kalavai-client 0.5.10__py3-none-any.whl → 0.5.13__py3-none-any.whl

kalavai_client/__init__.py

@@ -1,2 +1,2 @@
 
-__version__ = "0.5.10"
+__version__ = "0.5.13"

kalavai_client/assets/apps.yaml

@@ -139,7 +139,7 @@ releases:
     - name: replicas
       value: 2
     - name: image_tag
-      value: "v2025.01"
+      value: "v2025.01.9"
     - name: deployment.in_cluster
       value: "True"
     - name: deployment.use_auth_key

kalavai_client/assets/docker-compose-template.yaml

@@ -1,8 +1,9 @@
 services:
 {% if vpn %}
   {{vpn_name}}:
-    image: gravitl/netclient:v0.30.0
+    image: gravitl/netclient:v0.24.3
     container_name: {{vpn_name}}
+    #privileged: true
     cap_add:
       - NET_ADMIN
       - SYS_MODULE

kalavai_client/cli.py

@@ -15,6 +15,21 @@ import netifaces as ni
 import arguably
 from rich.console import Console
 
+from kalavai_client.env import (
+    USER_COOKIE,
+    USER_LOCAL_SERVER_FILE,
+    TEMPLATE_LABEL,
+    user_path
+)
+from kalavai_client.core import (
+    fetch_resources,
+    fetch_job_names,
+    fetch_job_details,
+    fetch_devices,
+    fetch_job_logs,
+    fetch_gpus,
+    load_gpu_models
+)
 from kalavai_client.utils import (
     check_gpu_drivers,
     run_cmd,
@@ -27,10 +42,8 @@ from kalavai_client.utils import (
     generate_table,
     request_to_server,
     resource_path,
-    user_path,
     safe_remove,
     leave_vpn,
-    join_vpn,
     get_vpn_details,
     load_server_info,
     user_login,
@@ -68,7 +81,6 @@ LOCAL_TEMPLATES_DIR = os.getenv("LOCAL_TEMPLATES_DIR", None)
 VERSION = 1
 RESOURCE_EXCLUDE = ["ephemeral-storage", "hugepages-1Gi", "hugepages-2Mi", "pods"]
 CORE_NAMESPACES = ["lws-system", "kube-system", "gpu-operator", "kalavai"]
-TEMPLATE_LABEL = "kalavai.job.name"
 RAY_LABEL = "kalavai.ray.name"
 PVC_NAME_LABEL = "kalavai.storage.name"
 DOCKER_COMPOSE_TEMPLATE = resource_path("kalavai_client/assets/docker-compose-template.yaml")
@@ -98,9 +110,7 @@ USER_COMPOSE_FILE = user_path("docker-compose-worker.yaml")
 USER_VPN_COMPOSE_FILE = user_path("docker-compose-vpn.yaml")
 USER_HELM_APPS_FILE = user_path("apps.yaml")
 USER_KUBECONFIG_FILE = user_path("kubeconfig")
-USER_LOCAL_SERVER_FILE = user_path(".server")
 USER_TEMPLATES_FOLDER = user_path("templates", create_path=True)
-USER_COOKIE = user_path(".user_cookie.pkl")
 
 
 console = Console()
@@ -118,7 +128,7 @@ CLUSTER = dockerCluster(
 ######################
 ## HELPER FUNCTIONS ##
 ######################
-    
+
 def check_seed_compatibility():
     """Check required packages to start pools"""
     logs = []
@@ -288,21 +298,11 @@ def select_ip_address(subnet=None):
             console.log("[red] Input error")
     return ips[option]
 
-def fetch_gpus():
-    data = request_to_server(
-        method="post",
-        endpoint="/v1/get_node_gpus",
-        data={},
-        server_creds=USER_LOCAL_SERVER_FILE,
-        user_cookie=USER_COOKIE
-    )
-    return data.items()
-
 def select_gpus(message):
     console.log(f"[yellow]{message}")
     gpu_models = ["Any/None"]
     gpu_models_full = ["Any/None"]
-    available_gpus = fetch_gpus()
+    available_gpus = load_gpu_models()
     for _, gpus in available_gpus:
         for gpu in gpus["gpus"]:
             #status = "free" if "ready" in gpu else "busy"
@@ -337,18 +337,19 @@ def select_token_type():
             break
     return {"admin": choice == 0, "user": choice == 1, "worker": choice == 2}
 
