ql3_runtime/);
+  assert.match(sql, /contract_version = 54/);
+  assert.match(sql, /"run_attempt_log_retention":1/);
+  assert.match(sql, /contract_version = 53/);
+  assert.match(
+    sql,
+    /migration_id = 'pg-0054-approval-management-boundary'/,
+  );
+});

packages/ql3-cluster-postgres/test/postgresqlSchemaReadiness.test.cjs

@@ -58,6 +58,8 @@ function validPrivileges() {
     tool_invocation_input_artifacts: [true, true, false, false],
     tool_invocation_preview_artifacts: [true, true, false, false],
     run_attempts: [true, true, true, false],
+    run_attempt_log_retention_controls: [true, true, true, true],
+    run_attempt_log_artifact_tombstones: [true, true, false, false],
     worker_sessions: [true, true, true, false],
     run_dispatch_leases: [true, true, true, false],
     worker_credentials: [false, false, false, false],
@@ -178,6 +180,8 @@ function validAdminPrivileges() {
     tool_invocation_input_artifacts: [false, false, false, false],
     tool_invocation_preview_artifacts: [false, false, false, false],
     run_attempts: [false, false, false, false],
+    run_attempt_log_retention_controls: [false, false, false, false],
+    run_attempt_log_artifact_tombstones: [false, false, false, false],
     worker_sessions: [false, false, false, false],
     run_dispatch_leases: [false, false, false, false],
     worker_credentials: [true, true, true, false],
@@ -692,7 +696,7 @@ test('accepts the exact PostgreSQL control schema and least-privilege runtime ro
     serverMajor: 16,
     currentUser: 'ql3_runtime',
     contractName: 'control-core',
-    contractVersion: 53,
+    contractVersion: 54,
     migrationIds: [
       'pg-0001-schema-capability',
       'pg-0002-run-core',
@@ -748,6 +752,7 @@ test('accepts the exact PostgreSQL control schema and least-privilege runtime ro
       'pg-0052-automation-management-identity-keyset-ledger',
       'pg-0053-plugin-package-workflow-run-list-index',
       'pg-0054-approval-management-boundary',
+      'pg-0055-run-attempt-log-retention',
     ],
   });
 });
@@ -778,10 +783,10 @@ test('accepts the exact schema and isolated least-privilege admin role', async (
     }),
   );
   assert.equal(report.currentUser, 'ql3_admin');
-  assert.equal(report.contractVersion, 53);
+  assert.equal(report.contractVersion, 54);
   assert.equal(
     report.migrationIds.at(-1),
-    'pg-0054-approval-management-boundary',
+    'pg-0055-run-attempt-log-retention',
   );
 });
 
@@ -794,10 +799,10 @@ test('accepts the isolated least-privilege automation manager role', async () =>
     }),
   );
   assert.equal(report.currentUser, 'ql3_automation_manager');
-  assert.equal(report.contractVersion, 53);
+  assert.equal(report.contractVersion, 54);
   assert.equal(
     report.migrationIds.at(-1),
-    'pg-0054-approval-management-boundary',
+    'pg-0055-run-attempt-log-retention',
   );
 
   const widened = automationManagerPrivileges();
@@ -826,10 +831,10 @@ test('accepts the isolated least-privilege human Approval manager role', async (
     }),
   );
   assert.equal(report.currentUser, 'ql3_approval_manager');
-  assert.equal(report.contractVersion, 53);
+  assert.equal(report.contractVersion, 54);
   assert.equal(
     report.migrationIds.at(-1),
-    'pg-0054-approval-management-boundary',
+    'pg-0055-run-attempt-log-retention',
   );
 
   const widened = approvalManagerPrivileges();
@@ -936,10 +941,10 @@ test('accepts the exact schema and isolated Worker ingress role', async () => {
     }),
   );
   assert.equal(report.currentUser, 'ql3_worker_ingress');
-  assert.equal(report.contractVersion, 53);
+  assert.equal(report.contractVersion, 54);
   assert.equal(
     report.migrationIds.at(-1),
-    'pg-0054-approval-management-boundary',
+    'pg-0055-run-attempt-log-retention',
   );
 });
 

packages/ql3-cluster-postgres/test/runAttemptLogRetentionClaimRepository.test.cjs

