knife-proxmox-ve 0.1.3 → 0.1.5

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 5d5214d06c2b3094351f1e2824fc0d3e9cceba436f2053bb32001b41fe18ea5c
-  data.tar.gz: fad9ae85bb184f7dd8fdb6e54cac5a15a8f9036f1f5f4a07b939cd0a4a0e3f3a
+  metadata.gz: 918852ed3cc22ff0316c1eb9f3ae5e9da00254225926e37f0642ca5986085919
+  data.tar.gz: b096cd71927d635a5b3782a9b77cd558e8d782256120b6ab7b2ead348dfd1dea
 SHA512:
-  metadata.gz: c4d3a3bdff34585a82c46888572244512f06a7a17b94475b135f2d5f520c2b47d983e07f69cb40baa2ebc2fca8526db2cd5c6c7928ab1f778d2db28061167a05
-  data.tar.gz: 9008cdcd204db73879b3f2a4602e202d821dd70b9df217ebc8ff90e59342e7ff805c516497cd7c62970c531fd6f04df7e69f2fac84305836ba1f9faf821f7630
+  metadata.gz: 15d9f3e83831755684014fff78b5ed931e1951ebd99343406172f60bb307590626765505685ed52f57816b9a128f125d10bdd9c150f7b84b2bae84b827449981
+  data.tar.gz: ded4dcf52c6e1ae6f5dc58a850919fec0b7684cc62deda036c6b0f8e4deed321f7591ede5c4269dec282c1589068a7476edb25de1b2b17e138f698e9c6562891

data/README.md

@@ -151,6 +151,8 @@ Both commands share the provisioning flags below.
 | `--linked-clone` | Linked clone instead of the default full clone (`--storage` is then ignored). |
 | `--storage NAME` / `--pool NAME` | Target storage / resource pool. |
 | `--cores N` / `--sockets N` / `--memory MiB` | CPU / RAM. |
+| `--disk-size SIZE` | Grow the disk before the first boot: `+20G` adds to the template's size, `50G` sets it. |
+| `--disk DISK` | Disk to resize (`scsi0`, `virtio0`, …). Defaults to the cloned VM's boot disk. |
 | `--bridge vmbrN` / `--vlan TAG` | net0 bridge and VLAN tag (needs a VLAN-aware bridge). |
 | `--ip CIDR\|dhcp` / `--gateway IP` / `--prefix N` | Static IPv4 (`10.0.10.5/24`, or a bare IP with `--prefix`) or `dhcp`. |
 | `--nameserver IP` / `--searchdomain DOMAIN` | cloud-init DNS. |
@@ -169,6 +171,23 @@ freshly-created host), `--yes`, etc. They do not apply to `vm create`, which nev
 clone (faster, thin) — it requires a storage that supports it, and any `--storage` (CLI or
 cluster default) is then dropped, because Proxmox rejects a target storage on a linked clone.
 
+**Disk resize.** `--disk-size` grows the clone's disk **between the clone and the first start**,
+so cloud-init's `growpart`/`resizefs` claim the extra space on that first boot and the guest
+filesystem comes up at full size — no second run, no manual step in the guest. The size format is
+Proxmox's own: `+20G` adds 20 GiB to whatever the template had (the portable choice), a bare `50G`
+sets an absolute size. **Proxmox cannot shrink a disk** — an absolute size below the current one is
+rejected by the API.
+
+Without `--disk`, the disk is the cloned VM's boot disk: the first entry of its `boot` order that is
+a real disk (cloud-init drives and CD-ROMs are skipped), falling back to its lowest-sorting disk key.
+Pass `--disk scsi1` to grow a specific one — naming a disk the VM does not have fails before the VM
+is started, rather than silently doing nothing. Growing the disk requires `VM.Config.Disk` plus
+`Datastore.AllocateSpace` on the target storage.
+
+```bash
+knife proxmox vm create web-01 --template ubuntu-2404-template --disk-size +40G
+```
+
 **Bootstrap product (CINC by default).** A `vm bootstrap`'d VM is bootstrapped with **CINC** out of the
 box — the install is pinned to the CINC omnitruck regardless of whether you invoke a CINC- or
 Chef-branded `knife`, which also avoids Chef's commercial license gate. To bootstrap with Chef

data/lib/chef/knife/helpers/proxmox_vm_provision.rb

@@ -28,6 +28,13 @@ class Chef
       # ENV override for the cloud-init password so it never lands in shell history.
       ENV_CIPASSWORD = "KNIFE_PROXMOX_CIPASSWORD"
 