-def generate_compose_config(role, node_name, is_public, node_labels=None, pool_ip=None, vpn_token=None, pool_token=None):
+def generate_compose_config(role, node_name, is_public, use_gpus=True, node_labels=None, pool_ip=None, vpn_token=None, pool_token=None):
     num_gpus = 0
-    try:
-        has_gpus = check_gpu_drivers()
-        if has_gpus:
-            max_gpus = int(run_cmd("nvidia-smi -L | wc -l").decode())
-            num_gpus = user_confirm(
-                question=f"{max_gpus} NVIDIA GPU(s) detected. How many GPUs would you like to include?",
-                options=range(max_gpus+1)
-            ) 
-    except:
-        console.log(f"[red]WARNING: error when fetching NVIDIA GPU info. GPUs will not be used on this local machine")
+    if use_gpus:
+        try:
+            has_gpus = check_gpu_drivers()
+            if has_gpus:
+                max_gpus = int(run_cmd("nvidia-smi -L | wc -l").decode())
+                num_gpus = user_confirm(
+                    question=f"{max_gpus} NVIDIA GPU(s) detected. How many GPUs would you like to include?",
+                    options=range(max_gpus+1)
+                ) 
+        except:
+            console.log(f"[red]WARNING: error when fetching NVIDIA GPU info. GPUs will not be used on this local machine")
     if node_labels is not None:
         node_labels = " ".join([f"--node-label {key}={value}" for key, value in node_labels.items()])
     compose_values = {
@@ -736,7 +737,7 @@ def pool__check_token(token, *others, public=False):
 
 
 @arguably.command
-def pool__join(token, *others, node_name=None, ip_address: str=None):
+def pool__join(token, *others, node_name=None):
     """
     Join Kalavai pool and start/resume sharing resources.
 
@@ -859,7 +860,7 @@ def pool__join(token, *others, node_name=None, ip_address: str=None):
         CLUSTER.start_worker_node()
     except Exception as e:
         console.log(f"[red] Error connecting to {cluster_name} @ {kalavai_seed_ip}. Check with the admin if the token is still valid.")
-        leave_vpn(container_name=DEFAULT_VPN_CONTAINER_NAME)
+        pool__stop()
         exit()
 
     # ensure we are connected
@@ -868,6 +869,22 @@ def pool__join(token, *others, node_name=None, ip_address: str=None):
         time.sleep(30)
         if is_watcher_alive(server_creds=USER_LOCAL_SERVER_FILE, user_cookie=USER_COOKIE):
             break
+
+    # send note to server to let them know the node is coming online
+    if not pre_join_check(node_name=node_name, server_url=watcher_service, server_key=auth_key):
+        console.log(f"[red] Failed pre join checks. Server offline or node '{node_name}' may already exist. Please specify a different one with [yellow]--node-name'")
+        pool__stop()
+        return
+    
+    # check the node has connected successfully
+    try:
+        while not CLUSTER.is_agent_running():
+            console.log("waiting for runner, may take a few minutes... Press <ctrl+c> to stop")
+            time.sleep(30)
+    except KeyboardInterrupt:
+        console.log("[red]Installation aborted. Leaving pool.")
+        pool__stop()
+        return
     
     init_user_workspace()
     
@@ -962,29 +979,24 @@ def pool__gpus(*others, available=False):
         console.log(f"[red]Problems with your pool: {str(e)}")
         return
 
-    try:
-        data = fetch_gpus()
-        columns, rows = [], []
-        for node, gpus in data:
-            row_gpus = []
-            for gpu in gpus["gpus"]:
-                status = gpu["ready"] if "ready" in gpu else True
-                if available and not status:
-                    continue
-                row_gpus.append( (f"{gpu['model']} ({math.floor(int(gpu['memory'])/1000)} GBs)", str(status)))
-            if len(row_gpus) > 0:
-                models, statuses = zip(*row_gpus)
-                rows.append([node, "\n".join(statuses), "\n".join(models), str(gpus["available"]), str(gpus["capacity"])])
-
-            columns = ["Ready", "GPU(s)", "Available", "Total"]
-        columns = ["Node"] + columns
-        console.print(
-            generate_table(columns=columns, rows=rows,end_sections=[n for n in range(len(rows))])
-        )
-        
-    except Exception as e:
-        console.log(f"[red]Error when connecting to kalavai service: {str(e)}")
-
+    gpus = fetch_gpus(available=available)
+    if "error" in gpus:
+        console.log(f"[red]Error when fetching gpus: {gpus}")
+        return
+    
+    columns = ["Node", "Ready", "GPU(s)", "Available", "Total"]
+    rows = []
+    for gpu in gpus:
+        rows.append([
+            gpu.node,
+            str(gpu.ready),
+            gpu.model,
+            str(gpu.available),
+            str(gpu.total)
+        ])
+    console.print(
+        generate_table(columns=columns, rows=rows,end_sections=[n for n in range(len(rows))])
+    )
 