@@ -0,0 +1,330 @@
+'use strict';
+
+const assert = require('node:assert/strict');
+const { test } = require('node:test');
+
+const {
+  createRunAttemptLogRetirementRecord,
+  RunAttemptLogRetentionUnavailableError,
+} = require('@qinglong/runtime-core/run-attempt-log-retention');
+const {
+  PostgresRunAttemptLogRetentionClaimRepository,
+} = require('../dist/entrypoints/runtime');
+
+const TOKEN = '00000000-0000-4000-8000-000000000055';
+const ARTIFACT_ID = `wlog-${'a'.repeat(30)}`;
+
+function claimRow(overrides = {}) {
+  return {
+    projectId: 'project-1',
+    runId: 'run-1',
+    attemptId: 'attempt-1',
+    logArtifactId: ARTIFACT_ID,
+    executorType: 'remote_worker',
+    finishedAtMs: '1000',
+    eligibleAtMs: '61000',
+    observedAtMs: '70000',
+    claimOwner: 'replica-a',
+    claimToken: TOKEN,
+    claimVersion: 1,
+    claimExpiresAtMs: '100000',
+    failureCount: 0,
+    ...overrides,
+  };
+}
+
+function claim() {
+  return Object.freeze({
+    candidate: Object.freeze({
+      projectId: 'project-1',
+      runId: 'run-1',
+      attemptId: 'attempt-1',
+      logArtifactId: ARTIFACT_ID,
+      executorType: 'remote_worker',
+      finishedAtMs: 1000,
+    }),
+    eligibleAtMs: 61000,
+    observedAtMs: 70000,
+    ownerId: 'replica-a',
+    token: TOKEN,
+    version: 1,
+    expiresAtMs: 100000,
+    failureCount: 0,
+  });
+}
+
+test('claims one bounded remote log page under a short database lease', async () => {
+  const calls = [];
+  let released = false;
+  const repository = new PostgresRunAttemptLogRetentionClaimRepository(
+    {
+      async connect() {
+        return {
+          async query(text, values = []) {
+            calls.push({ text, values });
+            if (text.includes('FOR UPDATE OF attempt SKIP LOCKED')) {
+              return { rows: [claimRow()], rowCount: 1 };
+            }
+            return { rows: [], rowCount: 0 };
+          },
+          release() {
+            released = true;
+          },
+        };
+      },
+      async query() {
+        throw new Error('pool query not expected');
+      },
+    },
+    () => TOKEN,
+  );
+
+  const page = await repository.claim({
+    ownerId: 'replica-a',
+    retentionMs: 60000,
+    limit: 4,
+    leaseMs: 30000,
+  });
+
+  assert.deepEqual(page, { claims: [claim()], hasMore: false });
+  assert.deepEqual(
+    calls.map(({ text }) => text.split('\n', 1)[0]),
+    [
+      'BEGIN ISOLATION LEVEL READ COMMITTED',
+      "SET LOCAL statement_timeout = '5000ms'",
+      "SET LOCAL lock_timeout = '1000ms'",
+      'WITH observation AS (',
+      'COMMIT',
+    ],
+  );
+  assert.deepEqual(calls[3].values, [
+    60000,
+    4,
+    'replica-a',
+    TOKEN,
+    30000,
+  ]);
+  assert.match(calls[3].text, /ON CONFLICT \(attempt_id\) DO UPDATE/);
+  assert.match(calls[3].text, /claim_expires_at_ms <= EXCLUDED\.updated_at_ms/);
+  assert.equal(released, true);
+});
+
+test('fences retry settlement by owner token version and database expiry', async () => {
+  const calls = [];
