diff --git a/src/fault_log_analyzer/__main__.py b/src/fault_log_analyzer/__main__.py
index 2e7c81a..bd81913 100644
--- a/src/fault_log_analyzer/__main__.py
+++ b/src/fault_log_analyzer/__main__.py
@@ -1 +1,51 @@
-（已提交至远程 main 分支）
\ No newline at end of file
+"""CLI 入口。
+
+用法：
+    python -m fault_log_analyzer --storage memory --api-host 127.0.0.1 --api-port 8080
+"""
+
+from __future__ import annotations
+
+import argparse
+import sys
+from typing import List, Optional
+
+from .api import run_server
+from .config import Config
+from .storage import MemoryStorage, Storage
+
+
+def build_storage(storage_name: str) -> Storage:
+    """根据配置构建存储。生产后端（mysql 等）在 integrations.py 中。"""
+    if storage_name in ("memory", "mem"):
+        return MemoryStorage()
+    if storage_name == "mysql":
+        from .integrations import MySQLFaultRepository
+
+        cfg = Config.from_env()
+        return MySQLFaultRepository(
+            host=cfg.mysql_host,
+            port=cfg.mysql_port,
+            user=cfg.mysql_user,
+            password=cfg.mysql_password,
+            db=cfg.mysql_db,
+        )
+    raise ValueError(f"unknown storage: {storage_name}")
+
+
+def main(argv: Optional[List[str]] = None) -> int:
+    cfg = Config.from_env()
+    parser = argparse.ArgumentParser(prog="fault-log-analyzer", description="HMS 故障日志捕获与分析模块")
+    parser.add_argument("--storage", default=cfg.storage, help="存储后端（memory/mysql）")
+    parser.add_argument("--api-host", default=cfg.api_host, help="REST API 监听地址")
+    parser.add_argument("--api-port", type=int, default=cfg.api_port, help="REST API 监听端口")
+    args = parser.parse_args(argv)
+
+    storage = build_storage(args.storage)
+    print(f"fault-log-analyzer serving on http://{args.api_host}:{args.api_port} (storage={args.storage})")
+    run_server(args.api_host, args.api_port, storage)
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/src/fault_log_analyzer/api.py b/src/fault_log_analyzer/api.py
index 2e7c81a..d35a30d 100644
--- a/src/fault_log_analyzer/api.py
+++ b/src/fault_log_analyzer/api.py
@@ -1 +1,191 @@
-（已提交至远程 main 分支）
\ No newline at end of file
+"""REST API（标准库 http.server 实现）。
+
+依据 api-design.md 第 6 节，提供：
+- GET  /healthz
+- GET  /api/v1/fault-logs                      故障日志列表（分页/过滤）
+- GET  /api/v1/fault-logs/{id}                 故障日志详情
+- GET  /api/v1/fault-logs/{id}/root-cause      根因结论
+- GET  /api/v1/fault-types                     故障类型列表
+- POST /api/v1/fault-types                     新建故障类型
+- GET  /api/v1/fault-filters                   过滤规则列表
+- POST /api/v1/fault-filters                   新建过滤规则
+
+统一响应包装：{"code": 0, "message": "ok", "data": ...}。
+"""
+
+from __future__ import annotations
+
+import json
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+from typing import Any, Dict, Optional, Type
+from urllib.parse import parse_qs, urlparse
+
+from .models import FaultFilterRule, FaultType
+from .storage import Storage
+
+
+class FaultLogAPIHandler(BaseHTTPRequestHandler):
+    """绑定 MemoryStorage 的请求处理器。storage 由 create_handler 注入。"""
+
+    storage: Storage = None  # type: ignore[assignment]
+    server_version = "FaultLogAnalyzer/0.1"
+
+    # ------------------------------------------------------------------
+    # 工具方法
+    # ------------------------------------------------------------------
+    def _send_json(self, status: int, payload: Dict[str, Any]) -> None:
+        body = json.dumps(payload, ensure_ascii=False, default=str).encode("utf-8")
+        self.send_response(status)
+        self.send_header("Content-Type", "application/json; charset=utf-8")
+        self.send_header("Content-Length", str(len(body)))
+        self.end_headers()
+        self.wfile.write(body)
+
+    def _ok(self, data: Any, status: int = 200) -> None:
+        self._send_json(status, {"code": 0, "message": "ok", "data": data})
+
+    def _error(self, code: int, message: str, status: int) -> None:
+        self._send_json(status, {"code": code, "message": message, "data": None})
+
+    def _read_body(self) -> Dict[str, Any]:
+        length = int(self.headers.get("Content-Length", "0") or "0")
+        if length <= 0:
+            return {}
+        raw = self.rfile.read(length)
+        try:
+            data = json.loads(raw.decode("utf-8"))
+        except (json.JSONDecodeError, UnicodeDecodeError):
+            return {}
+        return data if isinstance(data, dict) else {}
+
+    def _query(self) -> Dict[str, str]:
+        parsed = urlparse(self.path)
+        qs = parse_qs(parsed.query)
+        return {k: v[0] for k, v in qs.items() if v}
+
+    def log_message(self, fmt: str, *args: Any) -> None:  # noqa: A002
+        # 静默访问日志，避免污染测试输出
+        return
+
+    # ------------------------------------------------------------------
+    # 路由
+    # ------------------------------------------------------------------
+    def do_GET(self) -> None:  # noqa: N802
+        parsed = urlparse(self.path)
+        path = parsed.path.rstrip("/") or "/"
+
+        if path == "/healthz":
+            return self._send_json(200, {"status": "ok"})
+
+        if path == "/api/v1/fault-logs":
+            return self._list_fault_logs()
+
+        if path.startswith("/api/v1/fault-logs/"):
+            rest = path[len("/api/v1/fault-logs/") :]
+            if rest.endswith("/root-cause"):
+                return self._get_root_cause(rest[: -len("/root-cause")])
+            return self._get_fault_log(rest)
+
+        if path == "/api/v1/fault-types":
+            return self._list_fault_types()
+
+        if path == "/api/v1/fault-filters":
