kalavai-client 0.5.21__tar.gz → 0.5.22__tar.gz

{kalavai_client-0.5.21 → kalavai_client-0.5.22}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: kalavai-client
-Version: 0.5.21
+Version: 0.5.22
 Summary: Client app for kalavai platform
 License: Apache-2.0
 Keywords: LLM,platform

kalavai_client-0.5.22/kalavai_client/__init__.py

@@ -0,0 +1,2 @@
+
+__version__ = "0.5.22"

{kalavai_client-0.5.21 → kalavai_client-0.5.22}/kalavai_client/assets/apps.yaml

@@ -139,7 +139,7 @@ releases:
     - name: replicas
       value: 2
     - name: image_tag
-      value: "v2025.01.10"
+      value: "v2025.03.1"
     - name: deployment.in_cluster
       value: "True"
     - name: deployment.use_auth_key

{kalavai_client-0.5.21 → kalavai_client-0.5.22}/kalavai_client/assets/docker-compose-template.yaml

@@ -103,7 +103,7 @@ services:
   {% endif %}
 {% endif %}
 
-networks:
-  custom-network:
-    driver: bridge
+# networks:
+#   custom-network:
+#     driver: bridge
 

{kalavai_client-0.5.21 → kalavai_client-0.5.22}/kalavai_client/bridge_api.py

@@ -64,7 +64,8 @@ def pool_join(request: JoinPoolRequest):
     result = join_pool(
         token=request.token,
         num_gpus=request.num_gpus,
-        node_name=request.node_name
+        node_name=request.node_name,
+        ip_address=request.ip_address
     )
     return result
 

{kalavai_client-0.5.21 → kalavai_client-0.5.22}/kalavai_client/bridge_models.py

@@ -17,6 +17,7 @@ class DeleteNodesRequest(BaseModel):
 
 class JoinPoolRequest(BaseModel):
     token: str
+    ip_address: str = None
     node_name: str = None
     num_gpus: int = None
 

{kalavai_client-0.5.21 → kalavai_client-0.5.22}/kalavai_client/cli.py

@@ -511,11 +511,16 @@ def pool__join(token, *others, node_name=None):
         console.log("[red]Installation aborted")
         return
     
+    # select IP address (for external discovery)
+    console.log(f"Scanning for valid IPs")
+    ip_address = select_ip_address()
+    
     console.log("Connecting worker to the pool...")
     result = join_pool(
         token=token,
         node_name=node_name,
-        num_gpus=num_gpus
+        num_gpus=num_gpus,
+        ip_address=ip_address
     )
     if "error" in result:
         console.log(f"[red]Error when connecting: {result}")
@@ -1046,7 +1051,7 @@ def job__run(template_name, *others, values: str=None, force_namespace: str=None
     )
 
     if "error" in result:
-        console.log(f"[red]Error when deploying job: {str(e)}")
+        console.log(f"[red]Error when deploying job: {str(result['error'])}")
     else:
         console.log(f"[green]{template_name} job deployed")
 

{kalavai_client-0.5.21 → kalavai_client-0.5.22}/kalavai_client/core.py

@@ -267,6 +267,8 @@ def fetch_job_details(jobs: list[Job]):
                 if ns != namespace: # same job name, different namespace
                     continue
                 for _, values in ss.items():
+                    # TODO get 'restart_count' from values['conditions'][-1]["restart_count"]
+                    # TODO: get nodes involved in deployment (needs kubewatcher)
                     workers_status[values["status"]] += 1
             workers = "\n".join([f"{k}: {v}" for k, v in workers_status.items()])
             # get URL details
@@ -287,10 +289,12 @@ def fetch_job_details(jobs: list[Job]):
             urls = [f"http://{load_server_info(data_key=SERVER_IP_KEY, file=USER_LOCAL_SERVER_FILE)}:{node_port}" for node_port in node_ports]
             if "Ready" in workers_status and len(workers_status) == 1:
                 status = "running"
-            elif any([st in workers_status for st in ["Failed", "Completed"]]):
+            elif any([st in workers_status for st in ["Failed"]]):
                 status = "error"
-            elif any([st in workers_status for st in ["Pending"]]):
+            elif any([st in workers_status for st in ["Pending"]]) or len(workers_status) == 0:
                 status = "pending"
+            elif any([st in workers_status for st in ["Succeeded", "Completed"]]):
+                status = "completed"
             else:
                 status = "working"
             job_details.append(
@@ -569,7 +573,7 @@ def get_max_gpus():
     except:
         return 0
 
-def join_pool(token, num_gpus=None, node_name=None):
+def join_pool(token, num_gpus=None, node_name=None, ip_address=None):
     compatibility = check_worker_compatibility()
     if len(compatibility["issues"]) > 0:
         return {"error": compatibility["issues"]}
@@ -626,6 +630,7 @@ def join_pool(token, num_gpus=None, node_name=None):
     # Generate docker compose recipe
     generate_compose_config(
         role="agent",
+        node_ip_address=ip_address,
         pool_ip=f"https://{kalavai_seed_ip}:6443",
         pool_token=kalavai_token,
         num_gpus=num_gpus,
@@ -701,6 +706,9 @@ def create_pool(cluster_name: str, ip_address: str, app_values: str=None, pool_c
             node_labels[USER_NODE_LABEL] = user["username"]
         except Exception as e:
             return {"error": f"[red]Error when joining network: {str(e)}"}
+        
+    if num_gpus is None:
+        num_gpus = get_max_gpus()
     
     # Generate docker compose recipe
     generate_compose_config(

{kalavai_client-0.5.21 → kalavai_client-0.5.22}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name            = "kalavai-client"
-version         = "0.5.21"
+version         = "0.5.22"
 authors = [
   {name = "Carlos Fernandez Musoles", email = "carlos@kalavai.net"}
 ]

kalavai_client-0.5.21/kalavai_client/__init__.py

@@ -1,2 +0,0 @@
-
-__version__ = "0.5.21"