 @arguably.command
 def pool__resources(*others):
@@ -997,45 +1009,33 @@ def pool__resources(*others):
         console.log(f"[red]Problems with your pool: {str(e)}")
         return
 
-    try:
-        total = request_to_server(
-            method="get",
-            endpoint="/v1/get_cluster_total_resources",
-            data={},
-            server_creds=USER_LOCAL_SERVER_FILE,
-            user_cookie=USER_COOKIE
-        )
-        available = request_to_server(
-            method="get",
-            endpoint="/v1/get_cluster_available_resources",
-            data={},
-            server_creds=USER_LOCAL_SERVER_FILE,
-            user_cookie=USER_COOKIE
-        )
-        columns = []
-        total_values = []
-        available_values = []
-        for col in total.keys():
-            if col in RESOURCE_EXCLUDE:
-                continue
-            columns.append(col)
-            total_values.append(str(total[col]))
-            available_values.append(str(available[col]))
-        
-        columns = [""] + columns
-        total_values = ["Total"] + total_values
-        available_values = ["Available"] + available_values
-        
-        rows = [
-            tuple(available_values),
-            tuple(total_values)
-        ]
-        console.print(
-            generate_table(columns=columns, rows=rows, end_sections=[0, 1])
-        )
-        
-    except Exception as e:
+    data = fetch_resources()
+    if "error" in data:
         console.log(f"[red]Error when connecting to kalavai service: {str(e)}")
+        return
+    
+    columns = []
+    total_values = []
+    available_values = []
+    for col in data["total"].keys():
+        if col in RESOURCE_EXCLUDE:
+            continue
+        columns.append(col)
+        total_values.append(str(data["total"][col]))
+        available_values.append(str(data["available"][col]))
+    
+    columns = [""] + columns
+    total_values = ["Total"] + total_values
+    available_values = ["Available"] + available_values
+    
+    rows = [
+        tuple(available_values),
+        tuple(total_values)
+    ]
+    console.print(
+        generate_table(columns=columns, rows=rows, end_sections=[0, 1])
+    )
+        
 
 @arguably.command
 def pool__update(*others):
@@ -1102,6 +1102,10 @@ def pool__attach(token, *others, node_name=None):
     """
     Set creds in token on the local instance
     """
+
+    if node_name is None:
+        node_name = socket.gethostname()
+    
     # check that is not attached to another instance
     if os.path.exists(USER_LOCAL_SERVER_FILE):
         option = user_confirm(
@@ -1160,6 +1164,7 @@ def pool__attach(token, *others, node_name=None):
     
     # Generate docker compose recipe
     generate_compose_config(
+        use_gpus=False,
         role="",
         vpn_token=vpn["key"],
         node_name=node_name,
@@ -1327,22 +1332,18 @@ def node__list(*others):
         return
 
     try:
-        data = request_to_server(
-            method="get",
-            endpoint="/v1/get_nodes",
-            data={},
-            server_creds=USER_LOCAL_SERVER_FILE,
-            user_cookie=USER_COOKIE
-        )
+        devices = fetch_devices()
         rows = []
-        columns = ["Node name"]
-        for node, status in data.items():
-            row = [node]
-            for key, value in status.items():
-                if key not in columns:
-                    columns.append(key)
-                row.append(str(value))
-            rows.append(tuple(row))
+        columns = ["Node name", "Memory Pressure", "Disk pressure", "PID pressure", "Ready", "Unschedulable"]
+        for device in devices:
+            rows.append([
+                device.name,
+                str(device.memory_pressure),
+                str(device.disk_pressure),
+                str(device.pid_pressure),
+                str(device.ready),
+                str(device.unschedulable)
+            ])
         
         console.log("Nodes with 'unschedulable=True' will not receive workload")
         console.log("To make a node unschedulable (i.e. won't receive workloads) use [yellow]kalavai node cordon <node name>")
@@ -1510,6 +1511,64 @@ def job__run(template_name, *others, values: str=None, force_namespace: str=None
         console.log(f"[red]Error when connecting to kalavai service: {str(e)}")
         return
 
+@arguably.command
+def job__test(local_template_dir, *others, values, defaults, force_namespace: str=None):
+    """
+    Helper to test local templates, useful for development
+    """
+    try:
+        CLUSTER.validate_cluster()
+    except Exception as e:
+        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")
+        return
+    
+    # load template
+    with open(os.path.join(local_template_dir, "template.yaml"), "r") as f:
+        template_str = f.read()
+    
+    # load values
+    if not os.path.isfile(values):
+        console.log(f"[red]--values ({values}) is not a valid local file")
+        return
+    with open(values, "r") as f:
+        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()
+    
+    # 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)}")
+
 
