kalavai-client 0.5.27__py3-none-any.whl → 0.5.29__py3-none-any.whl

kalavai_client/__init__.py

@@ -1,2 +1,2 @@
 
-__version__ = "0.5.27"
+__version__ = "0.5.29"

kalavai_client/assets/apps.yaml

@@ -25,6 +25,19 @@ releases:
     chart: kalavai/kalavai-datashim
     version: "0.4.0"
     installed: false
+  - name: lago
+    namespace: kalavai
+    chart: kalavai/lago
+    installed: true
+    set:
+    - name: external.api.nodePort
+      value: 32000
+    - name: external.front.nodePort
+      value: 30080
+    - name: apiUrl
+      value: http://{{cluster_ip}}:32000
+    - name: frontUrl
+      value: http://{{cluster_ip}}:30080
   - name: minio
     needs: 
     - kalavai/longhorn
@@ -139,7 +152,7 @@ releases:
     - name: replicas
       value: 2
     - name: image_tag
-      value: "v2025.03.3"
+      value: "v2025.03.11"
     - name: deployment.in_cluster
       value: "True"
     - name: deployment.use_auth_key

kalavai_client/assets/docker-compose-gui.yaml

@@ -2,7 +2,26 @@ services:
   kalavai_gui:
     container_name: kalavai_gui
     image: bundenth/kalavai-gui:latest
-    network_mode: host
+    #network_mode: host
+    extra_hosts:
+    - "host.docker.internal:host-gateway"
+    networks:
+    - kalavai-net
+    environment:
+    - KALAVAI_BRIDGE_URL=http://host.docker.internal
+    - KALAVAI_BRIDGE_PORT={{bridge_port}}
+    entrypoint: ["reflex"]
+    command: >
+      run
+      --backend-port {{gui_backend_port}}
+      --frontend-port {{gui_frontend_port}}
+    ports:
+    - "{{gui_backend_port}}:{{gui_backend_port}}"
+    - "{{gui_frontend_port}}:{{gui_frontend_port}}"
     volumes:
       - "{{path}}:/root/.cache/kalavai"
     restart: always
+
+networks:
+  kalavai-net:
+    driver: bridge

kalavai_client/bridge_api.py

@@ -162,7 +162,8 @@ def job_deploy(request: DeployJobRequest):
     result = deploy_job(
         template_name=request.template_name,
         values_dict=request.values,
-        force_namespace=request.force_namespace
+        force_namespace=request.force_namespace,
+        target_labels=request.target_labels
     )
     return result
 

kalavai_client/bridge_models.py

@@ -37,6 +37,7 @@ class DeployJobRequest(BaseModel):
     template_name: str
     values: dict
     force_namespace: str = None
+    target_labels: dict[str, str] = None
 
 class DeleteJobRequest(BaseModel):
     name: str

kalavai_client/cli.py

@@ -34,6 +34,7 @@ from kalavai_client.env import (
     resource_path,
 )
 from kalavai_client.core import (
+    deploy_test_job,
     fetch_resources,
     fetch_job_names,
     fetch_job_details,
@@ -209,11 +210,15 @@ def input_gpus():
 @arguably.command
 def gui__start(*others, backend_only=False, gui_frontend_port=3000, gui_backend_port=8000, bridge_port=8001, log_level="critical"):
     """Run GUI (docker) and kalavai core backend (api)"""
-
+    if len(set([gui_frontend_port, gui_backend_port, bridge_port])) < 3:
+        console.log("[red]Error: ports must be unique")
+        return
+    
     if not backend_only:
         values = {
             "gui_frontend_port": gui_frontend_port,
             "gui_backend_port": gui_backend_port,
+            "bridge_port": bridge_port,
             "path": user_path("")
         }
         compose_yaml = load_template(
@@ -225,6 +230,9 @@ def gui__start(*others, backend_only=False, gui_frontend_port=3000, gui_backend_
         run_cmd(f"docker compose --file {USER_GUI_COMPOSE_FILE} up -d")
 
         console.log(f"[green]Loading GUI, may take a few minutes. It will be available at http://localhost:{gui_frontend_port}")
+    print(
+        "Deploying bridge API"
+    )
     run_api(port=bridge_port, log_level=log_level)
     
     if not backend_only:
@@ -1028,7 +1036,7 @@ def job__run(template_name, *others, values: str=None, force_namespace: str=None
         console.log(f"[green]{template_name} job deployed")
 
