2025-10-16 17:16:07 +08:00
5 changed files with 166 additions and 21 deletions
--- a/src/agent/README.md
+++ b/src/agent/README.md
@ -41,8 +41,11 @@ Agent 不再依赖配置文件；所有参数均由环境变量与主机名推
 主机名与元数据的解析优先级：

 1. 若设置 `AGENT_ENV` / `AGENT_USER` / `AGENT_INSTANCE` 且全部存在，则直接使用这些值。
-2. 否则按历史约定从主机名解析 `env-user-instance` 前缀。
-3. 如果两者都无法得到完整结果，Agent 启动会失败并提示需要提供上述环境变量。
+2. 否则检查历史 `node.json`（注册成功后由 Master 返回的信息），若包含 `env` / `user` / `instance` 则沿用。
+3. 若以上均不可用，则按历史约定从主机名解析 `env-user-instance` 前缀。
+4. 如果仍无法得到完整结果，Agent 启动会失败并提示需要提供上述环境变量。
+
+> 提示：在首次部署时需确保环境变量或主机名能够提供完整信息。完成注册后，Agent 会把 Master 返回的元数据写入 `node.json`，后续重启无需再次提供环境变量就能保持一致性。

 派生路径：

--- a/src/agent/app/config.py
+++ b/src/agent/app/config.py
@ -6,10 +6,14 @@ from dataclasses import dataclass
 from pathlib import Path
 from typing import Final

+from .state import load_node_state
 from .version import VERSION
+from .log import get_logger

 DEFAULT_REPORT_INTERVAL_SECONDS: Final[int] = 60

+LOGGER = get_logger("argus.agent.config")
+

@dataclass(frozen=True)
 class AgentConfig:
@ -50,7 +54,28 @@ def _resolve_hostname() -> str:
    return os.environ.get("AGENT_HOSTNAME") or socket.gethostname()


-def _resolve_metadata_fields(hostname: str) -> tuple[str, str, str]:
+def _load_metadata_from_state(node_file: str) -> tuple[str, str, str] | None:
+    state = load_node_state(node_file)
+    if not state:
+        return None
+
+    meta = state.get("meta_data") or {}
+    env = meta.get("env") or state.get("env")
+    user = meta.get("user") or state.get("user")
+    instance = meta.get("instance") or state.get("instance")
+
+    if env and user and instance:
+        LOGGER.debug("Metadata resolved from node state", extra={"node_file": node_file})
+        return env, user, instance
+
+    LOGGER.warning(
+        "node.json missing metadata fields; ignoring",
+        extra={"node_file": node_file, "meta_data": meta},
+    )
+    return None
+
+
+def _resolve_metadata_fields(hostname: str, node_file: str) -> tuple[str, str, str]:
    env = os.environ.get("AGENT_ENV")
    user = os.environ.get("AGENT_USER")
    instance = os.environ.get("AGENT_INSTANCE")
@ -59,23 +84,18 @@ def _resolve_metadata_fields(hostname: str) -> tuple[str, str, str]:
        return env, user, instance

    if any([env, user, instance]):
-        LOGGER = None
-        try:
-            from .log import get_logger
+        LOGGER.warning(
+            "Incomplete metadata environment variables; falling back to persisted metadata",
+            extra={
+                "has_env": bool(env),
+                "has_user": bool(user),
+                "has_instance": bool(instance),
+            },
+        )

-            LOGGER = get_logger("argus.agent.config")
-        except Exception:  # pragma: no cover - defensive
-            LOGGER = None
-        if LOGGER is not None:
-            LOGGER.warning(
-                "Incomplete metadata environment variables; falling back to hostname parsing",
-                extra={
-                    "has_env": bool(env),
-                    "has_user": bool(user),
-                    "has_instance": bool(instance),
-                },
-            )
-        env = user = instance = None
+    state_metadata = _load_metadata_from_state(node_file)
+    if state_metadata is not None:
+        return state_metadata

    from .collector import _parse_hostname  # Local import to avoid circular dependency

