kalavai-client 0.6.16__py3-none-any.whl → 0.6.17__py3-none-any.whl

kalavai_client/__init__.py

@@ -1,2 +1,2 @@
 
-__version__ = "0.6.16"
+__version__ = "0.6.17"

kalavai_client/assets/apps.yaml

@@ -18,6 +18,8 @@ repositories:
     url: https://opencost.github.io/opencost-helm-chart
   - name: minio
     url: https://charts.min.io/
+  - name: langfuse
+    url: https://langfuse.github.io/langfuse-k8s
 
 releases:
   - name: datashim
@@ -28,7 +30,7 @@ releases:
   - name: lago
     namespace: kalavai
     chart: kalavai/lago
-    installed: {{deploy_llm_sidecars}}
+    installed: {{deploy_lago}}
     set:
     - name: external.api.nodePort
       value: 32000
@@ -79,7 +81,7 @@ releases:
     namespace: kalavai
     chart: kalavai/kalavai-helios
     version: "0.1.11"
-    installed: false #{{deploy_llm_sidecars}}
+    installed: false
     set:
     - name: deployment.watcher_endpoint
       value: "http://{{watcher_service}}"
@@ -94,11 +96,11 @@ releases:
   - name: opencost
     namespace: opencost
     chart: opencost-charts/opencost
-    installed: {{deploy_llm_sidecars}}
+    installed: {{deploy_opencost}}
   - name: prometheus
     namespace: prometheus-system
     chart: prometheus/prometheus
-    installed: {{deploy_llm_sidecars}}
+    installed: {{deploy_prometheus}}
     set:
     - name: prometheus-pushgateway.enabled
       value: false
@@ -152,7 +154,7 @@ releases:
     - name: replicas
       value: 1
     - name: image_tag
-      value: "v2025.06.7"
+      value: "v2025.06.15"
     - name: deployment.in_cluster
       value: "True"
     - name: deployment.kalavai_username_key

kalavai_client/assets/apps_values.yaml

@@ -1,3 +1,24 @@
+### APS ###
+- name: deploy_lago
+  default: "False"
+  description: "Deploy Lago payment system"
+
+- name: deploy_opencost
+  default: "False"
+  description: "Deploy Opencost cost monitoring system"
+
+- name: deploy_prometheus
+  default: "True"
+  description: "Deploy Prometheus system monitoring system"
+
+- name: deploy_langfuse
+  default: "False"
+  description: "Deploy Langfuse LLM tracing system"
+
+######
+
+### VARIABLES ###
+
 - name: kalavai_api_endpoint
   default: https://platform.kalavai.net/_/api
   description: ""
@@ -78,4 +99,4 @@
 
 - name: minio_rootPassword
   default: "password"
-  description: ""
+  description: ""

kalavai_client/assets/pool_config_template.yaml

@@ -17,3 +17,4 @@
 #   unmapMarkSnapChainRemoved: "ignored"
 #   disableRevisionCounter: "true"
 #   dataEngine: "v1"
+

kalavai_client/assets/pool_config_values.yaml

@@ -1,12 +1,4 @@
-# STORAGE #
-- name: storage_label_selector
-  default: "kalavai.storage.enabled:True"
-  description: ""
-
 - name: storage_class_name
-  default: "longhorn-model"
+  default: "longhorn"
   description: ""
 
-- name: storage_replicas
-  default: 1
-  description: ""

kalavai_client/bridge_api.py

@@ -244,9 +244,9 @@ def send_pool_invites(request: InvitesRequest, api_key: str = Depends(verify_api
     summary="Fetch resources",
     description="Get available resources",
     response_description="Resource information")
-def resources(api_key: str = Depends(verify_api_key)):
+def resources(request: NodesActionRequest, api_key: str = Depends(verify_api_key)):
     """Get available resources"""
-    return fetch_resources()
+    return fetch_resources(node_names=request.nodes)
 