+            return self._list_fault_filters()
+
+        return self._error(40401, "not found", 404)
+
+    def do_POST(self) -> None:  # noqa: N802
+        parsed = urlparse(self.path)
+        path = parsed.path.rstrip("/") or "/"
+
+        if path == "/api/v1/fault-types":
+            return self._create_fault_type()
+
+        if path == "/api/v1/fault-filters":
+            return self._create_fault_filter()
+
+        return self._error(40401, "not found", 404)
+
+    # ------------------------------------------------------------------
+    # 处理器
+    # ------------------------------------------------------------------
+    def _list_fault_logs(self) -> None:
+        q = self._query()
+        try:
+            page = max(1, int(q.get("page", "1")))
+            page_size = max(1, min(int(q.get("page_size", "20")), 200))
+        except ValueError:
+            return self._error(40001, "invalid param: page/page_size", 400)
+
+        result = self.storage.query_fault_logs(
+            host_id=q.get("host_id") or None,
+            fault_type=q.get("fault_type") or None,
+            level=q.get("level") or None,
+            keyword=q.get("keyword") or None,
+            page=page,
+            page_size=page_size,
+        )
+        items = [l.to_dict() for l in result["items"]]
+        return self._ok({"total": result["total"], "items": items})
+
+    def _get_fault_log(self, fault_log_id: str) -> None:
+        log = self.storage.get_fault_log(fault_log_id)
+        if log is None:
+            return self._error(40401, "fault log not found", 404)
+        return self._ok(log.to_dict())
+
+    def _get_root_cause(self, fault_log_id: str) -> None:
+        root_cause = self.storage.get_root_cause(fault_log_id)
+        if root_cause is None:
+            return self._error(40401, "root cause not found", 404)
+        return self._ok(root_cause.to_dict())
+
+    def _list_fault_types(self) -> None:
+        return self._ok({"items": [ft.to_dict() for ft in self.storage.list_fault_types()]})
+
+    def _create_fault_type(self) -> None:
+        body = self._read_body()
+        if not body.get("fault_type") or not body.get("name"):
+            return self._error(40002, "fault_type and name are required", 400)
+        try:
+            ft = FaultType.from_dict(body)
+        except (TypeError, ValueError):
+            return self._error(40001, "invalid param", 400)
+        self.storage.add_fault_type(ft)
+        return self._ok(ft.to_dict(), status=201)
+
+    def _list_fault_filters(self) -> None:
+        return self._ok({"items": [r.to_dict() for r in self.storage.list_filter_rules()]})
+
+    def _create_fault_filter(self) -> None:
+        body = self._read_body()
+        if not body.get("name"):
+            return self._error(40002, "name is required", 400)
+        try:
+            rule = FaultFilterRule.from_dict(body)
+        except (TypeError, ValueError):
+            return self._error(40001, "invalid param", 400)
+        self.storage.add_filter_rule(rule)
+        return self._ok(rule.to_dict(), status=201)
+
+
+def create_handler(storage: Storage) -> Type[FaultLogAPIHandler]:
+    """创建绑定存储的处理器类。"""
+
+    class Handler(FaultLogAPIHandler):
+        pass
+
+    Handler.storage = storage
+    return Handler
+
+
+def run_server(host: str, port: int, storage: Storage) -> None:
+    """启动 HTTP 服务（阻塞）。"""
+    handler = create_handler(storage)
+    server = ThreadingHTTPServer((host, port), handler)
+    try:
+        server.serve_forever()
+    except KeyboardInterrupt:
+        pass
+    finally:
+        server.server_close()
diff --git a/src/fault_log_analyzer/integrations.py b/src/fault_log_analyzer/integrations.py
index 2e7c81a..68526df 100644
--- a/src/fault_log_analyzer/integrations.py
+++ b/src/fault_log_analyzer/integrations.py
@@ -1 +1,199 @@
-（已提交至远程 main 分支）
\ No newline at end of file
+"""可选真实后端适配（Kafka / Elasticsearch / MySQL / Redis）。
+
+所有外部 SDK 均为惰性导入：未安装对应依赖时，仅在实例化时抛出明确异常，
+不影响核心逻辑与单元测试运行。
+
+- KafkaProducerAdapter / KafkaConsumerAdapter  -> kafka-python
+- ElasticsearchSink                            -> elasticsearch
+- MySQLFaultRepository                         -> pymysql
+- RedisDedupCache                              -> redis
+"""
+
+from __future__ import annotations
+
+import json
+from datetime import datetime
+from typing import Any, Dict, Iterator, List, Optional
+
+from .models import Event, FaultFilterRule, FaultLog, FaultType, RootCause
+from .storage import Storage
+
+
+# ---------------------------------------------------------------------------
+# Kafka
+# ---------------------------------------------------------------------------
+class KafkaProducerAdapter:
+    """Kafka 生产者封装（用于生产 logs.fault）。"""
+
+    def __init__(self, bootstrap_servers: str) -> None:
+        try:
+            from kafka import KafkaProducer  # type: ignore
+        except ImportError as exc:  # pragma: no cover - 依赖未安装
+            raise RuntimeError("Kafka 后端需要安装 kafka-python") from exc
+        self._producer = KafkaProducer(
+            bootstrap_servers=bootstrap_servers,
+            value_serializer=lambda v: json.dumps(v, ensure_ascii=False, default=str).encode("utf-8"),
+        )
+
+    def send(self, topic: str, value: Dict[str, Any]) -> None:
+        self._producer.send(topic, value)
+
+    def flush(self) -> None:
+        self._producer.flush()
+
+    def close(self) -> None:
+        self._producer.close()
+
+
+class KafkaConsumerAdapter:
+    """Kafka 消费者封装（用于消费 logs.raw）。"""
+
+    def __init__(self, bootstrap_servers: str, group_id: str, topic: str) -> None:
