kitchen-proxmox 0.3.0 → 0.3.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 8b86aaf1627109dc3fc11a425f023a03da6601eb6ef5c19e7d426654f40a1363
-  data.tar.gz: 71a7a993ce7729691b6887f84df0854023b32953b843fd7b0c1f29a3d07c253f
+  metadata.gz: 8592a8c21a5245c5517893fb3de6c2c2a9cae921cdb8c499fd3a7bcfcc8dc35a
+  data.tar.gz: eb18680f687a41ef9a5a5c2f4b290e586e867de4795cdc5733998189fb3b4e85
 SHA512:
-  metadata.gz: 5c7640ad5482862220e38d3fb3ff3cdc3d6943b4656e5bdbb9c49169f5887c82c23044abcc6361e870625fda7fa944d7f81944e70d10255b095e230fbdd777b6
-  data.tar.gz: 9f919e44967c325720f05724b97318884719fd21b46717838d38199beebbb062c1a829c1a635092461d56fe9258c2f917e2ed97a5f18cf4af44d38b19b03c858
+  metadata.gz: 8da473dbe9ddf5891bff39a77fe3a55db3fd263a4413592e0c55d826387dbde2e60f77287516f31dd5083200d8d5c962ceaf811f43ad0b89e1a794311ece678b
+  data.tar.gz: c93d6729fa61fbc286cdd6b64e03febad2ea3995c1a29e346e5c35037f51339f84aef79370c90443ce96f4ec94fb494225fbf05a3ad27942378c97b54e2d9e36

data/lib/kitchen/driver/proxmox/api_client.rb

@@ -5,6 +5,7 @@ require 'json'
 require 'uri'
 require 'openssl'
 require 'kitchen'
+require_relative 'errors'
 
 module Kitchen
   module Driver
@@ -26,6 +27,10 @@ module Kitchen
           get('/api2/json/cluster/nextid')
         end
 
+        def validate_vmid(vm_id)
+          get("/api2/json/cluster/nextid?vmid=#{vm_id}")
+        end
+
         def clone_vm(node:, template_id:, new_id:, **options)
           full = options.fetch(:full, true)
           body = { newid: new_id, full: full ? 1 : 0, target: node }
@@ -75,7 +80,14 @@ module Kitchen
           deadline = Time.now + timeout
           loop do
             status = task_status(node:, upid:)
-            return status if status['status'] == 'stopped'
+            if status['status'] == 'stopped'
+              exitstatus = status['exitstatus'].to_s
+              unless exitstatus == 'OK'
+                raise ProxmoxErrors::ApiError.new(500, exitstatus)
+              end
+
+              return status
+            end
             raise "Task timeout after #{timeout}s: #{upid}" if Time.now > deadline
 
             sleep interval
@@ -145,7 +157,7 @@ module Kitchen
         end
 
         def handle_response(response)
-          raise "Proxmox API error #{response.code}: #{response.body}" unless response.is_a?(Net::HTTPSuccess)
+          raise ProxmoxErrors::ApiError.new(response.code, response.body) unless response.is_a?(Net::HTTPSuccess)
 
           JSON.parse(response.body)['data']
         end

data/lib/kitchen/driver/proxmox/errors.rb

@@ -0,0 +1,39 @@
+# frozen_string_literal: true
+
+module Kitchen
+  module Driver
+    module ProxmoxErrors
+      # Structured error raised by ApiClient on non-2xx responses.
+      class ApiError < ::StandardError
+        attr_reader :status_code, :response_body
+
+        def initialize(status_code, response_body)
+          @status_code = status_code.to_i
+          @response_body = response_body
+          super("Proxmox API error #{status_code}: #{response_body}")
+        end
+
+        # Returns true when the error indicates a VMID is already in use
+        # or cannot be locked (concurrent clone race).
+        def vmid_conflict?
+          return false unless status_code == 400 || status_code == 500
+
+          response_body.match?(/already exists/i) ||
+            response_body.match?(/unable to create VM \d+/i) ||
+            response_body.match?(/can't lock file.*lock-\d+/i)
+        end
+
+        # Returns true when the error indicates another process owns the VM
+        # (lost a VMID race — VM was started or configured by another clone).
+        def vmid_race_lost?
+          return false unless status_code == 400 || status_code == 500
+
+          response_body.match?(/already running/i) ||
+            response_body.match?(/hotplug problem/i) ||
+            response_body.match?(/does not exist/i) ||
+            response_body.match?(/can't lock file.*lock-\d+/i)
+        end
+      end
+    end
+  end
+end