 @arguably.command
 def job__defaults(template_name, *others):
@@ -1586,7 +1645,7 @@ def job__estimate(billion_parameters, *others, precision=32):
     
     average_vram = 8
     required_memory = float(billion_parameters) * (precision / 8) / 1.2
-    available_gpus = fetch_gpus()
+    available_gpus = load_gpu_models()
     vrams = []
     for _, gpus in available_gpus:
         for model in gpus["gpus"]:
@@ -1643,7 +1702,7 @@ def job__status(name, *others):
         return
 
 @arguably.command
-def job__list(*others, detailed=False):
+def job__list(*others):
     """
     List jobs in the cluster
     """
@@ -1653,106 +1712,22 @@ def job__list(*others, detailed=False):
         console.log(f"[red]Problems with your pool: {str(e)}")
         return
 
-    data = {
-        "group": "batch.volcano.sh",
-        "api_version": "v1alpha1",
-        "plural": "jobs"
-    }
-    try:
-        result = request_to_server(
-            method="post",
-            endpoint="/v1/get_objects_of_type",
-            data=data,
-            server_creds=USER_LOCAL_SERVER_FILE,
-            user_cookie=USER_COOKIE
-        )
-        all_deployments = defaultdict(list)
-        for ns, ds in result.items():
-            all_deployments[ns].extend([d["metadata"]["labels"][TEMPLATE_LABEL] for d in ds["items"]])
-        #deployments = {ns: d["metadata"]["labels"][TEMPLATE_LABEL] for ns, ds in result.items() for d in ds["items"]}
-    except Exception as e:
-        console.log(f"[red]Error when connecting to kalavai service: {str(e)}")
-        return
-    if len(all_deployments.keys()) == 0:
+    all_deployments = fetch_job_names()
+    if "error" in all_deployments:
+        console.log(f"[red]Error when connecting to kalavai service: {all_deployments}")
+        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
     columns = ["Owner", "Deployment", "Workers", "Endpoint"]
-    if detailed:
-        columns.append("Status")
-    rows = []
-    for namespace, deployments in all_deployments.items():
-        for deployment in deployments:
-            try:
-                # get status for deployment
-                if detailed:
-                    data = {
-                        "group": "batch.volcano.sh",
-                        "api_version": "v1alpha1",
-                        "plural": "jobs",
-                        # "group": "leaderworkerset.x-k8s.io",
-                        # "api_version": "v1",
-                        # "plural": "leaderworkersets",
-                        "name": deployment
-                    }
-                    result = request_to_server(
-                        method="post",
-                        endpoint="/v1/get_status_for_object",
-                        data=data,
-                        server_creds=USER_LOCAL_SERVER_FILE,
-                        user_cookie=USER_COOKIE
-                    )
-                    ss = [] # flatten results ({namespace: statuses})
-                    [ss.extend(values) for values in result.values()]
-                    if len(ss) > 0:
-                        last = ss[-1]
-                        statuses = f"[{last['lastTransitionTime']}] {last['status']}"
-                    else:
-                        statuses = "Unknown"
-                # get pod statuses
-                data = {
-                    "label": TEMPLATE_LABEL,
-                    "value": deployment
-                }
-                result = request_to_server(
-                    method="post",
-                    endpoint="/v1/get_pods_status_for_label",
-                    data=data,
-                    server_creds=USER_LOCAL_SERVER_FILE,
-                    user_cookie=USER_COOKIE
-                )
-                workers_status = defaultdict(int)
-                for ns, ss in result.items():
-                    if ns != namespace: # same job name, different namespace
-                        continue
-                    for _, values in ss.items():
-                        workers_status[values["status"]] += 1
-                workers = "\n".join([f"{k}: {v}" for k, v in workers_status.items()])
-                # get URL details
-                data = {
-                    "label": TEMPLATE_LABEL,
-                    "value": deployment,
-                    "types": ["NodePort"]
-                }
-                result = request_to_server(
-                    method="post",
-                    endpoint="/v1/get_ports_for_services",
-                    data=data,
-                    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"]]
-
-                urls = [f"http://{load_server_info(data_key=SERVER_IP_KEY, file=USER_LOCAL_SERVER_FILE)}:{node_port}" for node_port in node_ports]
-                row = [namespace, deployment, workers, "\n".join(urls)]
-                if detailed:
-                    row.append(statuses)
-                rows.append(row)
-
-            except Exception as e:
-                console.log(f"[red]Error when connecting to kalavai service: {str(e)}")
-                return
-        
+    rows = [[job.owner, job.name, job.workers, job.endpoint] for job in details]
+    
     console.print(
         generate_table(columns=columns, rows=rows, end_sections=range(len(rows)))
     )
@@ -1774,26 +1749,19 @@ def job__logs(name, *others, pod_name=None, stream=False, tail=100, force_namesp
     
