kalavai-client 0.6.12__py3-none-any.whl → 0.6.13__py3-none-any.whl

kalavai_client/__init__.py

@@ -1,2 +1,2 @@
 
-__version__ = "0.6.12"
+__version__ = "0.6.13"

kalavai_client/cli.py

@@ -186,7 +186,9 @@ def input_gpus(non_interactive=False):
     try:
         has_gpus = check_gpu_drivers()
         if has_gpus:
-            max_gpus = int(run_cmd("nvidia-smi -L | wc -l").decode())
+            max_gpus = len(
+                [r for r in run_cmd("nvidia-smi -L").decode().split("\n") if len(r.strip())>0]
+            )
             if non_interactive:
                 num_gpus = max_gpus
             else:

kalavai_client/cluster.py

@@ -1,4 +1,5 @@
 import os
+import platform
 import time
 from pathlib import Path
 from abc import ABC, abstractmethod
@@ -96,7 +97,7 @@ class dockerCluster(Cluster):
         # wait for container to be setup
         while True:
             try:
-                run_cmd(f"docker cp {self.container_name}:/etc/rancher/k3s/k3s.yaml {self.kubeconfig_file} >/dev/null 2>&1")
+                run_cmd(f"docker cp {self.container_name}:/etc/rancher/k3s/k3s.yaml {self.kubeconfig_file}", hide_output=True)
                 break
             except:
                 pass
@@ -115,15 +116,15 @@ class dockerCluster(Cluster):
     def update_dependencies(self, dependencies_file=None, debug=False, retries=3):
         if dependencies_file is not None:
             self.dependencies_file = dependencies_file
-        if debug:
-            output = ""
-        else:
-            output = " >/dev/null 2>&1"
         while True:
             try:
                 home = user_path("")
-                run_cmd(f"docker run --rm --net=host -v {home}:{home} ghcr.io/helmfile/helmfile:v0.169.2 helmfile sync --file {self.dependencies_file} --kubeconfig {self.kubeconfig_file} {output}")
-                #run_cmd(f"helmfile sync --file {self.dependencies_file} --kubeconfig {self.kubeconfig_file} {output}")
+                # convert path on host to path on container (will be different in windows os)
+                target_path = "/cache/kalavai"
+                kubeconfig_path = f"{target_path}/{Path(self.kubeconfig_file).name}"
+                dependencies_path = f"{target_path}/{Path(self.dependencies_file).name}"
+
+                run_cmd(f"docker run --rm --net=host -v {home}:{target_path} ghcr.io/helmfile/helmfile:v0.169.2 helmfile sync --file {dependencies_path} --kubeconfig {kubeconfig_path}", hide_output=not debug)
                 break
             except Exception as e:
                 if retries > 0:
@@ -142,11 +143,18 @@ class dockerCluster(Cluster):
     def is_agent_running(self):
         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:
+        try:
+            status = self.container_name in run_cmd(f"docker compose -f {self.compose_file} ps --services --status=running").decode()
+            if not status:
+                return False
+            if "windows" in platform.system().lower():
+                status = (0 == os.system(f'docker exec {self.container_name} ps aux | findstr /n /c:"k3s server" /c:"k3s agent"'))
+            else:
+                status = (0 == os.system(f'docker exec {self.container_name} ps aux | grep -v grep | grep -E "k3s (server|agent)"'))
+            return status
+        except Exception as e:
+            print(f"Error when checking agent. Is Docker installed and running?\n\n{str(e)}")
             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):
         if not os.path.isfile(self.compose_file):
@@ -154,7 +162,7 @@ class dockerCluster(Cluster):
         if not self.is_agent_running():
             return False
         try:
-            run_cmd(f"docker container exec {self.container_name} cat /var/lib/rancher/k3s/server/node-token >/dev/null 2>&1")
+            run_cmd(f"docker container exec {self.container_name} cat /var/lib/rancher/k3s/server/node-token", hide_output=True)
             return True
         except:
             return False
@@ -162,8 +170,12 @@ class dockerCluster(Cluster):
     def is_cluster_init(self):
         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 --all").decode()
-        return status
+        try:
+            status = self.container_name in run_cmd(f"docker compose -f {self.compose_file} ps --services --all").decode()
+            return status
+        except Exception as e:
+            print(f"Error when checking cluster. Is Docker installed and running?\n\n{str(e)}")
+            return False
 