 @arguably.command
-def job__test(local_template_dir, *others, values, defaults, force_namespace: str=None):
+def job__test(local_template_dir, *others, values, force_namespace: str=None):
     """
     Helper to test local templates, useful for development
     """
@@ -1038,13 +1046,18 @@ def job__test(local_template_dir, *others, values, defaults, force_namespace: st
         console.log(f"[red]Problems with your pool: {str(e)}")
         return
     
-    if not os.path.isdir(local_template_dir):
-        console.log(f"[red]--local_template_dir ({local_template_dir}) is not a directory")
+    if not os.path.isfile(os.path.join(local_template_dir, "template.yaml")):
+        console.log(f"[red]template.yaml not found under {local_template_dir}")
+        return
+    if not os.path.isfile(os.path.join(local_template_dir, "values.yaml")):
+        console.log(f"[red]values.yaml not found under {local_template_dir}")
         return
     
     # load template
     with open(os.path.join(local_template_dir, "template.yaml"), "r") as f:
         template_str = f.read()
+    with open(os.path.join(local_template_dir, "values.yaml"), "r") as f:
+        defaults = f.read()
     
     # load values
     if not os.path.isfile(values):
@@ -1054,37 +1067,17 @@ def job__test(local_template_dir, *others, values, defaults, force_namespace: st
         raw_values = yaml.load(f, Loader=yaml.SafeLoader)
         values_dict = {variable["name"]: variable['value'] for variable in raw_values}
 
-    # load defaults
-    if not os.path.isfile(defaults):
-        console.log(f"[red]--defaults ({defaults}) is not a valid local file")
-        return
-    with open(defaults, "r") as f:
-        defaults = f.read()
+    result = deploy_test_job(
+        template_str=template_str,
+        values_dict=values_dict,
+        default_values=defaults,
+        force_namespace=force_namespace)
     
-    # submit custom deployment
-    data = {
-        "template": template_str,
-        "template_values": values_dict,
-        "default_values": defaults
-    }
-    if force_namespace is not None:
-        data["force_namespace"] = force_namespace
-
-    try:
-        result = request_to_server(
-            method="post",
-            endpoint="/v1/deploy_custom_job",
-            data=data,
-            server_creds=USER_LOCAL_SERVER_FILE,
-            user_cookie=USER_COOKIE
-        )
-        console.log("Deployment result:")
-        print(
-            json.dumps(result,indent=3)
-        )
-    except Exception as e:
-        console.log(f"[red]Error when connecting to kalavai service: {str(e)}")
-
+    if "error" in result:
+        console.log(f"[red]Error: {result['error']}")
+    else:
+        console.log("[green]Successfully deployed:")
+        console.log(result)
 
 @arguably.command
 def job__defaults(template_name, *others):
@@ -1123,7 +1116,7 @@ def job__delete(name, *others, force_namespace: str=None):
     # deploy template with kube-watcher
     result = delete_job(name=name, force_namespace=force_namespace)
     if "error" in result:
-        console.log(f"[red]Error when deleting job: {str(e)}")
+        console.log(f"[red]Error when deleting job: {result['error']}")
     else:
         console.log(f"{result}")
 
@@ -1209,13 +1202,14 @@ def job__list(*others):
     all_deployments = fetch_job_names()
     if "error" in all_deployments:
         console.log(f"[red]Error when connecting to kalavai service: {all_deployments}")
-        return 
+        return
     
     if len(all_deployments) == 0:
         console.log("[green]No deployments found.")
         return
     
     details = fetch_job_details(jobs=all_deployments)
+
     if "error" in details:
         console.log(f"[red]{details}")
         return
@@ -1255,19 +1249,19 @@ def job__logs(name, *others, pod_name=None, stream=False, tail=100, force_namesp
     while True:
         try:
             if not stream:
-                for pod, logs in all_logs.items():
+                for pod, info in all_logs.items():
                     if pod_name is not None and pod_name != pod:
                         continue
-                    console.log(f"[yellow]Pod {pod}")
-                    console.log(f"[green]{logs}")
+                    console.log(f"[yellow]Pod {pod} in {info['pod']['spec']['node_name']}")
+                    console.log(f"[green]{info['logs']}")
                 break
             else:
                 os.system("clear")
-                for pod, logs in all_logs.items():
+                for pod, info in all_logs.items():
                     if pod_name is not None and pod_name != pod:
                         continue
-                    print(f"Pod {pod}")
-                    print(f"{logs}")
+                    print(f"Pod {pod} in {info['pod']['spec']['node_name']}")
+                    print(f"{info['logs']}")
                 time.sleep(1)
         except KeyboardInterrupt:
             break