     if force_namespace is not None:
         console.log("[WARNING][yellow]--force-namespace [white]requires an admin key. Request will fail if you are not an admin.")
-    
-    data = {
-        "label": TEMPLATE_LABEL,
-        "value": name,
-        "tail": tail
-    }
-    if force_namespace is not None:
-        data["force_namespace"] = force_namespace
+
+    all_logs = fetch_job_logs(
+        job_name=name,
+        pod_name=pod_name,
+        force_namespace=force_namespace,
+        tail=tail)
+    if "error" in all_logs:
+        console.log(f"[red]{all_logs}")
+        return
     while True:
         try:
-            # send tail as parameter (fetch only last _tail_ lines)
-            result = request_to_server(
-                method="post",
-                endpoint="/v1/get_logs_for_label",
-                data=data,
-                server_creds=USER_LOCAL_SERVER_FILE,
-                user_cookie=USER_COOKIE
-            )
             if not stream:
-                for pod, logs in result.items():
+                for pod, logs in all_logs.items():
                     if pod_name is not None and pod_name != pod:
                         continue
                     console.log(f"[yellow]Pod {pod}")
@@ -1801,7 +1769,7 @@ def job__logs(name, *others, pod_name=None, stream=False, tail=100, force_namesp
                 break
             else:
                 os.system("clear")
-                for pod, logs in result.items():
+                for pod, logs in all_logs.items():
                     if pod_name is not None and pod_name != pod:
                         continue
                     print(f"Pod {pod}")
@@ -1809,10 +1777,7 @@ def job__logs(name, *others, pod_name=None, stream=False, tail=100, force_namesp
                 time.sleep(1)
         except KeyboardInterrupt:
             break
-        except Exception as e:
-            console.log(f"[red]Error when connecting to kalavai service: {str(e)}")
-            console.log(f"Check if {name} is running with [yellow]kalavai job list")
-            return
+
 
 @arguably.command
 def job__manifest(*others, name, force_namespace: str=None):

kalavai_client/cluster.py

@@ -133,6 +133,9 @@ class dockerCluster(Cluster):
         if not os.path.isfile(self.compose_file):
             return False
         status = self.container_name in run_cmd(f"docker compose -f {self.compose_file} ps --services --status=running").decode()
+        if not status:
+            return False
+        status = (0 == os.system(f'docker exec {self.container_name} ps aux | grep -v grep | grep -E "k3s (server|agent)"'))
         return status
 
     def is_seed_node(self):