 @app.get("/fetch_job_names",
     summary="Fetch job names",
@@ -317,7 +317,7 @@ def job_templates(api_key: str = Depends(verify_api_key)):
 @app.get("/fetch_job_defaults",
     summary="Fetch job defaults",
     description="Get default values for a job template",
-    response_description="Job default values")
+    response_description="Job metadata values")
 def job_templates(name: str, api_key: str = Depends(verify_api_key)):
     """
     Get job defaults with the following parameters:

kalavai_client/bridge_models.py

@@ -18,7 +18,7 @@ class CreatePoolRequest(BaseModel):
     description: str = Field("", description="Description of the pool")
 
 class NodesActionRequest(BaseModel):
-    nodes: list[str] = Field(description="List of node names to perform the action on")
+    nodes: list[str] = Field(None, description="List of node names to perform the action on")
 
 class JoinPoolRequest(BaseModel):
     token: str = Field(description="Token to join the pool")

kalavai_client/cli.py

@@ -207,28 +207,43 @@ def input_gpus(non_interactive=False):
 @arguably.command
 def gui__start(
     *others,
-    gui_frontend_port=3000,
-    gui_backend_port=8000,
-    bridge_port=8001,
     log_level="critical",
     backend_only=False
 ):
     """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
+    ports_needed = 1 if backend_only else 3
+    # find 3 available ports
+    ip = socket.gethostbyname (socket.gethostname())
+    ports = []
+    for port in range(49152,65535):
+        try:
+            serv = socket.socket(socket.AF_INET,socket.SOCK_STREAM) # create a new socket
+            serv.bind((ip, port)) # bind socket with address
+            serv.close()
+            ports.append(port)
+        except:
+            #port closed
+            pass
+        if len(ports) >= ports_needed:
+            break
     
+    if len(ports) < ports_needed:
+        # if not found, error
+        console.log(f"[red]Cannot initialise GUI: Could not find {ports_needed} free ports in your machine")    
+        return
+    console.log(f"Using ports: {ports}")
+
     user_key = load_user_id()
     if user_key is not None:
         console.log(f"[green]Using user key: {user_key}")
     if not backend_only:
         values = {
-            "gui_frontend_port": gui_frontend_port,
-            "gui_backend_port": gui_backend_port,
-            "bridge_port": bridge_port,
+            "gui_frontend_port": ports[1],
+            "gui_backend_port": ports[2],
+            "bridge_port": ports[0],
             "path": user_path("", create_path=True),
             "protected_access": user_key
-    }
+        }
         compose_yaml = load_template(
             template_path=DOCKER_COMPOSE_GUI,
             values=values)
@@ -237,11 +252,11 @@ def gui__start(
         
         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}")
+        console.log(f"[green]Loading GUI, may take a few minutes. It will be available at http://localhost:{ports[1]}")
     print(
         "Deploying bridge API"
     )
-    run_api(port=bridge_port, log_level=log_level)
+    run_api(port=ports[0], log_level=log_level)
 
     if not backend_only:
         run_cmd(f"docker compose --file {USER_GUI_COMPOSE_FILE} down")
@@ -1096,12 +1111,22 @@ def job__defaults(template_name, *others):
         return
     
     # deploy template with kube-watcher
-    defaults = fetch_job_defaults(name=template_name)
+    data = fetch_job_defaults(name=template_name)
+    metadata = data["metadata"]
+    defaults = data["defaults"]
     if "error" in defaults:
         console.log(f"[red]Error when fetching job defaults: {defaults}")
     print(
         json.dumps(defaults, indent=3)
     )
+    print(
+        "*****************",
+        "Metadata",
+        "*****************"
+    )
+    print(
+        json.dumps(metadata, indent=3)
+    )
 