kalavai_client/core.py

@@ -291,14 +291,18 @@ def fetch_job_details(jobs: list[Job]):
                 user_cookie=USER_COOKIE
             )
             workers_status = defaultdict(int)
+            restart_counts = 0
             for ns, ss in result.items():
                 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)
+                    if "conditions" in values and values["conditions"] is not None:
+                        restart_counts = sum([c["restart_count"] for c in values["conditions"]])
                     workers_status[values["status"]] += 1
             workers = "\n".join([f"{k}: {v}" for k, v in workers_status.items()])
+            if restart_counts > 0:
+                workers += f"\n({restart_counts} restart)"
             # get URL details
             data = {
                 "label": TEMPLATE_LABEL,
@@ -338,7 +342,7 @@ def fetch_job_details(jobs: list[Job]):
     
     return job_details
 
-def deploy_job(template_name, values_dict, force_namespace=None):
+def deploy_job(template_name, values_dict, force_namespace=None, target_labels=None):
 
     # deploy template with kube-watcher
     data = {
@@ -347,6 +351,8 @@ def deploy_job(template_name, values_dict, force_namespace=None):
     }
     if force_namespace is not None:
         data["force_namespace"] = force_namespace
+    if target_labels is not None:
+        data["target_labels"] = target_labels
 
     try:
         result = request_to_server(
@@ -360,6 +366,29 @@ def deploy_job(template_name, values_dict, force_namespace=None):
     except Exception as e:
         return {"error": str(e)}  
     
+def deploy_test_job(template_str, values_dict, default_values, force_namespace=None):
+    
+    # submit custom deployment
+    data = {
+        "template": template_str,
+        "template_values": values_dict,
+        "default_values": default_values
+    }
+    if force_namespace is not None:
+        data["force_namespace"] = force_namespace
+
+    try:
+        result = request_to_server(
+            method="post",
+            endpoint="/v1/deploy_custom_job",
+            data=data,
+            server_creds=USER_LOCAL_SERVER_FILE,
+            user_cookie=USER_COOKIE
+        )
+        return result
+    except Exception as e:
+        return {"error": str(e)}
+    
 def delete_job(name, force_namespace=None):
     data = {
         "label": TEMPLATE_LABEL, # this ensures that both lws template and services are deleted
@@ -423,7 +452,7 @@ def fetch_job_logs(job_name, force_namespace=None, pod_name=None, tail=100):
             server_creds=USER_LOCAL_SERVER_FILE,
             user_cookie=USER_COOKIE
         )
-        return {pod: logs for pod, logs in all_logs.items() if pod_name is None or pod_name == pod}
+        return {pod: info for pod, info in all_logs.items() if pod_name is None or pod_name == pod}
 
     except Exception as e:
         return {"error": str(e)}

{kalavai_client-0.5.27.dist-info → kalavai_client-0.5.29.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: kalavai-client
-Version: 0.5.27
+Version: 0.5.29
 Summary: Client app for kalavai platform
 License: Apache-2.0
 Keywords: LLM,platform