+  const repository = new PostgresRunAttemptLogRetentionClaimRepository({
+    async connect() {
+      throw new Error('not expected');
+    },
+    async query(text, values) {
+      calls.push({ text, values });
+      return { rows: [{ attemptId: 'attempt-1' }], rowCount: 1 };
+    },
+  });
+
+  assert.equal(
+    await repository.settle(claim(), {
+      status: 'retry',
+      delayMs: 2500,
+      failureCode: 'artifact_unavailable',
+    }),
+    'settled',
+  );
+  assert.deepEqual(calls[0].values, [
+    'attempt-1',
+    'replica-a',
+    TOKEN,
+    1,
+    100000,
+    'retry',
+    2500,
+    'artifact_unavailable',
+  ]);
+  assert.match(calls[0].text, /claim_expires_at_ms > observation\.observed_at_ms/);
+});
+
+test('reads an exact durable tombstone for the profile-aware log route', async () => {
+  const record = createRunAttemptLogRetirementRecord({
+    ...claim().candidate,
+    eligibleAtMs: 61000,
+    retiredAtMs: 80000,
+    disposition: 'already_absent',
+    byteLength: 0,
+    truncation: { truncated: 'unknown' },
+  });
+  const rows = [
+    {
+      ...record,
+      finishedAtMs: String(record.finishedAtMs),
+      eligibleAtMs: String(record.eligibleAtMs),
+      retiredAtMs: String(record.retiredAtMs),
+      byteLength: String(record.byteLength),
+      truncated: 'unknown',
+      maximumBytes: null,
+      truncationObservedAtMs: null,
+    },
+  ];
+  const repository = new PostgresRunAttemptLogRetentionClaimRepository({
+    async connect() {
+      throw new Error('not expected');
+    },
+    async query(text, values) {
+      assert.match(text, /artifact_tombstones/);
+      assert.deepEqual(values, [ARTIFACT_ID]);
+      return { rows: rows.splice(0), rowCount: 1 };
+    },
+  });
+  const identity = {
+    projectId: 'project-1',
+    runId: 'run-1',
+    attemptId: 'attempt-1',
+    logArtifactId: ARTIFACT_ID,
+  };
+
+  assert.deepEqual(await repository.inspect(identity), {
+    status: 'retired',
+    record,
+  });
+  assert.deepEqual(await repository.inspect(identity), { status: 'active' });
+});
+
+test('records the exact tombstone and removes its claim atomically', async () => {
+  const record = createRunAttemptLogRetirementRecord({
+    ...claim().candidate,
+    eligibleAtMs: 61000,
+    retiredAtMs: 80000,
+    disposition: 'deleted',
+    byteLength: 11,
+    truncation: {
+      truncated: false,
+      maximumBytes: 1048576,
+      observedAtMs: 80000,
+    },
+  });
+  const calls = [];
+  let released = false;
+  const repository = new PostgresRunAttemptLogRetentionClaimRepository({
+    async connect() {
+      return {
+        async query(text, values = []) {
+          calls.push({ text, values });
+          if (text.includes('FOR UPDATE OF control')) {
+            return {
+              rows: [claimRow({ observedAtMs: '80000' })],
+              rowCount: 1,
+            };
+          }
+          if (text.startsWith('INSERT INTO "ql3"."run_attempt_log_artifact_tombstones"')) {
+            return {
+              rows: [{ recordDigest: record.recordDigest }],
