switch to proxmoxer waiter thingy and wait for QEMU agent

functions.py

@@ -1,4 +1,5 @@
 import logging
+import time
 
 from proxmoxer import ProxmoxAPI
 
@@ -55,12 +56,17 @@ def delete_vm(
     proxmox: ProxmoxAPI,
     node: str,
     vmid: int,
-) -> None:
-    proxmox.nodes(node).qemu(vmid).delete(
-        node=node,
-        vmid=vmid,
+) -> str:
+    upid: str = (
+        proxmox.nodes(node)
+        .qemu(vmid)
+        .delete(
+            node=node,
+            vmid=vmid,
+        )
     )
     logger.info("VM %s was deleted successfully!", vmid)
+    return upid
 
 
 def clone_template(
@@ -107,17 +113,27 @@ def get_vm_ip(
     node: str,
     vmid: int,
 ) -> str:
-    res: dict = (
-        proxmox.nodes(node).qemu(vmid).agent("network-get-interfaces").get()
-    )  # pyright: ignore
+    done = False
     vm_ip_address = None
-    for net in res["result"]:
-        if net["name"] == "eth0" or net["name"] == "ens19":
-            for ip in net["ip-addresses"]:
-                if ip["ip-address-type"] == "ipv4":
-                    vm_ip_address = ip["ip-address"]
+    while not done:
+        try:
+            res: dict = (
+                proxmox.nodes(node).qemu(vmid).agent("network-get-interfaces").get()
+            )  # pyright: ignore
+            for net in res["result"]:
+                if net["name"] == "eth0" or net["name"] == "ens19":
+                    for ip in net["ip-addresses"]:
+                        if ip["ip-address-type"] == "ipv4":
+                            vm_ip_address = ip["ip-address"]
+                            done = True
+        except KeyError:
+            time.sleep(1)
+        except Exception as e:
+            raise RuntimeError(
+                f"Could not get the ip address of VM {vmid} on node {node}"
+            ) from e
 
     if vm_ip_address is None:
-        raise ValueError(f"Could not get the ip address of VM {vmid} on node {node}")
+        raise RuntimeError(f"Could not get the ip address of VM {vmid} on node {node}")
 
     return vm_ip_address

main.py

@@ -1,15 +1,15 @@
-#!/usr/bin/python3
+#!/usr/bin/env python3
 
+import argparse
 import logging
 import os
-import sys
-from time import sleep
-from typing import Callable
+import time
 
 import urllib3
 import yaml
 from dotenv import load_dotenv
-from proxmoxer import ProxmoxAPI
+from proxmoxer import ProxmoxAPI, ResourceException
+from proxmoxer.tools import Tasks
 
 from functions import clone_template, delete_vm, get_vm_ip
 
@@ -28,15 +28,19 @@ PROXMOX_USER_FULL: str | None = os.environ.get("PROXMOX_USER_FULL")
 PROXMOX_PASSWORD: str | None = os.environ.get("PROXMOX_PASSWORD")
 
 
-def wait_status(
-    funcs: list[Callable[[], dict[str, str]]], status: str, sleep_time: float = 60
-) -> None:
-    statuses = [False] * len(funcs)
-    while not all(statuses):
-        logger.debug(statuses)
-        sleep(sleep_time)
-        for i in range(len(funcs)):
-            statuses[i] = funcs[i]()["status"] == status
+def tasks_wait_completion(proxmox: ProxmoxAPI, upids: list[str]) -> None:
+    for upid in upids:
+        Tasks.blocking_status(proxmox, upid)
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "branch",
+        help="used to pick the appropriate VM template to clone",
+        choices=["sisyphus", "c10f2", "p10"],
+    )
+    return parser.parse_args()
 
 
 def main() -> None:
@@ -45,6 +49,8 @@ def main() -> None:
     assert PROXMOX_USER_FULL is not None
     assert PROXMOX_PASSWORD is not None
 
+    args = parse_args()
+
     proxmox = ProxmoxAPI(
         PROXMOX_HOST,
         user=PROXMOX_USER_FULL,
@@ -57,10 +63,7 @@ def main() -> None:
     # node = min(nodes, key=lambda node: node['disk'])['node']
     node: str = "pve05"
 