     def pause_agent(self):
         status = False
@@ -177,7 +189,6 @@ class dockerCluster(Cluster):
     def restart_agent(self):
         try:
             run_cmd(f'docker compose -f {self.compose_file} start')
-
         except:
             pass
         time.sleep(5)
@@ -186,7 +197,6 @@ class dockerCluster(Cluster):
     def get_cluster_token(self):
         if self.is_seed_node():
             return run_cmd(f"docker container exec {self.container_name} cat /var/lib/rancher/k3s/server/node-token").decode()
-            #return run_cmd("sudo k3s token create --kubeconfig /etc/rancher/k3s/k3s.yaml --ttl 0").decode()
         else:
             return None
     
@@ -231,7 +241,7 @@ class k3sCluster(Cluster):
             flannel_iface = f"--flannel-iface {self.default_flannel_iface}"
         else:
             flannel_iface = ""
-        run_cmd(f'curl -sfL https://get.k3s.io | INSTALL_K3S_VERSION="{self.kube_version}" INSTALL_K3S_EXEC="server --node-ip {ip_address} --node-external-ip {ip_address} {flannel_iface} --flannel-backend wireguard-native {node_labels}" sh - >/dev/null 2>&1')
+        run_cmd(f'curl -sfL https://get.k3s.io | INSTALL_K3S_VERSION="{self.kube_version}" INSTALL_K3S_EXEC="server --node-ip {ip_address} --node-external-ip {ip_address} {flannel_iface} --flannel-backend wireguard-native {node_labels}" sh - ', hide_output=True)
         run_cmd(f"sudo cp /etc/rancher/k3s/k3s.yaml {self.kubeconfig_file}")
         run_cmd(f"sudo chown $USER {self.kubeconfig_file}")
 
@@ -245,8 +255,8 @@ class k3sCluster(Cluster):
             flannel_iface = f"--flannel-iface {self.default_flannel_iface}"
         else:
             flannel_iface = ""
-        command = f'curl -sfL https://get.k3s.io | INSTALL_K3S_VERSION="{self.kube_version}" INSTALL_K3S_EXEC="agent --token {token} --server https://{url}:6443 --node-name {node_name} --node-ip {ip_address} --node-external-ip {ip_address} {flannel_iface} {node_labels}" sh - >/dev/null 2>&1'
-        run_cmd(command)
+        command = f'curl -sfL https://get.k3s.io | INSTALL_K3S_VERSION="{self.kube_version}" INSTALL_K3S_EXEC="agent --token {token} --server https://{url}:6443 --node-name {node_name} --node-ip {ip_address} --node-external-ip {ip_address} {flannel_iface} {node_labels}" sh - '
+        run_cmd(command, hide_output=True)
         
 
     def update_dependencies(self, dependencies_file=None, debug=False, retries=3):
@@ -270,13 +280,13 @@ class k3sCluster(Cluster):
 
     def remove_agent(self):
         try:
-            run_cmd('/usr/local/bin/k3s-uninstall.sh >/dev/null 2>&1')
-            run_cmd('sudo rm -r /etc/rancher/node/ >/dev/null 2>&1')
+            run_cmd('/usr/local/bin/k3s-uninstall.sh', hide_output=True)
+            run_cmd('sudo rm -r /etc/rancher/node/', hide_output=True)
             return True
         except:
             pass
         try:
-            run_cmd('/usr/local/bin/k3s-agent-uninstall.sh >/dev/null 2>&1')
+            run_cmd('/usr/local/bin/k3s-agent-uninstall.sh', hide_output=True)
             return True
         except:
             pass
@@ -296,12 +306,12 @@ class k3sCluster(Cluster):
     def pause_agent(self):
         status = False
         try:
-            run_cmd('sudo systemctl stop k3s >/dev/null 2>&1')
+            run_cmd('sudo systemctl stop k3s', hide_output=True)
             status = True
         except:
             pass
         try:
-            run_cmd('sudo systemctl stop k3s-agent >/dev/null 2>&1')
+            run_cmd('sudo systemctl stop k3s-agent', hide_output=True)
             status = True
         except:
             pass
@@ -309,11 +319,11 @@ class k3sCluster(Cluster):
 