kalavai_client/core.py

@@ -0,0 +1,227 @@
+from collections import defaultdict
+import math
+
+from pydantic import BaseModel
+
+from kalavai_client.utils import (
+    request_to_server,
+    load_server_info
+)
+from kalavai_client.env import (
+    USER_COOKIE,
+    USER_LOCAL_SERVER_FILE,
+    TEMPLATE_LABEL,
+    SERVER_IP_KEY
+)
+
+class Job(BaseModel):
+    owner: str = None
+    name: str = None
+    workers: str = None
+    endpoint: str = None
+
+class DeviceStatus(BaseModel):
+    name: str
+    memory_pressure: bool
+    disk_pressure: bool
+    pid_pressure: bool
+    ready: bool
+    unschedulable: bool
+
+class GPU(BaseModel):
+    node: str
+    available: int
+    total: int
+    ready: bool
+    model: str
+
+
+def fetch_resources():
+    try:
+        total = request_to_server(
+            method="get",
+            endpoint="/v1/get_cluster_total_resources",
+            data={},
+            server_creds=USER_LOCAL_SERVER_FILE,
+            user_cookie=USER_COOKIE
+        )
+        available = request_to_server(
+            method="get",
+            endpoint="/v1/get_cluster_available_resources",
+            data={},
+            server_creds=USER_LOCAL_SERVER_FILE,
+            user_cookie=USER_COOKIE
+        )
+    except Exception as e:
+        return {"error": str(e)}
+        
+    return {"total": total, "available": available}
+
+def fetch_job_names():
+    data = {
+        "group": "batch.volcano.sh",
+        "api_version": "v1alpha1",
+        "plural": "jobs"
+    }
+    try:
+        jobs = request_to_server(
+            method="post",
+            endpoint="/v1/get_objects_of_type",
+            data=data,
+            server_creds=USER_LOCAL_SERVER_FILE,
+            user_cookie=USER_COOKIE
+        )
+        all_jobs = []
+        for ns, ds in jobs.items():
+            all_jobs.extend([Job(owner=ns, name=d["metadata"]["labels"][TEMPLATE_LABEL]) for d in ds["items"]])
+    except Exception as e:
+        return {"error": str(e)}
+    
+    return all_jobs  
+
+def fetch_job_details(jobs: list[Job]):
+    """Get job details. A job is a dict:
+    {
+        "namespace": ns,
+        "name": name
+    }
+    """
+    job_details = []
+    for job in jobs:
+        namespace = job.owner
+        deployment = job.name
+        try:
+            # get pod statuses
+            data = {
+                "label": TEMPLATE_LABEL,
+                "value": deployment
+            }
+            result = request_to_server(
+                method="post",
+                endpoint="/v1/get_pods_status_for_label",
+                data=data,
+                server_creds=USER_LOCAL_SERVER_FILE,
+                user_cookie=USER_COOKIE
+            )
+            workers_status = defaultdict(int)
+            for ns, ss in result.items():
+                if ns != namespace: # same job name, different namespace
+                    continue
+                for _, values in ss.items():
+                    workers_status[values["status"]] += 1
+            workers = "\n".join([f"{k}: {v}" for k, v in workers_status.items()])
+            # get URL details
+            data = {
+                "label": TEMPLATE_LABEL,
+                "value": deployment,
+                "types": ["NodePort"]
+            }
+            result = request_to_server(
+                method="post",
+                endpoint="/v1/get_ports_for_services",
+                data=data,
+                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"]]
+
+            urls = [f"http://{load_server_info(data_key=SERVER_IP_KEY, file=USER_LOCAL_SERVER_FILE)}:{node_port}" for node_port in node_ports]
+            job_details.append(
+                Job(owner=namespace,
+                    name=deployment,
+                    workers=workers,
+                    endpoint="\n".join(urls))
+            )
+
+        except Exception as e:
+            return {"error": str(e)}
+    
+    return job_details
+
+def fetch_devices():
+    """Load devices status info for all hosts"""
+    try:
+        data = request_to_server(
+            method="get",
+            endpoint="/v1/get_nodes",
+            data={},
+            server_creds=USER_LOCAL_SERVER_FILE,
+            user_cookie=USER_COOKIE
+        )
+        devices = []
+        for node, status in data.items():
+            devices.append(
+                DeviceStatus(
+                    name=node,
+                    memory_pressure=status["MemoryPressure"],
+                    disk_pressure=status["DiskPressure"],
+                    pid_pressure=status["PIDPressure"],
+                    ready=status["Ready"],
+                    unschedulable=status["unschedulable"]
+                )
+            )
+        return devices
+
+    except Exception as e:
+        return {"error": str(e)}
+
+def fetch_job_logs(job_name, force_namespace=None, pod_name=None, tail=100):
+    data = {
+        "label": TEMPLATE_LABEL,
+        "value": job_name,
+        "tail": tail
+    }
+    if force_namespace is not None:
+        data["force_namespace"] = force_namespace
+    try:
+        # send tail as parameter (fetch only last _tail_ lines)
+        all_logs = request_to_server(
+            method="post",
+            endpoint="/v1/get_logs_for_label",
+            data=data,
+            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}
+
+    except Exception as e:
+        return {"error": str(e)}
+    
+
+def load_gpu_models():
+    data = request_to_server(
+        method="post",
+        endpoint="/v1/get_node_gpus",
+        data={},
+        server_creds=USER_LOCAL_SERVER_FILE,
+        user_cookie=USER_COOKIE
+    )
+    return data.items()
+
+def fetch_gpus(available=False):
+    try:
+        data = load_gpu_models()
+        all_gpus = []
+        for node, gpus in data:
+            row_gpus = []
+            for gpu in gpus["gpus"]:
+                status = gpu["ready"] if "ready" in gpu else True
+                if available and not status:
+                    continue
+                row_gpus.append( (f"{gpu['model']} ({math.floor(int(gpu['memory'])/1000)} GBs)", str(status)))
+            if len(row_gpus) > 0:
+                models, statuses = zip(*row_gpus)
+                #rows.append([node, "\n".join(statuses), "\n".join(models), str(gpus["available"]), str(gpus["capacity"])])
+                all_gpus.extend([
+                    GPU(
+                        node=node,
+                        ready=status,
+                        model=model,
+                        available=gpus["available"],
+                        total=gpus["capacity"]
+                    ) for model, status in zip(models, statuses)
+                ])
+        return all_gpus
+
+    except Exception as e:
+        return {"error": str(e)}