+        try:
+            from kafka import KafkaConsumer  # type: ignore
+        except ImportError as exc:  # pragma: no cover - 依赖未安装
+            raise RuntimeError("Kafka 后端需要安装 kafka-python") from exc
+        self._consumer = KafkaConsumer(
+            topic,
+            bootstrap_servers=bootstrap_servers,
+            group_id=group_id,
+            value_deserializer=lambda raw: json.loads(raw.decode("utf-8")),
+            auto_offset_reset="earliest",
+        )
+
+    def poll(self, timeout_ms: int = 1000) -> List[Dict[str, Any]]:
+        batch = self._consumer.poll(timeout_ms=timeout_ms)
+        return [record.value for records in batch.values() for record in records]
+
+    def close(self) -> None:
+        self._consumer.close()
+
+
+# ---------------------------------------------------------------------------
+# Elasticsearch
+# ---------------------------------------------------------------------------
+class ElasticsearchSink:
+    """故障日志写入 Elasticsearch（索引 hms-fault-log-{yyyy.MM}）。"""
+
+    def __init__(self, hosts: str, index_prefix: str = "hms-fault-log") -> None:
+        try:
+            from elasticsearch import Elasticsearch  # type: ignore
+        except ImportError as exc:  # pragma: no cover - 依赖未安装
+            raise RuntimeError("Elasticsearch 后端需要安装 elasticsearch") from exc
+        host_list = [h.strip() for h in hosts.split(",") if h.strip()]
+        self._client = Elasticsearch(host_list or ["http://localhost:9200"])
+        self.index_prefix = index_prefix
+
+    @staticmethod
+    def _index_name(prefix: str, occurred_at: Optional[datetime]) -> str:
+        dt = occurred_at or datetime.utcnow()
+        return f"{prefix}-{dt:%Y.%m}"
+
+    def write(self, log: FaultLog) -> None:
+        index = self._index_name(self.index_prefix, log.occurred_at)
+        self._client.index(index=index, id=log.fault_log_id, document=log.to_dict())
+
+    def close(self) -> None:
+        self._client.close()
+
+
+# ---------------------------------------------------------------------------
+# MySQL
+# ---------------------------------------------------------------------------
+class MySQLFaultRepository:
+    """故障日志 / 根因结果 MySQL 落库（简化子集，生产可按需扩展）。"""
+
+    def __init__(self, host: str, port: int, user: str, password: str, db: str) -> None:
+        try:
+            import pymysql  # type: ignore
+        except ImportError as exc:  # pragma: no cover - 依赖未安装
+            raise RuntimeError("MySQL 后端需要安装 pymysql") from exc
+        self._conn = pymysql.connect(
+            host=host, port=port, user=user, password=password, database=db
+        )
+
+    def save_fault_log(self, log: FaultLog) -> None:
+        sql = (
+            "INSERT INTO fault_log (fault_log_id, host_id, fault_type, cluster_id, "
+            "fingerprint, level, service, message, trace_id, occurred_at, count) "
+            "VALUES (%s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s) "
+            "ON DUPLICATE KEY UPDATE fault_type=VALUES(fault_type), "
+            "cluster_id=VALUES(cluster_id), count=VALUES(count)"
+        )
+        with self._conn.cursor() as cursor:
+            cursor.execute(
+                sql,
+                (
+                    log.fault_log_id,
+                    log.host_id,
+                    log.fault_type,
+                    log.cluster_id,
+                    log.fingerprint,
+                    log.level,
+                    log.service,
+                    log.message,
+                    log.trace_id,
+                    log.occurred_at,
+                    log.count,
+                ),
+            )
+        self._conn.commit()
+
+    def save_root_cause(self, root_cause: RootCause) -> None:
+        sql = (
+            "INSERT INTO root_cause (fault_log_id, cause_type, evidence, confidence, analysis_at) "
+            "VALUES (%s, %s, %s, %s, %s) "
+            "ON DUPLICATE KEY UPDATE cause_type=VALUES(cause_type), "
+            "evidence=VALUES(evidence), confidence=VALUES(confidence)"
+        )
+        with self._conn.cursor() as cursor:
+            cursor.execute(
+                sql,
+                (
+                    root_cause.fault_log_id,
+                    root_cause.cause_type,
+                    json.dumps(root_cause.evidence, ensure_ascii=False, default=str),
+                    root_cause.confidence,
+                    root_cause.analysis_at,
+                ),
+            )
+        self._conn.commit()
+
+    def close(self) -> None:
+        self._conn.close()
+
+
+# ---------------------------------------------------------------------------
+# Redis
+# ---------------------------------------------------------------------------
+class RedisDedupCache:
+    """基于 Redis 的指纹去重（对应 Redis 键 hms:log:fp:{fingerprint}）。"""
+
+    def __init__(self, url: str, window_seconds: int = 300) -> None:
+        try:
+            import redis  # type: ignore
+        except ImportError as exc:  # pragma: no cover - 依赖未安装
+            raise RuntimeError("Redis 后端需要安装 redis") from exc
+        self._redis = redis.Redis.from_url(url)
+        self.window_seconds = window_seconds
+
+    def is_duplicate(self, fingerprint: str) -> bool:
+        key = f"hms:log:fp:{fingerprint}"
+        # SET NX EX：成功写入返回 True 表示首次出现；失败表示窗口内重复
+        return not bool(self._redis.set(key, "1", nx=True, ex=self.window_seconds))
+
+    def mark(self, fingerprint: str) -> None:
+        key = f"hms:log:fp:{fingerprint}"
+        self._redis.set(key, "1", ex=self.window_seconds)
+
+    def close(self) -> None:
+        self._redis.close()
+
+
+__all__ = [
+    "KafkaProducerAdapter",
+    "KafkaConsumerAdapter",
+    "ElasticsearchSink",
+    "MySQLFaultRepository",
+    "RedisDedupCache",
+]
diff --git a/src/fault_log_analyzer/pipeline.py b/src/fault_log_analyzer/pipeline.py
index 2e7c81a..800bfa2 100644
--- a/src/fault_log_analyzer/pipeline.py
+++ b/src/fault_log_analyzer/pipeline.py
@@ -1 +1,97 @@
-（已提交至远程 main 分支）
\ No newline at end of file
+"""故障日志捕获管道。
+
+依据 architecture.md 5.3.2：
+```
+logs.raw → 级别过滤(ERROR/FATAL) → 关键字/正则过滤 → 结构化解析
+        → 去重打标 → Kafka(logs.fault) → Elasticsearch(hms-fault-log-*)
+```
+
+本模块把「过滤 → 解析 → 指纹去重 → 启发式归类 → 落库 → 可选生产」串成
+一个可独立测试的管道。聚类与根因分析由 workers.py 异步完成。
+"""
+
+from __future__ import annotations
+
+import uuid
+from typing import Any, Callable, List, Optional
+
+from .classifier import Classifier
+from .filters import FaultFilter
+from .fingerprint import Deduplicator, MinHash
+from .models import FaultLog
+from .parser import LogParser
+from .storage import Storage
+
+
+class CapturePipeline:
+    """故障日志捕获管道。
+
+    参数：
+        storage: 故障日志存储。
+        capture_filter: 过滤规则，默认 FaultFilter（ERROR/FATAL）。
+        parser: 日志解析器，默认 LogParser。
+        classifier: 启发式归类器，默认 Classifier。
+        dedup: 去重器（实现 is_duplicate 接口），默认进程内 Deduplicator。
+        minhash: 指纹器，默认 MinHash。
+        producer: 可选回调，捕获成功后调用（生产到 logs.fault / ES）。
+    """
+
+    def __init__(
+        self,
+        storage: Storage,
+        capture_filter: Optional[FaultFilter] = None,
+        parser: Optional[LogParser] = None,
+        classifier: Optional[Classifier] = None,
+        dedup: Optional[Deduplicator] = None,
+        minhash: Optional[MinHash] = None,
+        producer: Optional[Callable[[FaultLog], None]] = None,
+    ) -> None:
+        self.storage = storage
+        self.capture_filter = capture_filter or FaultFilter()
+        self.parser = parser or LogParser()
+        self.classifier = classifier or Classifier()
+        self.dedup = dedup or Deduplicator()
+        self.minhash = minhash or MinHash()
+        self.producer = producer
+        self.captured = 0
+        self.duplicates = 0
+
+    def process(self, raw: Any) -> Optional[FaultLog]:
+        """处理一条原始日志；不是故障日志或重复时返回 None。"""
+        parsed = self.parser.parse(raw)
+
+        if not self.capture_filter.should_capture(parsed):
+            return None
+
+        fingerprint = self.minhash.fingerprint(parsed.message or "")
+        if self.dedup.is_duplicate(fingerprint):
+            self.duplicates += 1
+            return None
+
+        fault_type = self.classifier.classify(parsed.message or "")
+        log = FaultLog(
+            fault_log_id=uuid.uuid4().hex,
+            host_id=parsed.host_id or "",
+            message=parsed.message or "",
+            level=parsed.level or "ERROR",
+            fingerprint=fingerprint,
+            fault_type=fault_type,
+            service=parsed.service or None,
+            trace_id=parsed.trace_id,
+            occurred_at=parsed.timestamp,
+            count=1,
+        )
+        self.storage.save_fault_log(log)
+        if self.producer is not None:
+            self.producer(log)
+        self.captured += 1
+        return log
+
+    def process_batch(self, raws: List[Any]) -> List[FaultLog]:
+        """批量处理原始日志，返回捕获到的 FaultLog 列表。"""
+        results: List[FaultLog] = []
+        for raw in raws:
+            log = self.process(raw)
+            if log is not None:
+                results.append(log)
+        return results
diff --git a/src/fault_log_analyzer/storage.py b/src/fault_log_analyzer/storage.py
index 2e7c81a..1d34cd6 100644
--- a/src/fault_log_analyzer/storage.py
+++ b/src/fault_log_analyzer/storage.py
@@ -1 +1,180 @@
-（已提交至远程 main 分支）
\ No newline at end of file
+"""存储抽象与内存实现。
+
+依据 docs/01-design/database-design.md 中 fault_log / root_cause /
+fault_type / fault_filter_rule / event 表结构，提供统一的进程内存储接口。
+生产环境通过 integrations.py 对接 MySQL / Elasticsearch / Redis；
+本模块保证零第三方依赖，测试与离线运行使用 MemoryStorage。
+"""
+
+from __future__ import annotations
+
+import threading
+from datetime import datetime
+from typing import Dict, List, Optional
+
+from .models import Event, FaultFilterRule, FaultLog, FaultType, RootCause
+
+
+class Storage:
+    """存储接口（抽象基类）。
+
+    设计上把五类实体（故障日志 / 根因 / 故障类型 / 过滤规则 / 事件）收敛到
+    一个 Storage 门面，便于 pipeline / workers / api 复用同一份存储。
+    """
+
+    # ---- 故障日志 ------------------------------------------------------
+    def save_fault_log(self, log: FaultLog) -> None:
+        raise NotImplementedError
+
+    def get_fault_log(self, fault_log_id: str) -> Optional[FaultLog]:
+        raise NotImplementedError
+
+    def list_fault_logs(self) -> List[FaultLog]:
+        raise NotImplementedError
+
+    def query_fault_logs(
+        self,
+        host_id: Optional[str] = None,
+        fault_type: Optional[str] = None,
+        level: Optional[str] = None,
+        keyword: Optional[str] = None,
+        page: int = 1,
+        page_size: int = 20,
+    ) -> Dict:
+        raise NotImplementedError
+
+    # ---- 根因 ----------------------------------------------------------
+    def save_root_cause(self, root_cause: RootCause) -> None:
+        raise NotImplementedError
+
+    def get_root_cause(self, fault_log_id: str) -> Optional[RootCause]:
+        raise NotImplementedError
+
+    # ---- 故障类型 ------------------------------------------------------
+    def list_fault_types(self) -> List[FaultType]:
+        raise NotImplementedError
+
+    def get_fault_type(self, fault_type: str) -> Optional[FaultType]:
+        raise NotImplementedError
+
+    def add_fault_type(self, fault_type: FaultType) -> FaultType:
+        raise NotImplementedError
+
+    # ---- 过滤规则 ------------------------------------------------------
+    def list_filter_rules(self) -> List[FaultFilterRule]:
+        raise NotImplementedError
+
+    def add_filter_rule(self, rule: FaultFilterRule) -> FaultFilterRule:
+        raise NotImplementedError
+
+    # ---- 事件（根因分析关联输入） --------------------------------------
+    def save_event(self, event: Event) -> None:
+        raise NotImplementedError
+
+    def list_events_by_host(
+        self, host_id: str, start: datetime, end: datetime
+    ) -> List[Event]:
+        raise NotImplementedError
+
+
+class MemoryStorage(Storage):
+    """进程内内存实现，用于单元测试与离线运行。"""
+
+    def __init__(self) -> None:
+        self._lock = threading.RLock()
+        self._fault_logs: Dict[str, FaultLog] = {}
+        self._root_causes: Dict[str, RootCause] = {}
+        self._fault_types: Dict[str, FaultType] = {}
+        self._filter_rules: Dict[str, FaultFilterRule] = {}
+        self._events: List[Event] = []
+
+    # ---- 故障日志 ------------------------------------------------------
+    def save_fault_log(self, log: FaultLog) -> None:
+        with self._lock:
+            self._fault_logs[log.fault_log_id] = log
+
+    def get_fault_log(self, fault_log_id: str) -> Optional[FaultLog]:
+        with self._lock:
+            return self._fault_logs.get(fault_log_id)
+
+    def list_fault_logs(self) -> List[FaultLog]:
+        with self._lock:
+            return sorted(
+                self._fault_logs.values(),
+                key=lambda l: (l.occurred_at, l.fault_log_id),
+            )
+
+    def query_fault_logs(
+        self,
+        host_id: Optional[str] = None,
+        fault_type: Optional[str] = None,
+        level: Optional[str] = None,
+        keyword: Optional[str] = None,
+        page: int = 1,
+        page_size: int = 20,
+    ) -> Dict:
+        with self._lock:
+            items = list(self._fault_logs.values())
+        if host_id:
+            items = [l for l in items if l.host_id == host_id]
+        if fault_type:
+            items = [l for l in items if l.fault_type == fault_type]
+        if level:
+            items = [l for l in items if l.level.upper() == level.upper()]
+        if keyword:
+            needle = keyword.lower()
+            items = [l for l in items if needle in (l.message or "").lower()]
+        items.sort(key=lambda l: (l.occurred_at, l.fault_log_id), reverse=True)
+        total = len(items)
+        page = max(1, page)
+        page_size = max(1, min(page_size, 200))
+        start = (page - 1) * page_size
+        return {"total": total, "items": items[start : start + page_size]}
+
+    # ---- 根因 ----------------------------------------------------------
+    def save_root_cause(self, root_cause: RootCause) -> None:
+        with self._lock:
+            self._root_causes[root_cause.fault_log_id] = root_cause
+
+    def get_root_cause(self, fault_log_id: str) -> Optional[RootCause]:
+        with self._lock:
+            return self._root_causes.get(fault_log_id)
+
+    # ---- 故障类型 ------------------------------------------------------
+    def list_fault_types(self) -> List[FaultType]:
+        with self._lock:
+            return list(self._fault_types.values())
+
+    def get_fault_type(self, fault_type: str) -> Optional[FaultType]:
+        with self._lock:
+            return self._fault_types.get(fault_type)
+
+    def add_fault_type(self, fault_type: FaultType) -> FaultType:
+        with self._lock:
+            self._fault_types[fault_type.fault_type] = fault_type
+        return fault_type
+
+    # ---- 过滤规则 ------------------------------------------------------
+    def list_filter_rules(self) -> List[FaultFilterRule]:
+        with self._lock:
+            return list(self._filter_rules.values())
+
+    def add_filter_rule(self, rule: FaultFilterRule) -> FaultFilterRule:
+        with self._lock:
+            self._filter_rules[rule.name] = rule
+        return rule
+
+    # ---- 事件 ----------------------------------------------------------
+    def save_event(self, event: Event) -> None:
+        with self._lock:
+            self._events.append(event)
+
+    def list_events_by_host(
+        self, host_id: str, start: datetime, end: datetime
+    ) -> List[Event]:
+        with self._lock:
+            return [
+                e
+                for e in self._events
+                if e.host_id == host_id and start <= e.fired_at <= end
+            ]
diff --git a/src/fault_log_analyzer/workers.py b/src/fault_log_analyzer/workers.py
index 2e7c81a..c765f56 100644
--- a/src/fault_log_analyzer/workers.py
+++ b/src/fault_log_analyzer/workers.py
@@ -1 +1,108 @@
-（已提交至远程 main 分支）
\ No newline at end of file
+"""聚类与根因分析 worker。
+
+依据 architecture.md 5.3.3 / 5.3.4：
+- ClusterWorker：批量对未归类故障日志做 TF-IDF + DBSCAN 聚类，映射到 fault_type。
+- RootCauseWorker：批量关联同主机近期指标事件，产出 root_cause。
+"""
+
+from __future__ import annotations
+
+from datetime import timedelta
+from typing import Dict, List, Optional, Sequence
+
+from .classifier import Classifier
+from .cluster import Clusterer
+from .models import FaultLog, RootCause
+from .root_cause import RootCauseAnalyzer
+from .storage import Storage
+
+
+def _cluster_label(cluster_id: int) -> str:
+    return f"c-{cluster_id}"
+
+
+class ClusterWorker:
+    """批量聚类并归类故障日志。"""
+
+    def __init__(
+        self,
+        storage: Storage,
+        clusterer: Optional[Clusterer] = None,
+        classifier: Optional[Classifier] = None,