+              rowCount: 1,
+            };
+          }
+          if (text.startsWith('DELETE FROM "ql3"."run_attempt_log_retention_controls"')) {
+            return { rows: [], rowCount: 1 };
+          }
+          return { rows: [], rowCount: 0 };
+        },
+        release() {
+          released = true;
+        },
+      };
+    },
+    async query() {
+      throw new Error('not expected');
+    },
+  });
+
+  assert.equal(
+    await repository.settle(claim(), { status: 'retired', record }),
+    'settled',
+  );
+  assert.deepEqual(
+    calls.map(({ text }) => text.split('\n', 1)[0]),
+    [
+      'BEGIN ISOLATION LEVEL READ COMMITTED',
+      "SET LOCAL statement_timeout = '5000ms'",
+      "SET LOCAL lock_timeout = '1000ms'",
+      'WITH observation AS (',
+      'INSERT INTO "ql3"."run_attempt_log_artifact_tombstones" (',
+      'DELETE FROM "ql3"."run_attempt_log_retention_controls"',
+      'COMMIT',
+    ],
+  );
+  assert.equal(calls[4].values.at(-1), record.recordDigest);
+  assert.equal(released, true);
+});
+
+test('returns fenced without writing when the durable lease changed', async () => {
+  const calls = [];
+  const repository = new PostgresRunAttemptLogRetentionClaimRepository({
+    async connect() {
+      return {
+        async query(text) {
+          calls.push(text);
+          if (text.includes('FOR UPDATE OF control')) {
+            return { rows: [], rowCount: 0 };
+          }
+          return { rows: [], rowCount: 0 };
+        },
+        release() {},
+      };
+    },
+    async query() {
+      throw new Error('not expected');
+    },
+  });
+  const record = createRunAttemptLogRetirementRecord({
+    ...claim().candidate,
+    eligibleAtMs: 61000,
+    retiredAtMs: 80000,
+    disposition: 'already_absent',
+    byteLength: 0,
+    truncation: { truncated: 'unknown' },
+  });
+
+  assert.equal(
+    await repository.settle(claim(), { status: 'retired', record }),
+    'fenced',
+  );
+  assert.equal(
+    calls.some((text) => text.includes('artifact_tombstones" (')),
+    false,
+  );
+});
+
+test('rolls back and wraps claim failures without leaking the client', async () => {
+  const calls = [];
+  let released = false;
+  const repository = new PostgresRunAttemptLogRetentionClaimRepository(
+    {
+      async connect() {
+        return {
+          async query(text) {
+            calls.push(text);
+            if (text.includes('FOR UPDATE OF attempt')) throw new Error('offline');
+            return { rows: [], rowCount: 0 };
+          },
+          release() {
+            released = true;
+          },
+        };
+      },
+      async query() {
+        throw new Error('not expected');
+      },
+    },
+    () => TOKEN,
+  );
+
+  await assert.rejects(
+    repository.claim({
+      ownerId: 'replica-a',
+      retentionMs: 60000,
+      limit: 1,
+      leaseMs: 5000,
+    }),
+    RunAttemptLogRetentionUnavailableError,
+  );
+  assert.equal(calls.at(-1), 'ROLLBACK');
+  assert.equal(released, true);
+});