 
 @arguably.command

kalavai_client/core.py

@@ -13,7 +13,6 @@ import re
 
 from kalavai_client.cluster import CLUSTER
 from kalavai_client.utils import (
-    DEPLOY_LLM_SIDECARS_KEY,
     NODE_ROLE_LABEL,
     check_gpu_drivers,
     generate_join_token,
@@ -80,6 +79,7 @@ class Job(BaseModel):
     workers: Optional[str] = None
     endpoint: Optional[str] = None
     status: Optional[str] = None
+    host_nodes: Optional[str] = None
 
 class DeviceStatus(BaseModel):
     name: str
@@ -198,19 +198,22 @@ def get_ip_addresses(subnet=None):
             raise ValueError(f"No IPs available on subnet {subnet}")
     return ips
 
-def fetch_resources():
+def fetch_resources(node_names: list[str]=None):
+    data = {}
+    if node_names is not None:
+        data["node_names"] = node_names
     try:
         total = request_to_server(
             method="get",
             endpoint="/v1/get_cluster_total_resources",
-            data={},
+            data=data,
             server_creds=USER_LOCAL_SERVER_FILE,
             user_cookie=USER_COOKIE
         )
         available = request_to_server(
             method="get",
             endpoint="/v1/get_cluster_available_resources",
-            data={},
+            data=data,
             server_creds=USER_LOCAL_SERVER_FILE,
             user_cookie=USER_COOKIE
         )
@@ -224,14 +227,14 @@ def fetch_job_defaults(name):
         "template": name
     }
     try:
-        defaults = request_to_server(
+        metadata = request_to_server(
             method="get",
             endpoint="/v1/job_defaults",
             data=data,
             server_creds=USER_LOCAL_SERVER_FILE,
             user_cookie=USER_COOKIE
         )
-        return defaults
+        return metadata
     except Exception as e:
         return {"error": str(e)}
     
@@ -296,14 +299,18 @@ def fetch_job_details(jobs: list[Job]):
             )
             workers_status = defaultdict(int)
             restart_counts = 0
+            host_nodes = set()
             for ns, ss in result.items():
                 if ns != namespace: # same job name, different namespace
                     continue
                 for _, values in ss.items():
-                    # 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
+                    # get nodes involved in deployment (needs kubewatcher)
+                    if "node_name" in values:
+                        host_nodes.add(values["node_name"])
+
             workers = "\n".join([f"{k}: {v}" for k, v in workers_status.items()])
             if restart_counts > 0:
                 workers += f"\n({restart_counts} restart)"
@@ -320,7 +327,8 @@ def fetch_job_details(jobs: list[Job]):
                 server_creds=USER_LOCAL_SERVER_FILE,
                 user_cookie=USER_COOKIE
             )
-            node_ports = [f"{p['node_port']} (mapped to {p['port']})" for s in result.values() for p in s["ports"]]
+            #node_ports = [f"{p['node_port']} (mapped to {p['port']})" for s in result.values() for p in s["ports"]]
+            node_ports = [f"{p['node_port']}" for s in result.values() for p in s["ports"]]
 
             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:
@@ -338,7 +346,8 @@ def fetch_job_details(jobs: list[Job]):
                     name=deployment,
                     workers=workers,
                     endpoint="\n".join(urls),
-                    status=str(status))
+                    status=str(status),
+                    host_nodes=" ".join(host_nodes))
             )
 
         except Exception as e:
@@ -802,8 +811,7 @@ def create_pool(
         WATCHER_PORT_KEY: DEFAULT_WATCHER_PORT,
         WATCHER_SERVICE_KEY: watcher_service,
         USER_NODE_LABEL_KEY: USER_NODE_LABEL,
-        ALLOW_UNREGISTERED_USER_KEY: True, # Change this if only registered users are allowed,
-        DEPLOY_LLM_SIDECARS_KEY: location is not None
+        ALLOW_UNREGISTERED_USER_KEY: True, # Change this if only registered users are allowed
     }
 
     store_server_info(

kalavai_client/utils.py

@@ -38,7 +38,6 @@ CLUSTER_NAME_KEY = "cluster_name"
 AUTH_KEY = "watcher_admin_key"
 WRITE_AUTH_KEY = "watcher_write_key"
 ALLOW_UNREGISTERED_USER_KEY = "watcher_allow_unregistered_user"
-DEPLOY_LLM_SIDECARS_KEY = "deploy_llm_sidecars"
 NODE_ROLE_LABEL = "kalavai.node_role"
 USER_API_KEY = "user_api_key"
 READONLY_AUTH_KEY = "watcher_readonly_key"