     def restart_agent(self):
         try:
-            run_cmd('sudo systemctl start k3s >/dev/null 2>&1')
+            run_cmd('sudo systemctl start k3s', hide_output=True)
         except:
             pass
         try:
-            run_cmd('sudo systemctl start k3s-agent >/dev/null 2>&1')
+            run_cmd('sudo systemctl start k3s-agent', hide_output=True)
         except:
             pass
         return self.is_agent_running()

kalavai_client/core.py

@@ -33,6 +33,7 @@ from kalavai_client.utils import (
     get_public_seeds,
     load_template,
     is_storage_compatible,
+    get_max_gpus,
     NODE_NAME_KEY,
     MANDATORY_TOKEN_FIELDS,
     PUBLIC_LOCATION_KEY,
@@ -157,7 +158,7 @@ def check_seed_compatibility():
     logs = []
     # docker
     try:
-        run_cmd("docker version >/dev/null 2>&1")
+        run_cmd("docker ps", hide_output=True)
     except:
         logs.append("[red]Docker not installed. Install instructions:\n")
         logs.append("   Linux: https://docs.docker.com/engine/install/\n")
@@ -170,7 +171,7 @@ def check_worker_compatibility():
     logs = []
     # docker
     try:
-        run_cmd("docker version >/dev/null 2>&1")
+        run_cmd("docker ps", hide_output=True)
     except:
         logs.append("[red]Docker not installed. Install instructions:\n")
         logs.append("   Linux: https://docs.docker.com/engine/install/\n")
@@ -594,16 +595,6 @@ def attach_to_pool(token, node_name=None):
 
     return cluster_name
 
-def get_max_gpus():
-    try:
-        has_gpus = check_gpu_drivers()
-        if has_gpus:
-            return int(run_cmd("nvidia-smi -L | wc -l").decode())
-        else:
-            return 0
-    except:
-        return 0
-
 def generate_worker_package(num_gpus=0, node_name=None, ip_address="0.0.0.0", storage_compatible=True):
     # get pool data from token  
     token = get_pool_token(mode=TokenType.WORKER)
@@ -772,10 +763,9 @@ def create_pool(
         node_name=node_name,
         node_labels=node_labels
     )
-    
+
     # start server
     CLUSTER.start_seed_node()
-    
     while not CLUSTER.is_agent_running():
         time.sleep(10)
     

kalavai_client/utils.py

@@ -73,17 +73,20 @@ KALAVAI_AUTH = KalavaiAuth(
 
 ####### Methods to check OS compatibility ########
 def check_gpu_drivers():
-    value = run_cmd("command -v nvidia-smi")
-    if len(value.decode("utf-8")) == 0:
-        # no nvidia installed, no need to check nvidia any further
-        return False
-    else:
-        # check drivers are set correctly
-        try:
-            value = run_cmd("nvidia-smi")
-            return True
-        except:
-            raise ("Nvidia not configured properly. Please check your drivers are installed and configured")
+    value = run_cmd("nvidia-smi", hide_output=True)
+    return len(value.decode("utf-8")) == 0
+
+def get_max_gpus():
+    try:
+        has_gpus = check_gpu_drivers()
+        if has_gpus:
+            return len(
+                [r for r in run_cmd("nvidia-smi -L").decode().split("\n") if len(r.strip())>0]
+            )
+        else:
+            return 0
+    except:
+        return 0
 
 def is_storage_compatible():
     """
@@ -92,6 +95,9 @@ def is_storage_compatible():
     Exclude: WSL
     """
     try:
+        import platform
+        if "windows" in platform.system().lower():
+            return True
         flagged = any([
             "microsoft" in run_cmd("cat /proc/version").decode().lower()
         ])
@@ -231,9 +237,17 @@ def validate_poolconfig(poolconfig_file):
             return False
     return True
 
-def run_cmd(command):
+def run_cmd(command, hide_output=False):
     try:
-        return_value = subprocess.check_output(command, shell=True, executable="/bin/bash")
+        import platform
+        if "windows" in platform.system().lower():
+            if hide_output:
+                command = command + " > $nul 2>&1"
+            return_value = subprocess.check_output(command, shell=True)
+        else:
+            if hide_output:
+                command = command + " >/dev/null 2>&1"
+            return_value = subprocess.check_output(command, shell=True, executable="/bin/bash")
         return return_value
     except OSError as error:
         return error # for exit code