packages/ql3-runtime-core/package.json

@@ -238,6 +238,9 @@
       ],
       "run-attempt-log-retention": [
         "dist/run/log-retention/runAttemptLogRetention.d.ts"
+      ],
+      "cluster-run-attempt-log-retention": [
+        "dist/run/log-retention/clusterRunAttemptLogRetention.d.ts"
       ]
     }
   },
@@ -297,6 +300,11 @@
       "require": "./dist/run/log-retention/runAttemptLogRetention.js",
       "default": "./dist/run/log-retention/runAttemptLogRetention.js"
     },
+    "./cluster-run-attempt-log-retention": {
+      "types": "./dist/run/log-retention/clusterRunAttemptLogRetention.d.ts",
+      "require": "./dist/run/log-retention/clusterRunAttemptLogRetention.js",
+      "default": "./dist/run/log-retention/clusterRunAttemptLogRetention.js"
+    },
     "./task-definition": {
       "types": "./dist/task-definition/taskDefinition.d.ts",
       "require": "./dist/task-definition/taskDefinition.js",

packages/ql3-runtime-core/src/run/log-retention/clusterRunAttemptLogRetention.ts

@@ -0,0 +1,62 @@
+import type {
+  RunAttemptLogRetentionCandidate,
+  RunAttemptLogRetentionStateReader,
+  RunAttemptLogRetirementRecord,
+} from './runAttemptLogRetention';
+
+export const MAX_CLUSTER_RUN_ATTEMPT_LOG_RETENTION_CLAIMS = 16;
+export const MIN_CLUSTER_RUN_ATTEMPT_LOG_RETENTION_LEASE_MS = 5_000;
+export const MAX_CLUSTER_RUN_ATTEMPT_LOG_RETENTION_LEASE_MS = 5 * 60_000;
+export const MAX_CLUSTER_RUN_ATTEMPT_LOG_RETENTION_RETRY_DELAY_MS =
+  24 * 60 * 60_000;
+
+export type ClusterRunAttemptLogRetentionFailureCode =
+  | 'artifact_unavailable'
+  | 'artifact_integrity_mismatch'
+  | 'retirement_record_unavailable';
+
+export interface ClusterRunAttemptLogRetentionClaim {
+  readonly candidate: Readonly<RunAttemptLogRetentionCandidate>;
+  readonly eligibleAtMs: number;
+  readonly observedAtMs: number;
+  readonly ownerId: string;
+  readonly token: string;
+  readonly version: number;
+  readonly expiresAtMs: number;
+  readonly failureCount: number;
+}
+
+export interface ClusterRunAttemptLogRetentionClaimPage {
+  readonly claims: readonly Readonly<ClusterRunAttemptLogRetentionClaim>[];
+  readonly hasMore: boolean;
+}
+
+export type ClusterRunAttemptLogRetentionSettlement =
+  | Readonly<{
+      readonly status: 'retired';
+      readonly record: Readonly<RunAttemptLogRetirementRecord>;
+    }>
+  | Readonly<{
+      readonly status: 'retry';
+      readonly delayMs: number;
+      readonly failureCode: ClusterRunAttemptLogRetentionFailureCode;
+    }>
+  | Readonly<{
+      readonly status: 'manual';
+      readonly failureCode: ClusterRunAttemptLogRetentionFailureCode;
+    }>;
+
+export interface ClusterRunAttemptLogRetentionClaimRepository
+  extends RunAttemptLogRetentionStateReader {
+  claim(options: Readonly<{
+    readonly ownerId: string;
+    readonly retentionMs: number;
+    readonly limit: number;
+    readonly leaseMs: number;
+  }>): Promise<Readonly<ClusterRunAttemptLogRetentionClaimPage>>;
+
+  settle(
+    claim: Readonly<ClusterRunAttemptLogRetentionClaim>,
+    settlement: Readonly<ClusterRunAttemptLogRetentionSettlement>,
+  ): Promise<'settled' | 'fenced'>;
+}

test/back/ql3PackageBoundaryAudit.test.cjs

@@ -74,7 +74,7 @@ test('current QL3 workspace has exactly eighteen reviewed package boundaries', (
     [
       {
         directory: 'packages/ql3-cluster-postgres/src/migrations',
-        directSourceFiles: 57,
+        directSourceFiles: 58,
         reviewKind: 'ordered_ledger',
       },
       {
@@ -299,10 +299,10 @@ test('current QL3 workspace has exactly eighteen reviewed package boundaries', (
       rootSourceFileRoles: runtimeCore.rootSourceFileRoles,
     },
     {
-      sourceFiles: 150,
+      sourceFiles: 151,
       rootSourceFiles: 1,
       rootSourceLines: 160,
-      nestedSourceFiles: 149,
+      nestedSourceFiles: 150,
       rootSourceFileRoles: { 'index.ts': 'public_export' },
     },
   );
@@ -421,10 +421,10 @@ test('current QL3 workspace has exactly eighteen reviewed package boundaries', (
       rootSourceFileRoles: clusterPostgres.rootSourceFileRoles,
     },
     {
-      sourceFiles: 147,
+      sourceFiles: 149,
       rootSourceFiles: 1,
       rootSourceLines: 125,
-      nestedSourceFiles: 146,
+      nestedSourceFiles: 148,
       rootSourceFileRoles: { 'index.ts': 'public_export' },
     },
   );