dts-dance 0.2.6__py3-none-any.whl → 0.2.8__py3-none-any.whl

{dts_dance-0.2.6.dist-info → dts_dance-0.2.8.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dts-dance
-Version: 0.2.6
+Version: 0.2.8
 Summary: dts dance lib
 Keywords: observation,tools
 Requires-Python: >=3.12

{dts_dance-0.2.6.dist-info → dts_dance-0.2.8.dist-info}/RECORD

@@ -1,7 +1,7 @@
 dtsdance/__init__.py,sha256=Yl_jEZ5weYfcrklnDvwB4wSgCOvMBLRRgWx0gHs3qfM,49
 dtsdance/bytecloud.py,sha256=LZ-uw5NMmapyNpt89bMKhCgv4U2cGR-XjLzFFFEXGxM,5899
 dtsdance/ddutil.py,sha256=1ZUmW11nw8WW1NfAIRbjuKy8zeQVTFvoXl6QY3o4YJ0,1534
-dtsdance/dflow.py,sha256=jwsngCqgHUSAUFh-IUgq-fnQs8lAvHtj7uLLs_tupqI,7718
+dtsdance/dflow.py,sha256=v6tQv9O6-xjXhNuU14vys0lIxi4hKKv27nuBjLgRk6Y,7747
 dtsdance/dsyncer.py,sha256=PCatQIFXgi1l2qT60zNN-ALAqtv7_sg-D2oNMB_t8qw,9906
 dtsdance/feishu_base.py,sha256=dYYNnsCjGZYN2I6inR9FRQoUiDLB7j40rzydw50N38o,11825
 dtsdance/feishu_table.py,sha256=JleCjQJ-rcbC15v2UAc5BK-YioB2SMGsdGztQ9BaT4k,10779
@@ -10,6 +10,7 @@ dtsdance/s3.py,sha256=Bh-cwLksfO5PewNtIzE_Md3rRLDLI1DUVoOD7Pou5T8,1294
 dtsdance/spacex.py,sha256=W_YGt_Wo_Cy4eDhgXhNfskayPHcZ1kg2Mil2ghI_VLE,4791
 dtsdance/tcc_inner.py,sha256=QqVyhwP0kCdp1HFHFgtme9o0im-eQNn1HVGdLt3Xr8g,1627
 dtsdance/tcc_open.py,sha256=JIfVhL3yfm0ftS_vzONE2QNzfqcZuX1mPTvjovAtwmE,6893
-dts_dance-0.2.6.dist-info/METADATA,sha256=cUaSHfyZVfa3EzpzyvyIJuiot0s1e166G3Omzzk3vMU,826
-dts_dance-0.2.6.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
-dts_dance-0.2.6.dist-info/RECORD,,
+dtsdance/tce.py,sha256=QwqPimZu0GKeP--sdbMP9h1e22nGuZL2FHqBUykXDOw,11563
+dts_dance-0.2.8.dist-info/METADATA,sha256=YjipvJ24oD--koIvidnlXrhll5UzI5Efpl-Zog_kQZc,826
+dts_dance-0.2.8.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
+dts_dance-0.2.8.dist-info/RECORD,,

dtsdance/dflow.py

@@ -117,12 +117,13 @@ class DFlowClient:
         site_info = self.bytecloud_client.get_site_config(site)
         return f"{site_info.endpoint}/bytedts/datasync/detail/{task_id}"
 
-    def init_resources(self, site: str, ctrl_env: str) -> bool:
+    def init_resources(self, site: str, vregion: str, ctrl_env: str) -> bool:
         """
         初始化 CTRL 环境资源
 
         Args:
             site: 站点名称
+            vregion: 虚拟区域
             ctrl_env: 控制环境
 
         Returns:
@@ -135,21 +136,21 @@ class DFlowClient:
         # 构建请求数据
         payload = {"ctrl_env": ctrl_env}
 
-        response_data = make_request("POST", url, self.bytecloud_client.build_request_headers(site, None), payload)
+        response_data = make_request("POST", url, self.bytecloud_client.build_request_headers(site, vregion), payload)
 
         message = response_data.get("message")
         logger.info(f"int_resources {site} {ctrl_env}, message: {message}")
 
         return message == "ok"
 
-    def list_resources(self, site: str, ctrl_env: str, vregion: str | None = None) -> list[str]:
+    def list_resources(self, site: str, vregion: str, ctrl_env: str) -> list[str]:
         """
         列举 CTRL 环境资源列表，连同 agent 列表
 
         Args:
             site: 站点名称
-            ctrl_env: 控制环境
             vregion: 虚拟区域
+            ctrl_env: 控制环境
         Returns:
             list[str]: CTRL 环境资源列表
         """
@@ -164,7 +165,7 @@ class DFlowClient:
 
         # curl_cmd = output_curl(url, headers, payload)
         # print(f"Equivalent curl:\n{curl_cmd}")
-        
+
         try:
             data = response_data.get("data", {})
             items = data.get("items", [])

dtsdance/tce.py

@@ -0,0 +1,289 @@
+from typing import Any
+import requests
+
+from .bytecloud import ByteCloudClient
+
+
+class TCEError(Exception):
+    pass
+
+
+class TCEErrorServiceExist(Exception):
+    pass
+
+
+class TCEErrorClusterExist(Exception):
+    pass
+
+
+class TCEClient:
+    """Client for TCE deployment API."""
+
+    def __init__(self, bytecloud_client: ByteCloudClient) -> None:
+        """
+        Initialize TCE client.
+
+        Args:
+            base_url: TCE API base URL
+            jwt_token: JWT token for authentication (can be fetched automatically)
+        """
+        self.bytecloud_client = bytecloud_client
+
+    def create_service(self, site: str, ctrl_name: str, bytetree_parent_id: int) -> int | None:
+        site_info = self.bytecloud_client.get_site_config(site)
+        url = f"{site_info.endpoint}/api/v1/tce/open-apis/v1/deployment/service/create/"
+
+        service_info = {
+            "meta": {"name": f"bytedts.cm.{ctrl_name}", "psm": f"bytedts.cm.{ctrl_name}", "env": "prod"},
+            "auth": {
+                "galaxy_info": {"parent_id": bytetree_parent_id},
+                "approval_controls": ["reset"],
+                "iam_info": {"owners": [{"username": site_info.svc_account}, {"username": "mahang"}]},
+                "owner_name": site_info.svc_account,
+            },
+            "extra_features": {
+                "service_token": {
+                    "tech_stack": {"language": "go", "framework": "custom", "platform_types_v_2": {"tce": ["http"]}},
+                    "deploy": {"type": "stateless_compute", "scale": {"enable": True}, "descale": {"enable": True}},
+                    "level": "P2",
+                    "purpose": "convention",
+                    "contain_user_personal_info": False,
+                    "data_sensitivity_level": "L2",
+                }
+            },
+            "build": {
+                "dependency_mode": "default",
+                "base_image": "debian.bookworm.tce_service:latest",
+                "install_package_info": [],
+                "scm_repo_info": [
+                    {"name": "bytedts/dflow/dtscm", "remote_id": 387073, "main_repo": True, "path": "dts-cm"},
+                    {"name": "toutiao/load", "remote_id": 667, "path": "toutiao/load"},
+                    {"name": "toutiao/runtime", "remote_id": 631, "path": "toutiao/runtime"},
+                ],
+                "language": "go",
+                "framework_support": "none",
+                "tech_stack": "custom",
+            },
+            "runtime": {
+                "umount_ss_conf": False,
+                "new_ports": [
+                    {"intent": "primary", "port": 8088, "is_primary": True, "type": "tcp"},
+                    {"intent": "other", "port": 3033, "is_primary": False, "type": "tcp"},
+                ],
+                "is_runtime": False,
+                "app_path": "dts-cm/bootstrap.sh",
+                "enable_iast": False,
+            },
+        }
+
+        payload = {"service_info": service_info}
+
+        try:
+            response = requests.post(url, json=payload, headers=self.bytecloud_client.build_request_headers(site), timeout=60)
+            if response.status_code != 200:
+                if "already exists" in response.text:
+                    raise TCEErrorServiceExist("TCE cluster with the same name already exists")
+
+                raise TCEError(f"TCE API request failed with status {response.status_code}\n" f"Response: {response.text}")
+
+            result = response.json()
+            if result.get("code", 0) == -1:
+                raise TCEError(f"TCE API error: {result.get('error')}")
+
+            return result.get("data", {}).get("service", 0)
+
+        except requests.RequestException as e:
+            raise TCEError(f"Failed to create TCE cluster: {str(e)}") from e
+
+    def create_cluster(self, site: str, cluster_info: dict[str, Any], service: int) -> int | None:
+        """
+        Create TCE cluster.
+
+        Args:
+            cluster_info: Cluster configuration (meta, resource, runtime)
+            service: Service ID
+
+        Returns:
+            The created cluster ID, or None if creation failed
+
+        Raises:
+            TCEError: If API call fails
+        """
+        site_info = self.bytecloud_client.get_site_config(site)
+        url = f"{site_info.endpoint}/api/v1/tce/open-apis/v1/deployment/cluster/create/"
+        payload = {"cluster_info": cluster_info, "service": service}
+
+        try:
+            response = requests.post(url, json=payload, headers=self.bytecloud_client.build_request_headers(site), timeout=60)
+            if response.status_code != 200:
+                if "a same named TCE cluster already exists" in response.text:
+                    raise TCEErrorClusterExist("TCE cluster with the same name already exists")
+
+                raise TCEError(f"TCE API request failed with status {response.status_code}\n" f"Response: {response.text}")
+
+            result = response.json()
+
+            # Check for errors in response
+            if not isinstance(result, dict):
+                raise TCEError(f"Unexpected response format: {result}")
+
+            # TCE API may return error information in different formats
+            if result.get("code", 0) == -1:
+                raise TCEError(f"TCE API error: {result.get('error')}")
+
+            data = result.get("data", {})
+            create_success = data.get("deployment_type", "") == "create" and data.get("status", "") == "pending"
+            if not create_success:
+                raise TCEError(f"TCE cluster creation failed: {result}")
+
+            cluster_id = data.get("cluster_info", {}).get("meta", {}).get("id", 0)
+            # Ensure cluster_id is an integer
+            return int(cluster_id) if cluster_id else 0
+
+        except requests.RequestException as e:
+            raise TCEError(f"Failed to create TCE cluster: {str(e)}") from e
+
+    def delete_cluster(self, site: str, cluster_id: int, service_id: int, cluster_name: str) -> int | None:
+        """
+        Delete TCE cluster (reset/destroy).
+
+        Args:
+            site: Site identifier
+            cluster_id: Cluster ID to delete
+            service_id: Service ID
+            cluster_name: Cluster name
+
+        Returns:
+            Response data
+
+        Raises:
+            TCEError: If API call fails
+        """
+        site_info = self.bytecloud_client.get_site_config(site)
+        url = f"{site_info.endpoint}/api/v1/tce/open-apis/v1/deployment/cluster/reset/"
+        payload = {"cluster": cluster_id, "service": service_id}
+
+        # print(f"Deleting TCE cluster {cluster_id} for service {service_id}")
+        try:
+            response = requests.post(url, json=payload, headers=self.bytecloud_client.build_request_headers(site), timeout=60)
+            if response.status_code != 200:
+                raise TCEError(f"TCE API request failed with status {response.status_code}\n" f"Response: {response.text}")
+
+            result = response.json()
+            if result.get("code", 0) == -1:
+                error_msg = result.get("error", "")
+                if "do not exist" in error_msg:
+                    print(f"cluster {cluster_name} not exist, maybe already deleted")
+                    return None
+
+                raise TCEError(f"TCE API error: {error_msg}")
+
+            data = result.get("data", {})
+            # 查找 "人工确认" step 的 id
