agent options check

Dockerfile

@@ -46,5 +46,4 @@ COPY agent /workdir/agent
 #ENV PYTHONPATH /workdir:/workdir/src:/workdir/supervisely_lib/worker_proto:$PYTHONPATH
 WORKDIR /workdir/agent
 
-ENTRYPOINT ["sh", "-c", "python -u /workdir/agent/main.py"]
-
+ENTRYPOINT ["python", "-u", "/workdir/agent/main.py"]

agent/main.py

@@ -213,12 +213,22 @@ def init_envs():
     try:
         agent_utils.check_instance_version()
         new_envs, new_volumes, ca_cert = agent_utils.updated_agent_options()
-    except agent_utils.AgentOptionsNotAvailable:
-        sly.logger.debug("Can not update agent options", exc_info=True)
-        sly.logger.warning(
-            "Can not update agent options. Agent will be started with current options"
-        )
-        return
+    except Exception as e:
+        if not agent_utils.is_agent_container_ready_to_continue():
+            sly.logger.error(
+                "Agent options are not available. Agent will be stopped. Please, check the connection to the server"
+            )
+            raise
+
+        if isinstance(e, agent_utils.AgentOptionsNotAvailable):
+            sly.logger.debug("Can not update agent options", exc_info=True)
+            sly.logger.warning(
+                "Can not update agent options. Agent will be started with current options"
+            )
+            return
+
+        raise
+
     if new_envs.get(constants._FORCE_CPU_ONLY, "false") == "true":
         runtime = "runc"
         runtime_changed = _is_runtime_changed(runtime)

agent/worker/agent_utils.py

@@ -623,7 +623,7 @@ def get_agent_options(server_address=None, token=None, timeout=60) -> dict:
         method,
         data={"token": token},
     )
-    if resp.status_code != requests.codes.ok:  # pylint: disable=no-member
+    if resp is None or resp.status_code != requests.codes.ok:  # pylint: disable=no-member
         try:
             text = resp.text
         except:
@@ -641,8 +641,8 @@ def get_instance_version(server_address=None, timeout=60):
 
     api = sly.Api(server_address=server_address)
     resp = api.get("instance.version", {})
-    if resp.status_code != requests.codes.ok:  # pylint: disable=no-member
-        if resp.status_code in (400, 401, 403, 404):
+    if resp is None or resp.status_code != requests.codes.ok:  # pylint: disable=no-member
+        if resp is not None and resp.status_code in (400, 401, 403, 404):
             return None
         try:
             text = resp.text
@@ -934,6 +934,20 @@ def _ca_cert_changed(ca_cert) -> str:
     return cert_path
 
 
+def is_agent_container_ready_to_continue():
+    container_info = get_container_info()
+    volumes = binds_to_volumes_dict(container_info.get("HostConfig", {}).get("Binds", []))
+
+    # should contain at least 3 volumes:
+    # docker socket
+    # agent data files
+    # apps data files
+    if len(volumes) < 3:
+        return False
+
+    return True
+
+
 def get_options_changes(envs: dict, volumes: dict, ca_cert: str) -> Tuple[dict, dict, str]:
     return _envs_changes(envs), _volumes_changes(volumes), _ca_cert_changed(ca_cert)