kalavai_client/env.py

@@ -0,0 +1,19 @@
+import os
+from pathlib import Path
+
+
+def user_path(relative_path, create_path=False):
+    """Transform a relative path into the user's cache folder path"""
+    base = os.path.expanduser("~")
+    kalavai_user_path = os.path.join(base, ".cache/kalavai")
+    full_path = os.path.join(kalavai_user_path, relative_path)
+    if create_path:
+        Path(full_path).mkdir(parents=True, exist_ok=True)
+    
+    return full_path
+
+
+USER_LOCAL_SERVER_FILE = user_path(".server")
+USER_COOKIE = user_path(".user_cookie.pkl")
+TEMPLATE_LABEL = "kalavai.job.name"
+SERVER_IP_KEY = "server_ip"

kalavai_client/utils.py

@@ -14,6 +14,9 @@ import yaml
 
 
 from kalavai_client.auth import KalavaiAuthClient
+from kalavai_client.env import (
+    SERVER_IP_KEY
+)
 
 
 GITHUB_ORG = "kalavai-net"
@@ -22,7 +25,6 @@ GITHUB_TEMPLATE_PATH = "templates"
 USER_NODE_LABEL_KEY = "user_node_label"
 CLUSTER_IP_KEY = "cluster_ip"
 CLUSTER_TOKEN_KEY = "cluster_token"
-SERVER_IP_KEY = "server_ip"
 NODE_NAME_KEY = "node_name"
 PUBLIC_LOCATION_KEY = "public_location"
 CLUSTER_NAME_KEY = "cluster_name"
@@ -284,8 +286,11 @@ def request_to_server(
         json=data,
         headers=headers
     )
-    result = response.json()
-    return result
+    try:
+        result = response.json()
+        return result
+    except Exception as e:
+        raise ValueError(f"Error with HTTP request: {response.text}\n{str(e)}")
 
 
 def generate_table(columns, rows, end_sections=None):
@@ -394,16 +399,6 @@ def resource_path(relative_path: str):
         return None
     return resource
 
-def user_path(relative_path, create_path=False):
-    """Transform a relative path into the user's cache folder path"""
-    base = os.path.expanduser("~")
-    kalavai_user_path = os.path.join(base, ".cache/kalavai")
-    full_path = os.path.join(kalavai_user_path, relative_path)
-    if create_path:
-        Path(full_path).mkdir(parents=True, exist_ok=True)
-    
-    return full_path
-
 def safe_remove(filepath, force=True):
     if not os.path.exists(filepath):
         return

{kalavai_client-0.5.10.dist-info → kalavai_client-0.5.13.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: kalavai-client
-Version: 0.5.10
+Version: 0.5.13
 Summary: Client app for kalavai platform
 License: Apache-2.0
 Keywords: LLM,platform
@@ -89,6 +89,9 @@ https://github.com/user-attachments/assets/0d2316f3-79ea-46ac-b41e-8ef720f52672
 
 ### News updates
 
+<img src="docs/docs/assets/images/DeepSeek-Emblem.png" width="100">
+
+- 6 February 2025: 🔥🔥🔥 Access  **DeepSeek R1 model for free** when you join our [public LLM pool](https://kalavai-net.github.io/kalavai-client/public_llm_pool/)
 - 31 January 2025: `kalavai-client` is now a [PyPI package](https://pypi.org/project/kalavai-client/), easier to install than ever!
 - 27 January 2025: Support for accessing pools from remote computers
 - 9 January 2025: Added support for [Aphrodite Engine](https://github.com/aphrodite-engine/aphrodite-engine) models
@@ -140,7 +143,7 @@ From release **v0.5.0, you can now install `kalavai-client` in non-worker comput
 For workers sharing resources with the pool:
 
 - A laptop, desktop or Virtual Machine
-- Docker engine installed (for [linux](https://docs.docker.com/engine/install/), [Windows and MacOS](https://docs.docker.com/desktop/)) with [privilege access](https://docs.docker.com/engine/containers/run/#runtime-privilege-and-linux-capabilities).
+- Docker engine installed (for [linux](https://docs.docker.com/engine/install/ubuntu/), [Windows and MacOS](https://docs.docker.com/desktop/)) with [privilege access](https://docs.docker.com/engine/containers/run/#runtime-privilege-and-linux-capabilities).
 