+    ) -> None:
+        self.storage = storage
+        self.clusterer = clusterer or Clusterer()
+        self.classifier = classifier or Classifier()
+
+    def run(self, logs: Optional[Sequence[FaultLog]] = None) -> Dict:
+        """对故障日志聚类并持久化归类结果。
+
+        返回：{"clustered": n, "clusters": m, "noise": k}
+        """
+        items: List[FaultLog] = list(logs) if logs is not None else self.storage.list_fault_logs()
+        if not items:
+            return {"clustered": 0, "clusters": 0, "noise": 0}
+
+        messages = [l.message or "" for l in items]
+        labels = self.clusterer.fit_predict(messages)
+
+        # 按簇收集代表消息，用于簇级归类
+        cluster_messages: Dict[int, List[str]] = {}
+        for label, message in zip(labels, messages):
+            if label >= 0:
+                cluster_messages.setdefault(label, []).append(message)
+
+        fault_type_by_cluster: Dict[int, Optional[str]] = {}
+        for label, msgs in cluster_messages.items():
+            fault_type_by_cluster[label] = self.classifier.classify_cluster(label, msgs)
+
+        clustered = 0
+        for log, label in zip(items, labels):
+            if label >= 0:
+                log.cluster_id = _cluster_label(label)
+                log.fault_type = fault_type_by_cluster[label] or log.fault_type
+                clustered += 1
+            self.storage.save_fault_log(log)
+
+        noise = sum(1 for lbl in labels if lbl < 0)
+        return {
+            "clustered": clustered,
+            "clusters": self.clusterer.n_clusters,
+            "noise": noise,
+        }
+
+
+class RootCauseWorker:
+    """批量执行根因分析并持久化结论。"""
+
+    def __init__(
+        self,
+        storage: Storage,
+        analyzer: Optional[RootCauseAnalyzer] = None,
+        window_minutes: int = 5,
+    ) -> None:
+        self.storage = storage
+        self.window_minutes = window_minutes
+        self.analyzer = analyzer or RootCauseAnalyzer(window_minutes=window_minutes)
+
+    def run(self, logs: Optional[Sequence[FaultLog]] = None) -> List[RootCause]:
+        """对尚未分析过的故障日志执行根因分析，返回新产出的结论列表。"""
+        items: List[FaultLog] = list(logs) if logs is not None else self.storage.list_fault_logs()
+        results: List[RootCause] = []
+        window = timedelta(minutes=self.window_minutes)
+
+        for log in items:
+            if self.storage.get_root_cause(log.fault_log_id) is not None:
+                continue
+            start = log.occurred_at - window
+            end = log.occurred_at + window
+            events = (
+                self.storage.list_events_by_host(log.host_id, start, end)
+                if log.host_id
+                else []
+            )
+            root_cause = self.analyzer.analyze(log, events)
+            self.storage.save_root_cause(root_cause)
+            results.append(root_cause)
+
+        return results
diff --git a/tests/test_config.py b/tests/test_config.py
index 2e7c81a..eaf88f8 100644
--- a/tests/test_config.py
+++ b/tests/test_config.py
@@ -1 +1,75 @@
-（已提交至远程 main 分支）
\ No newline at end of file
+"""Config 配置模型与环境变量加载测试。"""
+import os
+import unittest
+
+from fault_log_analyzer.config import Config, _env, _env_bool, _env_float, _env_int
+
+
+class TestEnvHelpers(unittest.TestCase):
+    def test_env_default(self):
+        self.assertEqual(_env("HMS_TEST_UNSET_XYZ", "dflt"), "dflt")
+
+    def test_env_int_parse_and_fallback(self):
+        os.environ["HMS_TEST_INT"] = "42"
+        self.assertEqual(_env_int("HMS_TEST_INT", 1), 42)
+        os.environ["HMS_TEST_INT"] = "abc"
+        self.assertEqual(_env_int("HMS_TEST_INT", 7), 7)
+        del os.environ["HMS_TEST_INT"]
+
+    def test_env_float_parse_and_fallback(self):
+        os.environ["HMS_TEST_FLOAT"] = "0.75"
+        self.assertAlmostEqual(_env_float("HMS_TEST_FLOAT", 1.0), 0.75)
+        os.environ["HMS_TEST_FLOAT"] = "x"
+        self.assertEqual(_env_float("HMS_TEST_FLOAT", 2.5), 2.5)
+        del os.environ["HMS_TEST_FLOAT"]
+
+    def test_env_bool_variants(self):
+        for raw in ("1", "true", "TRUE", "yes", "on"):
+            os.environ["HMS_TEST_BOOL"] = raw
+            self.assertTrue(_env_bool("HMS_TEST_BOOL", False))
+        for raw in ("0", "false", "no", "off"):
+            os.environ["HMS_TEST_BOOL"] = raw
+            self.assertFalse(_env_bool("HMS_TEST_BOOL", True))
+        del os.environ["HMS_TEST_BOOL"]
+
+
+class TestConfig(unittest.TestCase):
+    def test_defaults(self):
+        cfg = Config()
+        self.assertEqual(cfg.kafka_bootstrap_servers, "localhost:9092")
+        self.assertEqual(cfg.kafka_logs_raw_topic, "logs.raw")
+        self.assertEqual(cfg.kafka_logs_fault_topic, "logs.fault")
+        self.assertEqual(cfg.cluster_eps, 0.75)
+        self.assertEqual(cfg.cluster_min_samples, 5)
+        self.assertEqual(cfg.api_port, 8080)
+        self.assertEqual(cfg.storage, "memory")
+
+    def test_from_env_defaults(self):
+        cfg = Config.from_env()
+        self.assertEqual(cfg.api_host, "127.0.0.1")
+        self.assertEqual(cfg.root_cause_window_minutes, 5)
+
+    def test_from_env_override(self):
+        os.environ["KAFKA_BOOTSTRAP_SERVERS"] = "kafka:29092"
+        os.environ["CLUSTER_EPS"] = "0.6"
+        os.environ["API_PORT"] = "9090"
+        os.environ["STORAGE"] = "mysql"
+        try:
+            cfg = Config.from_env()
+            self.assertEqual(cfg.kafka_bootstrap_servers, "kafka:29092")
+            self.assertAlmostEqual(cfg.cluster_eps, 0.6)
+            self.assertEqual(cfg.api_port, 9090)
+            self.assertEqual(cfg.storage, "mysql")