-    assert len(sys.argv) >= 2
-    branch: str = sys.argv[1]
-
-    match branch:
+    match args.branch:
         case "sisyphus":
             template_id: int = 374
         case "p10":
@@ -92,15 +95,24 @@ def main() -> None:
 
     if delete != "0":
         logger.info("Stopping VMs %s...", vm_ids)
-        proxmox.nodes(node).qemu(vm_ids["master"]).status.stop.post()
-        proxmox.nodes(node).qemu(vm_ids["worker1"]).status.stop.post()
-        proxmox.nodes(node).qemu(vm_ids["worker2"]).status.stop.post()
-        sleep(15)
+        tasks_wait_completion(
+            proxmox,
+            [
+                proxmox.nodes(node).qemu(vm_ids["master"]).status.stop.post(),
+                proxmox.nodes(node).qemu(vm_ids["worker1"]).status.stop.post(),
+                proxmox.nodes(node).qemu(vm_ids["worker2"]).status.stop.post(),
+            ],
+        )
+
         logger.info("Deleting VMs %s...", vm_ids)
-        delete_vm(proxmox, node, vm_ids["master"])
-        delete_vm(proxmox, node, vm_ids["worker1"])
-        delete_vm(proxmox, node, vm_ids["worker2"])
-        sleep(15)
+        tasks_wait_completion(
+            proxmox,
+            [
+                delete_vm(proxmox, node, vm_ids["master"]),
+                delete_vm(proxmox, node, vm_ids["worker1"]),
+                delete_vm(proxmox, node, vm_ids["worker2"]),
+            ],
+        )
 
     vm_id_master, upid_master = clone_template(
         proxmox,
@@ -130,35 +142,42 @@ def main() -> None:
     assert vm_id_worker2 == vm_ids["worker2"]
 
     logger.info("Waiting for cloning to complete...")
-    wait_status(
+    tasks_wait_completion(
+        proxmox,
         [
-            proxmox.nodes(node).tasks(upid_master).status.get,
-            proxmox.nodes(node).tasks(upid_worker1).status.get,
-            proxmox.nodes(node).tasks(upid_worker2).status.get,
-        ],  # pyright: ignore
-        "stopped",
-        sleep_time=40,
+            upid_master,
+            upid_worker1,
+            upid_worker2,
+        ],
     )
-
     logger.info("Cloning completed!")
 
     logger.info("Starting VMs...")
-    proxmox.nodes(node).qemu(vm_ids["master"]).status.start.post()
-    proxmox.nodes(node).qemu(vm_ids["worker1"]).status.start.post()
-    proxmox.nodes(node).qemu(vm_ids["worker2"]).status.start.post()
-    wait_status(
+    tasks_wait_completion(
+        proxmox,
         [
-            proxmox.nodes(node).qemu(vm_ids["master"]).status.current.get,
-            proxmox.nodes(node).qemu(vm_ids["worker1"]).status.current.get,
-            proxmox.nodes(node).qemu(vm_ids["worker2"]).status.current.get,
-        ],  # pyright: ignore
-        "running",
-        sleep_time=10,
+            proxmox.nodes(node).qemu(vm_ids["master"]).status.start.post(),
+            proxmox.nodes(node).qemu(vm_ids["worker1"]).status.start.post(),
+            proxmox.nodes(node).qemu(vm_ids["worker2"]).status.start.post(),
+        ],
     )
-
-    sleep(40)
     logger.info("VMs are running!")
 
+    logger.info("Waiting for QEMU guest agent to start")
+    done = False
+    while not done:
+        try:
+            _ = proxmox.nodes(node).qemu(vm_ids["master"]).agent.ping.post()
+            _ = proxmox.nodes(node).qemu(vm_ids["worker1"]).agent.ping.post()
+            _ = proxmox.nodes(node).qemu(vm_ids["worker2"]).agent.ping.post()
+        except ResourceException:
+            time.sleep(1)
+        except Exception as e:
+            raise RuntimeError("Failed to ping QEMU agent") from e
+        else:
+            done = True
+    logger.info("QEMU guest agent is running")
+
     vm_ip_master: str = get_vm_ip(proxmox, node, vm_ids["master"])
     vm_ip_worker1: str = get_vm_ip(proxmox, node, vm_ids["worker1"])
     vm_ip_worker2: str = get_vm_ip(proxmox, node, vm_ids["worker2"])