@ -93,9 +113,9 @@ def load_config() -> AgentConfig:
    """从环境变量推导配置，移除了外部配置文件依赖。"""

    hostname = _resolve_hostname()
-    environment, user, instance = _resolve_metadata_fields(hostname)
-
    node_file = f"/private/argus/agent/{hostname}/node.json"
+    environment, user, instance = _resolve_metadata_fields(hostname, node_file)
+
    health_dir = f"/private/argus/agent/{hostname}/health/"

    master_endpoint_env = os.environ.get("MASTER_ENDPOINT")
--- a/src/agent/dist/argus-agent
+++ b/src/agent/dist/argus-agent
--- a/src/agent/tests/scripts/07_restart_agent_and_reregister.sh
+++ b/src/agent/tests/scripts/07_restart_agent_and_reregister.sh
@ -6,9 +6,18 @@ TEST_ROOT="$(cd "$SCRIPT_DIR/.." && pwd)"
 TMP_ROOT="$TEST_ROOT/tmp"
 API_BASE="http://localhost:32300/api/v1/master"
 NODE_ID="$(cat "$TMP_ROOT/node_id")"
+ENV_NODE_ID_FILE="$TMP_ROOT/node_id_host_abc"
+if [[ ! -f "$ENV_NODE_ID_FILE" ]]; then
+  echo "[ERROR] Env agent node id file missing at $ENV_NODE_ID_FILE" >&2
+  exit 1
+fi
+
+ENV_NODE_ID="$(cat "$ENV_NODE_ID_FILE")"
 AGENT_HOSTNAME="dev-e2euser-e2einst-pod-0"
+ENV_AGENT_HOSTNAME="host_abc"
 NETWORK_NAME="tests_default"
 NEW_AGENT_IP="172.28.0.200"
+NEW_ENV_AGENT_IP="172.28.0.210"
 ENTRYPOINT_SCRIPT="$SCRIPT_DIR/agent_entrypoint.sh"
 VERIFY_SCRIPT="$TEST_ROOT/../scripts/agent_deployment_verify.sh"
 ENV_FILE="$TEST_ROOT/.env"
@ -80,15 +89,37 @@ if [[ "$prev_ip" != "$initial_ip" ]]; then
  exit 1
 fi

+env_before_file="$TMP_ROOT/env_before_restart.json"
+curl -sS "$API_BASE/nodes/$ENV_NODE_ID" -o "$env_before_file"
+env_prev_last_updated=$(python3 - "$env_before_file" <<'PY'
+import json, sys
+with open(sys.argv[1]) as handle:
+    node = json.load(handle)
+print(node.get("last_updated", ""))
+PY
+)
+env_prev_ip=$(python3 - "$env_before_file" <<'PY'
+import json, sys
+with open(sys.argv[1]) as handle:
+    node = json.load(handle)
+print(node["meta_data"].get("ip", ""))
+PY
+)
+
 pushd "$TEST_ROOT" >/dev/null
 compose rm -sf agent
+compose rm -sf agent_env
 popd >/dev/null

 docker container rm -f argus-agent-e2e >/dev/null 2>&1 || true
+docker container rm -f argus-agent-env-e2e >/dev/null 2>&1 || true

 AGENT_DIR="$TEST_ROOT/private/argus/agent/$AGENT_HOSTNAME"
 HEALTH_DIR="$TEST_ROOT/private/argus/agent/$AGENT_HOSTNAME/health"

+ENV_AGENT_DIR="$TEST_ROOT/private/argus/agent/$ENV_AGENT_HOSTNAME"
+ENV_HEALTH_DIR="$TEST_ROOT/private/argus/agent/$ENV_AGENT_HOSTNAME/health"
+
 # 先以 sleep 方式启动容器，确保我们掌握注册时的网络状态
 if ! docker run -d \
  --name argus-agent-e2e \
@ -148,3 +179,76 @@ if [[ "$success" != true ]]; then
 fi