data/lib/kitchen/driver/proxmox.rb

@@ -2,6 +2,7 @@
 
 require 'kitchen'
 require_relative 'proxmox_version'
+require_relative 'proxmox/errors'
 require_relative 'proxmox/api_client'
 
 module Kitchen
@@ -16,6 +17,8 @@ module Kitchen
 
       plugin_version Kitchen::Driver::PROXMOX_VERSION
 
+      ApiError = Kitchen::Driver::ProxmoxErrors::ApiError
+
       required_config :proxmox_url
       required_config :proxmox_token_id
       required_config :proxmox_token_secret
@@ -32,6 +35,9 @@ module Kitchen
       default_config :clone_timeout, 300
       default_config :start_timeout, 300
       default_config :ip_wait_timeout, 120
+      default_config :clone_retries, 5
+      default_config :vmid_range_min, 900_000
+      default_config :vmid_range_max, 999_999
 
       def create(state)
         return if state[:vm_id]
@@ -63,23 +69,66 @@ module Kitchen
       end
 
       def clone_and_start(state)
-        info("Creating Proxmox VM from template #{config[:template_id]}...")
+        retries = config[:clone_retries]
+        last_error = nil
+
+        retries.times do |attempt|
+          info("Creating Proxmox VM from template #{config[:template_id]}...")
+
+          vm_id, vm_name = allocate_and_clone
+          state[:vm_id] = vm_id
+          state[:vm_name] = vm_name
+
+          begin
+            configure_hardware(vm_id)
+            start_and_wait_for_ip(state, vm_id)
+            info("Proxmox VM #{vm_name} (#{vm_id}) created.")
+            return
+          rescue ApiError => e
+            raise unless e.vmid_race_lost?
+
+            last_error = e
+            warn("VMID #{vm_id} race lost (attempt #{attempt + 1}/#{retries}): #{e.message}")
+            warn("Another process owns VM #{vm_id} — abandoning and retrying...")
+            clear_state(state)
+            sleep backoff_delay(attempt)
+          end
+        end
+
+        raise last_error
+      end
 
-        vm_id = allocate_vm_id
-        vm_name = generate_vm_name(instance.name)
+      def allocate_and_clone
+        retries = config[:clone_retries]
+        last_error = nil
 
-        clone_template(vm_id, vm_name)
-        configure_hardware(vm_id)
-        start_and_wait_for_ip(state, vm_id)
+        retries.times do |attempt|
+          vm_id = allocate_vm_id
+          vm_name = generate_vm_name(instance.name)
 
-        state[:vm_id] = vm_id
-        state[:vm_name] = vm_name
+          begin
+            clone_template(vm_id, vm_name)
+            return [vm_id, vm_name]
+          rescue ApiError => e
+            raise unless e.vmid_conflict?
+
+            last_error = e
+            warn("VMID #{vm_id} conflict (attempt #{attempt + 1}/#{retries}), retrying...")
+            sleep backoff_delay(attempt)
+          end
+        end
+
+        raise last_error
+      end
 
-        info("Proxmox VM #{vm_name} (#{vm_id}) created.")
+      def backoff_delay(attempt)
+        (0.5 * (2**attempt)) + rand(0.0..1.0)
       end
 
       def allocate_vm_id
-        Integer(api_client.next_vm_id)
+        vm_id = rand(config[:vmid_range_min]..config[:vmid_range_max])
+        api_client.validate_vmid(vm_id)
+        vm_id
       end
 
       def generate_vm_name(suite_name)

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: kitchen-proxmox
 version: !ruby/object:Gem::Version
-  version: 0.3.0
+  version: 0.3.2
 platform: ruby
 authors:
 - Richard Nixon
 autorequire:
 bindir: bin
 cert_chain: []
-date: 2026-04-30 00:00:00.000000000 Z
+date: 2026-05-04 00:00:00.000000000 Z
 dependencies: []
 description: A Test Kitchen driver for Proxmox VE. Manages VM lifecycle (create, destroy)
   via the Proxmox REST API. Supports cloning from templates.
@@ -22,6 +22,7 @@ files:
 - README.md
 - lib/kitchen/driver/proxmox.rb
 - lib/kitchen/driver/proxmox/api_client.rb
+- lib/kitchen/driver/proxmox/errors.rb
 - lib/kitchen/driver/proxmox_version.rb
 homepage: https://github.com/trickyearlobe-chef/kitchen-proxmox
 licenses: