NAS-130995 / 25.04 / Allow retrieving logs of apps in crashed state

src/middlewared/middlewared/plugins/apps/crud.py

@@ -18,6 +18,7 @@
 from .ix_apps.path import get_installed_app_path, get_installed_app_version_path
 from .ix_apps.query import list_apps
 from .ix_apps.setup import setup_install_app_dir
+from .ix_apps.utils import AppState
 from .version_utils import get_latest_version_from_app_versions
 
 
@@ -33,7 +34,7 @@ class Config:
         'app_entry',
         Str('name'),
         Str('id'),
-        Str('state', enum=['STOPPED', 'DEPLOYING', 'RUNNING']),
+        Str('state', enum=[state.value for state in AppState]),
         Bool('upgrade_available'),
         Str('human_version'),
         Str('version'),

src/middlewared/middlewared/plugins/apps/ix_apps/query.py

@@ -6,7 +6,7 @@
 from .metadata import get_collective_config, get_collective_metadata
 from .lifecycle import get_current_app_config
 from .path import get_app_parent_config_path
-from .utils import get_app_name_from_project_name, normalize_reference, PROJECT_PREFIX
+from .utils import AppState, ContainerState, get_app_name_from_project_name, normalize_reference, PROJECT_PREFIX
 
 
 COMPOSE_SERVICE_KEY: str = 'com.docker.compose.service'
@@ -86,12 +86,18 @@ def list_apps(
         # When we stop docker service and start it again - the containers can be in exited
         # state which means we need to account for this.
         state = 'STOPPED'
+        exited_containers = 0
         for container in workloads['container_details']:
-            if container['state'] == 'starting':
-                state = 'DEPLOYING'
+            if container['state'] == ContainerState.STARTING.value:
+                state = AppState.DEPLOYING.value
                 break
-            elif container['state'] == 'running':
-                state = 'RUNNING'
+            elif container['state'] == ContainerState.RUNNING.value:
+                state = AppState.RUNNING.value
+            elif container['state'] == ContainerState.EXITED.value:
+                exited_containers += 1
+        else:
+            if exited_containers != 0 and exited_containers == len(workloads['container_details']):
+                state = AppState.CRASHED.value
 
         app_metadata = metadata[app_name]
         active_workloads = get_default_workload_values() if state == 'STOPPED' else workloads
@@ -127,7 +133,7 @@ def list_apps(
                 'name': entry.name,
                 'id': entry.name,
                 'active_workloads': get_default_workload_values(),
-                'state': 'STOPPED',
+                'state': AppState.STOPPED.value,
                 'upgrade_available': upgrade_available_for_app(train_to_apps_version_mapping, app_metadata),
                 'image_updates_available': False,
                 **app_metadata | {'portals': normalize_portal_uris(app_metadata['portals'], host_ip)}
@@ -187,9 +193,12 @@ def translate_resources_to_desired_workflow(app_resources: dict) -> dict:
 
         if container['State']['Status'].lower() == 'running':
             if health_config := container['State'].get('Health'):
-                state = 'running' if health_config['Status'] == 'healthy' else 'starting'
+                if health_config['Status'] == 'healthy':
+                    state = ContainerState.RUNNING.value
+                else:
+                    state = ContainerState.STARTING.value
             else:
-                state = 'running'
+                state = ContainerState.RUNNING.value
         else:
             state = 'exited'
 

src/middlewared/middlewared/plugins/apps/ix_apps/utils.py

@@ -1,8 +1,23 @@
+import enum
+
 from catalog_reader.library import RE_VERSION  # noqa
 from middlewared.plugins.apps_images.utils import normalize_reference  # noqa
 from middlewared.plugins.apps.schema_utils import CONTEXT_KEY_NAME  # noqa
 from middlewared.plugins.apps.utils import IX_APPS_MOUNT_PATH, PROJECT_PREFIX, run  # noqa
 
 
+class AppState(enum.Enum):
+    CRASHED = 'CRASHED'
+    DEPLOYING = 'DEPLOYING'
+    RUNNING = 'RUNNING'
+    STOPPED = 'STOPPED'
+
+
+class ContainerState(enum.Enum):
+    EXITED = 'exited'
+    RUNNING = 'running'
+    STARTING = 'starting'
+
+
 def get_app_name_from_project_name(project_name: str) -> str:
     return project_name[len(PROJECT_PREFIX):]

src/middlewared/middlewared/plugins/apps/logs.py

@@ -8,6 +8,7 @@
 from middlewared.service import CallError
 from middlewared.validators import Range
 
+from .ix_apps.utils import AppState
 from .ix_apps.docker.utils import get_docker_client
 
 
@@ -38,8 +39,8 @@ def __init__(self, *args, **kwargs):
 
     def validate_log_args(self, app_name, container_id):
         app = self.middleware.call_sync('app.get_instance', app_name)
-        if app['state'] not in ('RUNNING', 'DEPLOYING'):
-            raise CallError(f'App "{app_name}" is not running')
+        if app['state'] not in (AppState.CRASHED.value, AppState.RUNNING.value, AppState.DEPLOYING.value):
+            raise CallError(f'Unable to retrieve logs of stopped {app_name!r} app')
 
         if not any(c['id'] == container_id for c in app['active_workloads']['container_details']):
             raise CallError(f'Container "{container_id}" not found in app "{app_name}"', errno=errno.ENOENT)

src/middlewared/middlewared/plugins/apps/resources.py

@@ -3,6 +3,7 @@
 
 from middlewared.utils.gpu import get_nvidia_gpus
 
+from .ix_apps.utils import ContainerState
 from .resources_utils import get_normalized_gpu_choices
 
 
@@ -43,7 +44,9 @@ async def container_ids(self, app_name, options):
                 'id': c['id'],
             } for c in (
                 await self.middleware.call('app.get_instance', app_name)
-            )['active_workloads']['container_details'] if (options['alive_only'] is False or c['state'] == 'running')
+            )['active_workloads']['container_details'] if (
+                options['alive_only'] is False or ContainerState(c['state']) == ContainerState.RUNNING
+            )
         }
 
     @accepts(Str('app_name'), roles=['APPS_READ'])