 echo "[INFO] Agent restart produced successful re-registration with IP change"
+
+# ---- Restart env-driven agent without metadata environment variables ----
+
+if [[ ! -d "$ENV_AGENT_DIR" ]]; then
+  echo "[ERROR] Env agent data dir missing at $ENV_AGENT_DIR" >&2
+  exit 1
+fi
+
+if [[ ! -d "$ENV_HEALTH_DIR" ]]; then
+  mkdir -p "$ENV_HEALTH_DIR"
+fi
+
+if ! docker run -d \
+  --name argus-agent-env-e2e \
+  --hostname "$ENV_AGENT_HOSTNAME" \
+  --network "$NETWORK_NAME" \
+  --ip "$NEW_ENV_AGENT_IP" \
+  -v "$ENV_AGENT_DIR:/private/argus/agent/$ENV_AGENT_HOSTNAME" \
+  -v "$ENV_HEALTH_DIR:/private/argus/agent/$ENV_AGENT_HOSTNAME/health" \
+  -v "$TEST_ROOT/private/argus/etc:/private/argus/etc" \
+  -v "$AGENT_BINARY:/usr/local/bin/argus-agent:ro" \
+  -v "$ENTRYPOINT_SCRIPT:/usr/local/bin/agent-entrypoint.sh:ro" \
+  -v "$VERIFY_SCRIPT:/usr/local/bin/agent_deployment_verify.sh:ro" \
+  -e MASTER_ENDPOINT=http://master.argus.com:3000 \
+  -e REPORT_INTERVAL_SECONDS=2 \
+  -e ARGUS_BUILD_UID="$AGENT_UID" \
+  -e ARGUS_BUILD_GID="$AGENT_GID" \
+  --entrypoint /usr/local/bin/agent-entrypoint.sh \
+  ubuntu:22.04 >/dev/null; then
+  echo "[ERROR] Failed to start env-driven agent container without metadata env" >&2
+  exit 1
+fi
+
+env_success=false
+env_detail_file="$TMP_ROOT/env_post_restart.json"
+for _ in {1..20}; do
+  sleep 3
+  if ! curl -sS "$API_BASE/nodes/$ENV_NODE_ID" -o "$env_detail_file"; then
+    continue
+  fi
+  if python3 - "$env_detail_file" "$env_prev_last_updated" "$ENV_NODE_ID" "$env_prev_ip" "$NEW_ENV_AGENT_IP" <<'PY'
+import json, sys
+with open(sys.argv[1]) as handle:
+    node = json.load(handle)
+prev_last_updated = sys.argv[2]
+expected_id = sys.argv[3]
+old_ip = sys.argv[4]
+expected_ip = sys.argv[5]
+last_updated = node.get("last_updated")
+current_ip = node["meta_data"].get("ip")
+meta = node.get("meta_data", {})
+assert node["id"] == expected_id
+if current_ip != expected_ip:
+    raise SystemExit(1)
+if current_ip == old_ip:
+    raise SystemExit(1)
+if not last_updated or last_updated == prev_last_updated:
+    raise SystemExit(1)
+if meta.get("env") != "prod" or meta.get("user") != "ml" or meta.get("instance") != "node-3":
+    raise SystemExit(1)
+PY
+  then
+    env_success=true
+    break
+  fi
+done
+
+if [[ "$env_success" != true ]]; then
+  echo "[ERROR] Env-driven agent did not reuse persisted metadata after restart" >&2
+  exit 1
+fi
+
+echo "[INFO] Env-driven agent restart succeeded with persisted metadata"
--- a/src/agent/tests/test_config_metadata.py
+++ b/src/agent/tests/test_config_metadata.py
@ -60,6 +60,24 @@ class LoadConfigMetadataTests(unittest.TestCase):
        self.assertEqual(config.instance, "abc")
        mock_mkdir.assert_called()

+    @patch("app.config._load_metadata_from_state", return_value=("prod", "ops", "node-1"))
+    @patch("app.config.Path.mkdir")
+    def test_metadata_from_node_state(self, mock_mkdir, mock_state):
+        with temp_env(
+            MASTER_ENDPOINT="http://master.local",
+            AGENT_HOSTNAME="host_abc",
+            AGENT_ENV=None,
+            AGENT_USER=None,
+            AGENT_INSTANCE=None,
+        ):
+            config = load_config()
+
+        self.assertEqual(config.environment, "prod")
+        self.assertEqual(config.user, "ops")
+        self.assertEqual(config.instance, "node-1")
+        mock_state.assert_called_once()
+        mock_mkdir.assert_called()
+
    @patch("app.config.Path.mkdir")
    def test_partial_environment_variables_fallback(self, mock_mkdir):
        with temp_env(