{kalavai_client-0.6.12.dist-info → kalavai_client-0.6.13.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: kalavai-client
-Version: 0.6.12
+Version: 0.6.13
 Summary: Client app for kalavai platform
 License: Apache-2.0
 Keywords: LLM,platform

{kalavai_client-0.6.12.dist-info → kalavai_client-0.6.13.dist-info}/RECORD

@@ -1,4 +1,4 @@
-kalavai_client/__init__.py,sha256=RnWDr4VWcRjJDXe1q7REqBSUCBo44AvIa_8rGwJtT44,23
+kalavai_client/__init__.py,sha256=RtAfZnVOn5ru1jOt5ukVyb0s2GiJ7s39Qp5_KBjCL-A,23
 kalavai_client/__main__.py,sha256=WQUfxvRsBJH5gsCJg8pLz95QnZIj7Ol8psTO77m0QE0,73
 kalavai_client/assets/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 kalavai_client/assets/apps.yaml,sha256=AhTA3VZI27y05xHoHJCA9nvGnk8sWMhFDruBI2is3LM,6365
@@ -13,13 +13,13 @@ kalavai_client/assets/user_workspace_values.yaml,sha256=G0HOzQUxrDMCwuW9kbWUZaKM
 kalavai_client/auth.py,sha256=EB3PMvKUn5_KAQkezkEHEt-OMZXyfkZguIQlUFkEHcA,3243
 kalavai_client/bridge_api.py,sha256=ZLyFOOz_o4agm-7DrHzoSBFrH65y__hZUoEe4diBTOA,7557
 kalavai_client/bridge_models.py,sha256=GbIaqGFAVs-3ikVUQZldwTTc06SsxmP6iAifH0oVDro,1219
-kalavai_client/cli.py,sha256=tBn3l12XhOJF17cWeX84xm10HPG7rSzalyOyugzcLhw,46900
-kalavai_client/cluster.py,sha256=ojUBXp2bR3hVyikIEkiGDbXvQfhBXBSk_mCqLxvyP0c,12943
-kalavai_client/core.py,sha256=QOzNSxNJZ8tXhUTUPZUEFI3PX8UlVVnONYqVA61NU6U,34268
+kalavai_client/cli.py,sha256=mmwLqqSYfl9k6vqveMcbHTq7g5FFd84YUUQCSH4J0k0,46967
+kalavai_client/cluster.py,sha256=Z2PIXbZuSAv9xmw-MyZP1M41BpVMpirLzG51bqGA-zc,13548
+kalavai_client/core.py,sha256=u8a4uYqGS0mMJh0ArcXG2hwp2uDUSuwM5ROGXRQkHZg,34051
 kalavai_client/env.py,sha256=YsfZj7LWf6ABquDsoIFFkXCFYwenpDk8zVnGsf7qv98,2823
-kalavai_client/utils.py,sha256=EmjWVLjTBILjw87_6ih_v5CbVqihsYHSKxfD-C_Z-es,12276
-kalavai_client-0.6.12.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-kalavai_client-0.6.12.dist-info/METADATA,sha256=ShcxZWQofBxd_VVg6LbOGaPiPDrWSix_yLIbhn7MiJ4,13354
-kalavai_client-0.6.12.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
-kalavai_client-0.6.12.dist-info/entry_points.txt,sha256=9T6D45gxwzfVbglMm1r6XPdXuuZdHfy_7fCeu2jUphc,50
-kalavai_client-0.6.12.dist-info/RECORD,,
+kalavai_client/utils.py,sha256=yHz9n4hGwhpGUTc1ZcG5JHtesoUEzSngmOSTvu2jDic,12674
+kalavai_client-0.6.13.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+kalavai_client-0.6.13.dist-info/METADATA,sha256=YUeJh_a81RDslXa8QNlgNwmXfQLHxAbqcCK2HWP6t-4,13354
+kalavai_client-0.6.13.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
+kalavai_client-0.6.13.dist-info/entry_points.txt,sha256=9T6D45gxwzfVbglMm1r6XPdXuuZdHfy_7fCeu2jUphc,50
+kalavai_client-0.6.13.dist-info/RECORD,,