{kalavai_client-0.6.16.dist-info → kalavai_client-0.6.17.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: kalavai-client
-Version: 0.6.16
+Version: 0.6.17
 Summary: Client app for kalavai platform
 License: Apache-2.0
 Keywords: LLM,platform

kalavai_client-0.6.17.dist-info/RECORD

@@ -0,0 +1,25 @@
+kalavai_client/__init__.py,sha256=6D1LEuHzwFI-n2PmIawhX9C0K4e_BwIIJUAUnpkzCWE,23
+kalavai_client/__main__.py,sha256=WQUfxvRsBJH5gsCJg8pLz95QnZIj7Ol8psTO77m0QE0,73
+kalavai_client/assets/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+kalavai_client/assets/apps.yaml,sha256=zVtfPqesNhoBLpNlhIPAVtxgXLqEQU2pK1GTzKGEqiQ,6395
+kalavai_client/assets/apps_values.yaml,sha256=dvsAnMC1uk9oDsnITIYJc5CIg9LTwGzXldjPZTwRQyE,2069
+kalavai_client/assets/docker-compose-gui.yaml,sha256=shqN78YLw0QP7bqTKveI4ppz5E-5b1JowmsSB4OG3nA,778
+kalavai_client/assets/docker-compose-template.yaml,sha256=KHIwJ2WWX7Y7wQKiXRr82Jqd3IKRyls5zhTyl8mSmrc,1805
+kalavai_client/assets/nginx.conf,sha256=drVVCg8GHucz7hmt_BI6giAhK92OV71257NTs3LthwM,225
+kalavai_client/assets/pool_config_template.yaml,sha256=MhBZQsEMKrBgbUVSKgIGmXWhybeGKG6l5XvJb38y5GI,577
+kalavai_client/assets/pool_config_values.yaml,sha256=_iAnugramLiwJaaDcPSetThvOdR7yFiCffdMri-SQCU,68
+kalavai_client/assets/user_workspace.yaml,sha256=wDvlMYknOPABAEo0dsQwU7bac8iubjAG9tdkFbJZ5Go,476
+kalavai_client/assets/user_workspace_values.yaml,sha256=G0HOzQUxrDMCwuW9kbWUZaKMzDDPVwDwzBHCL2Xi2ZM,542
+kalavai_client/auth.py,sha256=EB3PMvKUn5_KAQkezkEHEt-OMZXyfkZguIQlUFkEHcA,3243
+kalavai_client/bridge_api.py,sha256=5tYqI8UdG7K1Qskywk97kC0TpvYruUZxqxvbn-2nve4,15405
+kalavai_client/bridge_models.py,sha256=t1fJGaF6YDMQdOnEU3XT8zTBHU8eUWJ1yhM5s7c6vMo,2546
+kalavai_client/cli.py,sha256=SzKG7_ZG0ehMQsECQRWSvqj2Fju2Gd5O7uBa60bFBAY,47830
+kalavai_client/cluster.py,sha256=Z2PIXbZuSAv9xmw-MyZP1M41BpVMpirLzG51bqGA-zc,13548
+kalavai_client/core.py,sha256=dJVX5mhFzIshazCfAzb-AqpqWjkp_djgbMyNXzuAF48,34650
+kalavai_client/env.py,sha256=YsfZj7LWf6ABquDsoIFFkXCFYwenpDk8zVnGsf7qv98,2823
+kalavai_client/utils.py,sha256=bhvQzF12q7L2hGVrbcmXRDXXIsAdlzcsTms6RQRxGU4,12733
+kalavai_client-0.6.17.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+kalavai_client-0.6.17.dist-info/METADATA,sha256=J5TEqnwPm3ZopGEi2MRn_ddy_VV101JcB8sl6MGL4iI,12655
+kalavai_client-0.6.17.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
+kalavai_client-0.6.17.dist-info/entry_points.txt,sha256=9T6D45gxwzfVbglMm1r6XPdXuuZdHfy_7fCeu2jUphc,50
+kalavai_client-0.6.17.dist-info/RECORD,,