+        finally:
+            for k in ("KAFKA_BOOTSTRAP_SERVERS", "CLUSTER_EPS", "API_PORT", "STORAGE"):
+                os.environ.pop(k, None)
+
+    def test_to_dict_roundtrip(self):
+        cfg = Config(api_port=9001)
+        data = cfg.to_dict()
+        self.assertEqual(data["api_port"], 9001)
+        self.assertIn("cluster_eps", data)
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/tests/test_filters.py b/tests/test_filters.py
index 2e7c81a..f89158c 100644
--- a/tests/test_filters.py
+++ b/tests/test_filters.py
@@ -1 +1,63 @@
-（已提交至远程 main 分支）
\ No newline at end of file
+"""故障日志过滤规则测试。"""
+import unittest
+
+from fault_log_analyzer.filters import DEFAULT_LEVELS, FaultFilter
+from fault_log_analyzer.models import FaultFilterRule, ParsedLog
+
+
+def _log(level="ERROR", message="boom"):
+    return ParsedLog(level=level, message=message)
+
+
+class TestDefaultFilter(unittest.TestCase):
+    def test_error_captured(self):
+        self.assertTrue(FaultFilter().should_capture(_log("ERROR", "x")))
+
+    def test_fatal_captured(self):
+        self.assertTrue(FaultFilter().should_capture(_log("FATAL", "x")))
+
+    def test_info_not_captured(self):
+        self.assertFalse(FaultFilter().should_capture(_log("INFO", "x")))
+
+    def test_warn_not_captured(self):
+        self.assertFalse(FaultFilter().should_capture(_log("WARN", "x")))
+
+    def test_default_levels_set(self):
+        self.assertEqual(DEFAULT_LEVELS, {"ERROR", "FATAL"})
+
+
+class TestRuleFilter(unittest.TestCase):
+    def test_level_rule(self):
+        f = FaultFilter([FaultFilterRule(name="warn", level="WARN")])
+        self.assertTrue(f.should_capture(_log("WARN", "x")))
+        self.assertFalse(f.should_capture(_log("ERROR", "x")))
+
+    def test_pattern_rule(self):
+        f = FaultFilter([FaultFilterRule(name="oom", level="INFO", pattern="out of memory")])
+        self.assertTrue(f.should_capture(_log("INFO", "out of memory occurred")))
+        self.assertFalse(f.should_capture(_log("INFO", "nothing")))
+
+    def test_exclude_pattern(self):
+        f = FaultFilter(
+            [FaultFilterRule(name="no_heartbeat", level="ERROR", pattern=".*", exclude_pattern="heartbeat")]
+        )
+        self.assertFalse(f.should_capture(_log("ERROR", "heartbeat timeout")))
+        self.assertTrue(f.should_capture(_log("ERROR", "disk full")))
+
+    def test_disabled_rule_skipped(self):
+        f = FaultFilter([FaultFilterRule(name="off", level="WARN", enabled=False)])
+        self.assertFalse(f.should_capture(_log("WARN", "x")))
+
+    def test_add_rule(self):
+        f = FaultFilter()
+        f.add_rule(FaultFilterRule(name="warn", level="WARN"))
+        self.assertTrue(f.should_capture(_log("WARN", "x")))
+
+    def test_compile_rules(self):
+        rules = FaultFilter.compile_rules([{"name": "r1", "level": "WARN"}])
+        self.assertEqual(len(rules), 1)
+        self.assertEqual(rules[0].name, "r1")
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/tests/test_models.py b/tests/test_models.py
index 2e7c81a..3985592 100644
--- a/tests/test_models.py
+++ b/tests/test_models.py
@@ -1 +1,125 @@
-（已提交至远程 main 分支）
\ No newline at end of file
+"""数据模型测试。"""
+import unittest
+from datetime import datetime, timezone
+
+from fault_log_analyzer.models import (
+    Event,
+    FaultFilterRule,
+    FaultLog,
+    FaultType,
+    ParsedLog,
+    RootCause,
+    _to_dt,
+)
+
+
+class TestToDt(unittest.TestCase):
+    def test_none_returns_now(self):
+        self.assertIsInstance(_to_dt(None), datetime)
+
+    def test_datetime_naive_gets_utc(self):
+        dt = datetime(2025, 1, 1, 10, 0, 0)
+        out = _to_dt(dt)
+        self.assertEqual(out.tzinfo, timezone.utc)
+
+    def test_datetime_aware_kept(self):
+        dt = datetime(2025, 1, 1, 10, 0, 0, tzinfo=timezone.utc)
+        self.assertEqual(_to_dt(dt), dt)
+
+    def test_timestamp(self):
+        ts = 1735722000.0
+        out = _to_dt(ts)
+        self.assertEqual(out.tzinfo, timezone.utc)
+        self.assertEqual(int(out.timestamp()), int(ts))
+
+    def test_iso_string_z(self):
+        out = _to_dt("2025-01-01T10:18:00Z")
+        self.assertEqual(out.tzinfo, timezone.utc)
+        self.assertEqual(out.hour, 10)
+
+    def test_iso_string_space(self):
+        out = _to_dt("2025-01-01 10:18:00")
+        self.assertEqual(out.hour, 10)
+
+
+class TestFaultLog(unittest.TestCase):
+    def test_to_from_dict_roundtrip(self):
+        log = FaultLog(
+            fault_log_id="fl-1",
+            host_id="h-1",
+            message="No space left on device",
+            level="ERROR",
+            fingerprint="abc",
+            fault_type="disk_full",
+            cluster_id="c-1",
+            occurred_at=datetime(2025, 1, 1, 10, 18, 0, tzinfo=timezone.utc),
+        )
+        data = log.to_dict()
+        self.assertEqual(data["fault_log_id"], "fl-1")
+        restored = FaultLog.from_dict(data)
+        self.assertEqual(restored.fault_log_id, "fl-1")
+        self.assertEqual(restored.occurred_at, log.occurred_at)
+        self.assertEqual(restored.fault_type, "disk_full")
+
+    def test_from_dict_ignores_unknown_fields(self):
+        log = FaultLog.from_dict({"fault_log_id": "x", "host_id": "h", "message": "m", "extra": 1})
+        self.assertEqual(log.fault_log_id, "x")
+        self.assertFalse(hasattr(log, "extra"))
+
+
+class TestRootCause(unittest.TestCase):
+    def test_to_from_dict(self):
+        rc = RootCause(
+            fault_log_id="fl-1",
+            cause_type="disk_full",