+            steps = data.get("steps", [])
+            for step in steps:
+                if step.get("step_type") == "ConfirmStep":
+                    return step.get("id")
+
+            return None
+
+        except requests.RequestException as e:
+            raise TCEError(f"Failed to delete TCE cluster: {cluster_name}, msg: {str(e)}") from e
+
+    def get_cluster_info(self, site: str, cluster_id: int) -> dict[str, Any]:
+        site_info = self.bytecloud_client.get_site_config(site)
+        url = f"{site_info.endpoint}/api/v1/tce/open-apis/v1/clusters/{cluster_id}/"
+
+        try:
+            response = requests.get(url, headers=self.bytecloud_client.build_request_headers(site), timeout=60)
+            if response.status_code != 200:
+                raise TCEError(f"TCE API request failed with status {response.status_code}\n" f"Response: {response.text}")
+
+            # print(f"get_cluster_info response.text: {response.text}")
+            result = response.json()
+            if result.get("code", 0) == -1:
+                raise TCEError(f"TCE API error: {result.get('error')}")
+
+            return result.get("data", {})
+
+        except requests.RequestException as e:
+            raise TCEError(f"Failed to create TCE cluster: {str(e)}") from e
+
+    def is_cluster_running(self, site: str, cluster_id: int | None) -> bool:
+        if cluster_id is None:
+            return False
+
+        cluster_info = self.get_cluster_info(site, cluster_id)
+        return cluster_info.get("meta", {}).get("status", "unknown") == "running"
+
+    def restart_cluster(self, site: str, service_id: int, cluster_id: int) -> list[int]:
+        site_info = self.bytecloud_client.get_site_config(site)
+        url = f"{site_info.endpoint}/api/v1/tce/open-apis/v1/deployment/cluster/pod_exec/"
+
+        payload = {
+            "service": service_id,
+            "cluster": cluster_id,
+            "exec_command": "rebuild",
+            "exec_params": {
+                "surge_percent": 25,
+                "ready_seconds": 10,
+                "timeout": 120,
+                "tolerance_num": 3,
+                "failed_policy": "remain",
+                "need_check": True,
+            },
+            "exec_type": "rebuild",
+            "parallel_mode": True,
+        }
+
+        try:
+            response = requests.post(url, json=payload, headers=self.bytecloud_client.build_request_headers(site), timeout=60)
+            if response.status_code != 200:
+                raise TCEError(f"TCE API request failed with status {response.status_code}\n" f"Response: {response.text}")
+
+            result = response.json()
+            if result.get("code", 0) == -1:
+                raise TCEError(f"TCE API error: {result.get('error')}")
+
+            steps = result.get("data", {}).get("steps", [])
+            step_ids = [step.get("id", 0) for step in steps]
+            return step_ids
+
+        except requests.RequestException as e:
+            raise TCEError(f"Failed to create TCE cluster: {str(e)}") from e
+
+    def confirm_step(self, site: str, step_id: int, action: str = "confirm") -> bool:
+        """
+        Confirm a step in TCE cluster deployment.
+
+        Args:
+            step_id: Step ID
+
+        Returns:
+            Response data
+
+        Raises:
+            TCEError: If API call fails
+        """
+        site_info = self.bytecloud_client.get_site_config(site)
+        url = f"{site_info.endpoint}/api/v1/tce/open-apis/v1/deployment_steps/{step_id}/actions/"
+        payload = {"action": action}
+
+        try:
+            response = requests.post(url, json=payload, headers=self.bytecloud_client.build_request_headers(site), timeout=60)
+            # print(f"Confirm step response: {response.text}")
+
+            response.raise_for_status()
+            result = response.json()
+
+            # Check for errors in response
+            if not isinstance(result, dict):
+                raise TCEError(f"Unexpected response format: {result}")
+
+            if result.get("code", 0) == -1:
+                raise TCEError(f"TCE API error: {result.get('error')}")
+
+            return True
+
+        except requests.RequestException as e:
+            raise TCEError(f"Failed to confirm ticket, msg: {str(e)}") from e