kalavai_client-0.6.16.dist-info/RECORD

@@ -1,25 +0,0 @@
-kalavai_client/__init__.py,sha256=1--FABNdIxbiNn1wQox38stjIswkk1wPeOgoYJXMsNU,23
-kalavai_client/__main__.py,sha256=WQUfxvRsBJH5gsCJg8pLz95QnZIj7Ol8psTO77m0QE0,73
-kalavai_client/assets/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-kalavai_client/assets/apps.yaml,sha256=Qe3RtY4kQbzZnF9K724FUbtqnkuCGfNUfK-WWtamATg,6365
-kalavai_client/assets/apps_values.yaml,sha256=WRew3bS1MztjzcJfphuJcKn0n2T1ICRupPpr_Csjt_s,1644
-kalavai_client/assets/docker-compose-gui.yaml,sha256=shqN78YLw0QP7bqTKveI4ppz5E-5b1JowmsSB4OG3nA,778
-kalavai_client/assets/docker-compose-template.yaml,sha256=KHIwJ2WWX7Y7wQKiXRr82Jqd3IKRyls5zhTyl8mSmrc,1805
-kalavai_client/assets/nginx.conf,sha256=drVVCg8GHucz7hmt_BI6giAhK92OV71257NTs3LthwM,225
-kalavai_client/assets/pool_config_template.yaml,sha256=fFz4w2-fMKD5KvyzFdfcWD_jSneRlmnjLc8hCctweX0,576
-kalavai_client/assets/pool_config_values.yaml,sha256=VrM3XHQfQo6QLZ68qvagooUptaYgl1pszniY_JUtemk,233
-kalavai_client/assets/user_workspace.yaml,sha256=wDvlMYknOPABAEo0dsQwU7bac8iubjAG9tdkFbJZ5Go,476
-kalavai_client/assets/user_workspace_values.yaml,sha256=G0HOzQUxrDMCwuW9kbWUZaKMzDDPVwDwzBHCL2Xi2ZM,542
-kalavai_client/auth.py,sha256=EB3PMvKUn5_KAQkezkEHEt-OMZXyfkZguIQlUFkEHcA,3243
-kalavai_client/bridge_api.py,sha256=O65aIh5lUl0KldRekHzLC-xdv1YJmrR14kt5-3UgCco,15351
-kalavai_client/bridge_models.py,sha256=5ALGbkb6cxKwXbrzeTa9ja0kiZkJBvnY3J1IsmXTn0U,2540
-kalavai_client/cli.py,sha256=_LK5OrCM5PYcYZo7lwXyfI3mlNzLFhL-BicKYbJkxeY,47123
-kalavai_client/cluster.py,sha256=Z2PIXbZuSAv9xmw-MyZP1M41BpVMpirLzG51bqGA-zc,13548
-kalavai_client/core.py,sha256=R8UBTTzMHVPHuM9nB70cIxUxVCHyBspEq1cAWH1OyOQ,34304
-kalavai_client/env.py,sha256=YsfZj7LWf6ABquDsoIFFkXCFYwenpDk8zVnGsf7qv98,2823
-kalavai_client/utils.py,sha256=kGtfEuXVG5LgMJk289ksFgYrsMHwKXN7yvS5wCIou8s,12781
-kalavai_client-0.6.16.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-kalavai_client-0.6.16.dist-info/METADATA,sha256=K5mzqy8pSDdK6WWFSt8YZNTJLENfeV3OOGELq417dYs,12655
-kalavai_client-0.6.16.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
-kalavai_client-0.6.16.dist-info/entry_points.txt,sha256=9T6D45gxwzfVbglMm1r6XPdXuuZdHfy_7fCeu2jUphc,50
-kalavai_client-0.6.16.dist-info/RECORD,,