 > **Support for Windows and MacOS workers is experimental**: kalavai workers run on docker containers that require access to the host network interfaces, thus systems that do not support containers natively (Windows and MacOS) may have difficulties finding each other.
 

kalavai_client-0.5.13.dist-info/RECORD

@@ -0,0 +1,22 @@
+kalavai_client/__init__.py,sha256=OKM-UDxm0absUf9IgE89lC_PpDG9RbBD4It-hbz8ORM,23
+kalavai_client/__main__.py,sha256=WQUfxvRsBJH5gsCJg8pLz95QnZIj7Ol8psTO77m0QE0,73
+kalavai_client/assets/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+kalavai_client/assets/apps.yaml,sha256=yC-vtYTPE960KUQihTk5pee8xZz9RD8Reuyh1nSpRWk,5981
+kalavai_client/assets/apps_values.yaml,sha256=CjKVelPQHd-hm-DTMEuya92feKiphU9mh3HrosLYYPE,1676
+kalavai_client/assets/docker-compose-template.yaml,sha256=mo8LUam9-AzB_0w72wTyMyreKr4Ns-pxZGc4GVWcUHA,2747
+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=QsBh28L2LwjBBK6pTUE4Xu36lLDTyetyU1YfS1Hbb6g,1717
+kalavai_client/cli.py,sha256=_oUZAYV397_-BQAHsTcyK0pkyK5iusdyYrJU5z6lecM,66312
+kalavai_client/cluster.py,sha256=z9HCD6ZUemjabcDenszQhqB_IUiVi_vpFbfAkKwHiEU,12292
+kalavai_client/core.py,sha256=Vb-5MBHjpuR590FIDOnytJpP1Xjt7hYqehPV2rh6P68,6863
+kalavai_client/env.py,sha256=RAi37vJtIGfPR25PNxZYMZNkkEKR4AyUPN_htFiFesM,575
+kalavai_client/utils.py,sha256=kQk_1QOs8u08rcfhkcfo_oC-cZzww0cij-1R_jK1ER8,12185
+kalavai_client-0.5.13.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+kalavai_client-0.5.13.dist-info/METADATA,sha256=fQus2g5Q39Wu_HglzdiOGnalva3GcSG48o5iyyikbt4,14347
+kalavai_client-0.5.13.dist-info/WHEEL,sha256=IYZQI976HJqqOpQU6PHkJ8fb3tMNBFjg-Cn-pwAbaFM,88
+kalavai_client-0.5.13.dist-info/entry_points.txt,sha256=9T6D45gxwzfVbglMm1r6XPdXuuZdHfy_7fCeu2jUphc,50
+kalavai_client-0.5.13.dist-info/RECORD,,

kalavai_client-0.5.10.dist-info/RECORD

@@ -1,20 +0,0 @@
-kalavai_client/__init__.py,sha256=PoEyxKkxe5kSNJwy0utYwv_ANRK98k-p4OLs19lJaBA,23
-kalavai_client/__main__.py,sha256=WQUfxvRsBJH5gsCJg8pLz95QnZIj7Ol8psTO77m0QE0,73
-kalavai_client/assets/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-kalavai_client/assets/apps.yaml,sha256=_8BgT9F611c8uZJvhTE_0CLbQqnLaUQosqxZjzOslXQ,5979
-kalavai_client/assets/apps_values.yaml,sha256=CjKVelPQHd-hm-DTMEuya92feKiphU9mh3HrosLYYPE,1676
-kalavai_client/assets/docker-compose-template.yaml,sha256=gJ0NkhcG2c-gZPmSd385dadrXkZrWruTJkiaxcaKkQ0,2725
-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=QsBh28L2LwjBBK6pTUE4Xu36lLDTyetyU1YfS1Hbb6g,1717
-kalavai_client/cli.py,sha256=4qTZuYNFhsdJbnER-MSBJHPNgJc_lWzbWR0Bj2YeQe0,68889
-kalavai_client/cluster.py,sha256=fULTAad4KXEGeWZmp4_VBoBwT5eED_HOBUsXIKmf0CU,12119
-kalavai_client/utils.py,sha256=RTJNgY7ho52Q5WzV68ZK5uHNYRAwXkelIr1PmBLBJsk,12420
-kalavai_client-0.5.10.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-kalavai_client-0.5.10.dist-info/METADATA,sha256=u_E2mWeRmmVE2pWCToOVwwxdgu--UWkzhkImQb67qy4,14101
-kalavai_client-0.5.10.dist-info/WHEEL,sha256=IYZQI976HJqqOpQU6PHkJ8fb3tMNBFjg-Cn-pwAbaFM,88
-kalavai_client-0.5.10.dist-info/entry_points.txt,sha256=9T6D45gxwzfVbglMm1r6XPdXuuZdHfy_7fCeu2jUphc,50
-kalavai_client-0.5.10.dist-info/RECORD,,