Merge pull request #1997 from DSD-DBS/custom-promtail-conf

feat: Add custom promtail / logging configuration for tools

backend/capellacollab/sessions/hooks/log_collector.py

@@ -9,6 +9,7 @@
 from capellacollab.config import config
 from capellacollab.sessions import models as sessions_models
 from capellacollab.sessions.operators import models as operators_models
+from capellacollab.tools import models as tools_models
 from capellacollab.users.workspaces import crud as users_workspaces_crud
 
 from . import interface
@@ -23,10 +24,7 @@ def post_session_creation_hook(
         self,
         request: interface.PostSessionCreationHookRequest,
     ) -> interface.PostSessionCreationHookResult:
-        if (
-            not self._loki_enabled
-            or request.db_session.type == sessions_models.SessionType.READONLY
-        ):
+        if not self._log_collection_enabled(request.db_session):
             return interface.PostSessionCreationHookResult()
 
         workspaces = users_workspaces_crud.get_workspaces_for_user(
@@ -40,7 +38,7 @@ def post_session_creation_hook(
             data=self._promtail_configuration(
                 username=request.user.name,
                 session_type=request.db_session.type.value,
-                tool_name=request.db_session.tool.name,
+                tool=request.db_session.tool,
                 version_name=request.db_session.version.name,
             ),
         )
@@ -62,7 +60,7 @@ def post_session_creation_hook(
             operators_models.PersistentVolume(
                 name="workspace",
                 read_only=False,
-                container_path=pathlib.PurePosixPath("/var/log/promtail"),
+                container_path=pathlib.PurePosixPath("/workspace"),
                 volume_name=workspaces[0].pvc_name,
             ),
         ]
@@ -84,23 +82,29 @@ def pre_session_termination_hook(
         self,
         request: interface.PreSessionTerminationHookRequest,
     ) -> interface.PreSessionTerminationHookResult:
-        if (
-            not self._loki_enabled
-            or request.session.type == sessions_models.SessionType.READONLY
-        ):
+        if not self._log_collection_enabled(request.session):
             return interface.PostSessionCreationHookResult()
 
         request.operator._delete_config_map(name=request.session.id)
         request.operator._delete_pod(name=f"{request.session.id}-promtail")
 
         return interface.PreSessionTerminationHookResult()
 
+    def _log_collection_enabled(
+        self, session: sessions_models.DatabaseSession
+    ) -> bool:
+        return (
+            self._loki_enabled
+            and session.type == sessions_models.SessionType.PERSISTENT
+            and session.tool.config.monitoring.logging.enabled
+        )
+
     @classmethod
     def _promtail_configuration(
         cls,
         username: str,
         session_type: str,
-        tool_name: str,
+        tool: tools_models.DatabaseTool,
         version_name: str,
     ) -> dict:
         cfg = config.k8s.promtail
@@ -121,7 +125,7 @@ def _promtail_configuration(
                         }
                     ],
                     "positions": {
-                        "filename": "/var/log/promtail/positions.yaml"
+                        "filename": f"/workspace/.promtail/positions-tool-{tool.id}.yaml"
                     },
                     "scrape_configs": [
                         {
@@ -139,9 +143,9 @@ def _promtail_configuration(
                                     "labels": {
                                         "username": username,
                                         "session_type": session_type,
-                                        "tool": tool_name,
+                                        "tool": tool.name,
                                         "version": version_name,
-                                        "__path__": "/var/log/promtail/**/*.log",
+                                        "__path__": tool.config.monitoring.logging.path,
                                     },
                                 }
                             ],

backend/capellacollab/tools/models.py

@@ -15,6 +15,7 @@
 from sqlalchemy import orm
 
 from capellacollab import core
+from capellacollab.config import config
 from capellacollab.core import database
 from capellacollab.core import pydantic as core_pydantic
 from capellacollab.core.database import decorator
@@ -246,11 +247,29 @@ class PrometheusConfiguration(core_pydantic.BaseModel):
     path: str = pydantic.Field(default="/prometheus")
 
 
+class LoggingConfiguration(core_pydantic.BaseModel):
+    """Side-car container to push logs to Grafana Loki"""
+
+    enabled: bool = pydantic.Field(
+        default=config.k8s.promtail.loki_enabled,
+        description="If enabled, logs will be pushed to Grafana Loki.",
+    )
+
+    path: str = pydantic.Field(
+        default="/workspace/**/*.log",
+        description="Path to the log files, can be a glob string.",
+    )
+
+
 class SessionMonitoring(core_pydantic.BaseModel):
     prometheus: PrometheusConfiguration = pydantic.Field(
         default=PrometheusConfiguration(),
         description="Configuration for monitoring and garbage collection.",
     )
+    logging: LoggingConfiguration = pydantic.Field(
+        default=LoggingConfiguration(),
+        description="Configuration for side-car logging container.",
+    )
 
 
 class ToolModelProvisioning(core_pydantic.BaseModel):

backend/capellacollab/tools/routes.py

@@ -8,9 +8,11 @@
 
 import capellacollab.projects.toolmodels.crud as projects_models_crud
 import capellacollab.settings.modelsources.t4c.instance.crud as settings_t4c_crud
+from capellacollab.config import config
 from capellacollab.core import database
 from capellacollab.core import exceptions as core_exceptions
 from capellacollab.core.authentication import injectables as auth_injectables
+from capellacollab.core.logging import exceptions as logging_exceptions
 from capellacollab.users import models as users_models
 
 from . import crud, exceptions, injectables, models
@@ -66,6 +68,12 @@ def create_tool(
     To use this route, the user role `administrator` is required.
     """
 
+    if (
+        body.config.monitoring.logging.enabled
+        and not config.k8s.promtail.loki_enabled
+    ):
+        raise logging_exceptions.GrafanaLokiDisabled()
+
     return crud.create_tool(db, body)
 
 
@@ -85,6 +93,12 @@ def update_tool(
     tool: models.DatabaseTool = fastapi.Depends(injectables.get_existing_tool),
     db: orm.Session = fastapi.Depends(database.get_db),
 ) -> models.DatabaseTool:
+    if (
+        body.config.monitoring.logging.enabled
+        and not config.k8s.promtail.loki_enabled
+    ):
+        raise logging_exceptions.GrafanaLokiDisabled()
+
     return crud.update_tool(db, tool, body)
 
 

frontend/src/storybook/tool.ts

@@ -83,6 +83,10 @@ const defaultToolConfig: ToolSessionConfigurationOutput = {
     prometheus: {
       path: '/metrics',
     },
+    logging: {
+      enabled: true,
+      path: '/workspace/*.log',
+    },
   },
   supported_project_types: ['general', 'training'],
 };