{kalavai_client-0.5.27.dist-info → kalavai_client-0.5.29.dist-info}/RECORD

@@ -1,9 +1,9 @@
-kalavai_client/__init__.py,sha256=Wek6gRlCzMJXNHhlUntIAj24wWPV3k64gGvcy5WL7rU,23
+kalavai_client/__init__.py,sha256=GKwLB1LKAMgkwQjAmbvcykT9kbP8Rg1TdG1GyM2RMJw,23
 kalavai_client/__main__.py,sha256=WQUfxvRsBJH5gsCJg8pLz95QnZIj7Ol8psTO77m0QE0,73
 kalavai_client/assets/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-kalavai_client/assets/apps.yaml,sha256=d13TzkWtqdwpWOxuoG7eG0Jp0UhVUsboS28496H8iH4,5982
+kalavai_client/assets/apps.yaml,sha256=NDlPoV0MPpD9WxqJIu-mksb-FPDtIQpxvH4icpmn2a8,6304
 kalavai_client/assets/apps_values.yaml,sha256=CjKVelPQHd-hm-DTMEuya92feKiphU9mh3HrosLYYPE,1676
-kalavai_client/assets/docker-compose-gui.yaml,sha256=oE-neG3HC2PhdH-mIxrZdQlhqnycrSP_p8fRxjaxDFE,192
+kalavai_client/assets/docker-compose-gui.yaml,sha256=6OHZIDDTl_PwXSYo1d05JasWfT0iiUDhrja0nQDjrlw,692
 kalavai_client/assets/docker-compose-template.yaml,sha256=ii24Nn-dM5cZk9lxFgrzxnmK7yv_6kIIw7KUlWhvYeI,2831
 kalavai_client/assets/nginx.conf,sha256=drVVCg8GHucz7hmt_BI6giAhK92OV71257NTs3LthwM,225
 kalavai_client/assets/pool_config_template.yaml,sha256=fFz4w2-fMKD5KvyzFdfcWD_jSneRlmnjLc8hCctweX0,576
@@ -11,15 +11,15 @@ kalavai_client/assets/pool_config_values.yaml,sha256=VrM3XHQfQo6QLZ68qvagooUptaY
 kalavai_client/assets/user_workspace.yaml,sha256=wDvlMYknOPABAEo0dsQwU7bac8iubjAG9tdkFbJZ5Go,476
 kalavai_client/assets/user_workspace_values.yaml,sha256=G0HOzQUxrDMCwuW9kbWUZaKMzDDPVwDwzBHCL2Xi2ZM,542
 kalavai_client/auth.py,sha256=QsBh28L2LwjBBK6pTUE4Xu36lLDTyetyU1YfS1Hbb6g,1717
-kalavai_client/bridge_api.py,sha256=rXTz6WpzQtsDmBDUUOqPEjWx8vfiIWEfvP1iM4MYDGM,5501
-kalavai_client/bridge_models.py,sha256=WwGIaWBIk4s32YemgDB2CcrrCWC-KeZjTT3iBi-kaa0,936
-kalavai_client/cli.py,sha256=zQ205vqPW69oEt9EAO5wC8_yEvWbFEtLUnI90Oqt4t4,46642
+kalavai_client/bridge_api.py,sha256=xBd3KGoDcruuo53uNWOwA30CIPYgQLX8IkJWJgNHq9s,5546
+kalavai_client/bridge_models.py,sha256=k4ILxa8jfAcfgIIBJqK1DunDzHh_oNUpuiEpmhAfTP0,977
+kalavai_client/cli.py,sha256=MNmn8DgnEuyaCV8cMqr17JvfjalHcrP5Udpjn2jVXp8,46713
 kalavai_client/cluster.py,sha256=gwjmdsd--YrffT0BmZDOEpbrdm3lPskUuN5jdgcrOR0,12947
-kalavai_client/core.py,sha256=Trv2DDdlBAsBYUaTKnosWrfboYRrZnRcv-jfvmG7-LU,32288
+kalavai_client/core.py,sha256=m5FOPzix8oNBlLqgllgJ0PwLXxo-s438fjJ0FPRVIZs,33266
 kalavai_client/env.py,sha256=Zg2pP-xGJpQumo56KMBxBLgIsBmcNN0S9R-ZP2-s630,2604
 kalavai_client/utils.py,sha256=OPmrsycNyrs2ZpTsjAzBuPN8hQNJtsYDLPKU13tnf-U,13862
-kalavai_client-0.5.27.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-kalavai_client-0.5.27.dist-info/METADATA,sha256=5-vseYyG7Ya57UyFAqdEabXKGoFrzTMRnJgPb6FRZ6Q,14443
-kalavai_client-0.5.27.dist-info/WHEEL,sha256=XbeZDeTWKc1w7CSIyre5aMDU_-PohRwTQceYnisIYYY,88
-kalavai_client-0.5.27.dist-info/entry_points.txt,sha256=9T6D45gxwzfVbglMm1r6XPdXuuZdHfy_7fCeu2jUphc,50
-kalavai_client-0.5.27.dist-info/RECORD,,
+kalavai_client-0.5.29.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+kalavai_client-0.5.29.dist-info/METADATA,sha256=ZsOolX-q_zOLr2N4eKqnMnkpxHuJquFlWK3KgMxODik,14443
+kalavai_client-0.5.29.dist-info/WHEEL,sha256=XbeZDeTWKc1w7CSIyre5aMDU_-PohRwTQceYnisIYYY,88
+kalavai_client-0.5.29.dist-info/entry_points.txt,sha256=9T6D45gxwzfVbglMm1r6XPdXuuZdHfy_7fCeu2jUphc,50
+kalavai_client-0.5.29.dist-info/RECORD,,