+            evidence=[{"type": "event", "metric": "disk_used_percent"}],
+            confidence=0.92,
+        )
+        data = rc.to_dict()
+        restored = RootCause.from_dict(data)
+        self.assertEqual(restored.cause_type, "disk_full")
+        self.assertEqual(restored.confidence, 0.92)
+        self.assertEqual(restored.evidence[0]["metric"], "disk_used_percent")
+
+
+class TestFaultType(unittest.TestCase):
+    def test_to_from_dict(self):
+        ft = FaultType(fault_type="disk_full", name="磁盘空间不足", pattern="no space left")
+        restored = FaultType.from_dict(ft.to_dict())
+        self.assertEqual(restored.fault_type, "disk_full")
+        self.assertEqual(restored.pattern, "no space left")
+
+
+class TestFaultFilterRule(unittest.TestCase):
+    def test_to_from_dict(self):
+        rule = FaultFilterRule(name="oom", level="ERROR", pattern="out of memory")
+        restored = FaultFilterRule.from_dict(rule.to_dict())
+        self.assertEqual(restored.name, "oom")
+        self.assertEqual(restored.level, "ERROR")
+
+
+class TestEvent(unittest.TestCase):
+    def test_to_from_dict(self):
+        event = Event(
+            event_id="e-1",
+            host_id="h-1",
+            metric="disk_used_percent",
+            value=97.2,
+            threshold=90.0,
+            fired_at=datetime(2025, 1, 1, 10, 18, 0, tzinfo=timezone.utc),
+        )
+        restored = Event.from_dict(event.to_dict())
+        self.assertEqual(restored.event_id, "e-1")
+        self.assertEqual(restored.value, 97.2)
+
+
+class TestParsedLog(unittest.TestCase):
+    def test_to_dict(self):
+        pl = ParsedLog(host_id="h-1", level="ERROR", message="boom", raw="boom")
+        data = pl.to_dict()
+        self.assertEqual(data["host_id"], "h-1")
+        self.assertIn("timestamp", data)
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/tests/test_parser.py b/tests/test_parser.py
index 2e7c81a..5790161 100644
--- a/tests/test_parser.py
+++ b/tests/test_parser.py
@@ -1 +1,101 @@
-（已提交至远程 main 分支）
\ No newline at end of file
+"""日志解析与模板化测试。"""
+import unittest
+
+from fault_log_analyzer.models import ParsedLog
+from fault_log_analyzer.parser import LogParser
+
+
+class TestParseJson(unittest.TestCase):
+    def setUp(self):
+        self.parser = LogParser()
+
+    def test_json_dict(self):
+        raw = {
+            "timestamp": "2025-01-01T10:18:00Z",
+            "level": "ERROR",
+            "message": "disk full",
+            "host_id": "h-1",
+            "service": "svc",
+            "trace_id": "tr-1",
+        }
+        parsed = self.parser.parse(raw)
+        self.assertEqual(parsed.level, "ERROR")
+        self.assertEqual(parsed.message, "disk full")
+        self.assertEqual(parsed.host_id, "h-1")
+        self.assertEqual(parsed.service, "svc")
+        self.assertEqual(parsed.trace_id, "tr-1")
+
+    def test_json_dict_uses_msg_alias(self):
+        parsed = self.parser.parse({"msg": "hello", "severity": "warn"})
+        self.assertEqual(parsed.message, "hello")
+        self.assertEqual(parsed.level, "WARN")
+
+    def test_json_line(self):
+        line = '{"timestamp":"2025-01-01T10:00:00Z","level":"FATAL","message":"oom"}'
+        parsed = self.parser.parse(line)
+        self.assertEqual(parsed.level, "FATAL")
+        self.assertEqual(parsed.message, "oom")
+
+
+class TestParseLog4j(unittest.TestCase):
+    def setUp(self):
+        self.parser = LogParser()
+
+    def test_log4j(self):
+        line = "2025-01-01 10:18:00 ERROR [app] No space left on device"
+        parsed = self.parser.parse(line)
+        self.assertEqual(parsed.level, "ERROR")
+        self.assertEqual(parsed.service, "app")
+        self.assertEqual(parsed.message, "No space left on device")
+
+    def test_log4j_with_trace(self):
+        line = "2025-01-01 10:18:00 ERROR [svc] trace_id=abc-123-456 connection refused"
+        parsed = self.parser.parse(line)
+        self.assertEqual(parsed.trace_id, "abc-123-456")
+
+
+class TestParseSyslog(unittest.TestCase):
+    def setUp(self):
+        self.parser = LogParser()
+
+    def test_syslog(self):
+        line = "<134>Jan  1 10:18:00 web-01 nginx[123]: connect timeout to upstream"
+        parsed = self.parser.parse(line)
+        self.assertEqual(parsed.host_id, "web-01")
+        self.assertEqual(parsed.service, "nginx")
+        self.assertIn("connect timeout", parsed.message)
+
+
+class TestParseFallback(unittest.TestCase):
+    def setUp(self):
+        self.parser = LogParser()
+
+    def test_fallback_level_guess(self):
+        parsed = self.parser.parse("some random FATAL failure")
+        self.assertEqual(parsed.level, "FATAL")
+        self.assertEqual(parsed.message, "some random FATAL failure")
+
+    def test_passthrough_parsed_log(self):
+        pl = ParsedLog(level="ERROR", message="x")
+        self.assertIs(self.parser.parse(pl), pl)
+
+
+class TestTemplate(unittest.TestCase):
+    def test_ip_uuid_number_replacement(self):
+        msg = "connect from 10.0.0.1 uuid 123e4567-e89b-12d3-a456-426614174000 retry 42"
+        templated = LogParser.template(msg)
+        self.assertIn("<IP>", templated)
+        self.assertIn("<UUID>", templated)
+        self.assertIn("<NUM>", templated)
+        self.assertNotIn("10.0.0.1", templated)
+
+    def test_tokenize_filters_short(self):
+        tokens = LogParser.tokenize("a b disk full error")
+        self.assertIn("disk", tokens)
+        self.assertIn("error", tokens)
+        self.assertNotIn("a", tokens)
+        self.assertNotIn("b", tokens)
+
+
+if __name__ == "__main__":
+    unittest.main()