+      # A qemu disk key: bus name plus its index (scsi0, virtio1, sata0, ide0).
+      DISK_PATTERN = /\A(?:scsi|virtio|sata|ide)\d+\z/
+
+      # Proxmox's own size format for the resize endpoint: an optional "+" (grow by), a
+      # number, and an optional K/M/G/T unit (no unit means bytes).
+      DISK_SIZE_PATTERN = /\A\+?\d+(\.\d+)?[KMGT]?\z/
+
       def self.included(includer)
         includer.class_eval do
           # --- Source template / clone placement --------------------------------
@@ -72,6 +79,15 @@ class Chef
             long:        "--memory MiB",
             description: "Memory in MiB."
 
+          option :disk,
+            long:        "--disk DISK",
+            description: "Disk to resize (e.g. scsi0). Defaults to the cloned VM's boot disk."
+
+          option :disk_size,
+            long:        "--disk-size SIZE",
+            description: "Resize the disk before the first boot: '+20G' adds to the template's " \
+                         "size, '50G' sets it. Proxmox cannot shrink a disk."
+
           # --- Networking -------------------------------------------------------
 
           option :bridge,
@@ -147,6 +163,7 @@ class Chef
 
         validate_scalars!
         validate_network!
+        validate_disk!
         resolve_ssh_auth!(require_credential: require_ssh_auth)
         true
       end
@@ -222,6 +239,23 @@ class Chef
         ui.fatal!("--gateway #{gateway.inspect} is not a valid IPv4 address")
       end
 
+      # The size format is Proxmox's, so it is checked here rather than server-side: a typo
+      # like "20GB" would otherwise surface as an opaque HTTP 400 after the clone already ran.
+      def validate_disk!
+        disk = config[:disk]
+        size = config[:disk_size]
+
+        if !disk.nil? && !DISK_PATTERN.match?(disk.to_s)
+          ui.fatal!("--disk #{disk.inspect} is not a disk key (e.g. scsi0, virtio0, sata0, ide0)")
+        end
+        ui.fatal!("--disk needs --disk-size; naming a disk alone changes nothing") if !disk.nil? && size.nil?
+        return if size.nil?
+        return if DISK_SIZE_PATTERN.match?(size.to_s)
+
+        ui.fatal!("--disk-size #{size.inspect} is not a Proxmox size " \
+                  "(e.g. +20G to grow by 20 GiB, or 50G for an absolute size)")
+      end
+
       def positive_integer!(key, value)
         int = integer!(key, value)
         ui.fatal!("--#{key.to_s.tr("_", "-")} must be greater than 0 (got #{int})") unless int > 0
@@ -305,6 +339,10 @@ class Chef
           # clones only — for a linked clone the CLI/cluster storage is dropped entirely.
           storage:       full ? from_cli_or_cluster(:storage) : nil,
           pool:          config[:pool],
+          # Disk growth is applied to the clone before its first start (see Provisioner);
+          # nil disk_size means the template's disk is kept as-is.
+          disk:          config[:disk],
+          disk_size:     config[:disk_size],
           clone_timeout: config[:clone_timeout],
           boot_timeout:  config[:boot_timeout],
           # Post-start wait policy (see Provisioner#wait_for_ip). Defaults suit bootstrap;

data/lib/chef/knife/proxmox_vm_bootstrap.rb

@@ -28,13 +28,30 @@ class Chef
       CINC_PRODUCT = "cinc"
       CINC_INSTALL_URL = "https://omnitruck.cinc.sh/install.sh"
 
-      # A freshly cloned VM often still runs cloud-init (which itself drives apt) when SSH first
-      # answers. Installing the client then races cloud-init for the dpkg lock and fails
-      # non-deterministically. Block on cloud-init completion before the omnibus install so a
-      # single `vm bootstrap` is reliable. Guarded so non-cloud-init images don't error, and
-      # `|| true` so a degraded/errored cloud-init state still lets the bootstrap proceed.
-      CLOUD_INIT_WAIT_COMMAND =
-        "if command -v cloud-init >/dev/null 2>&1; then cloud-init status --wait >/dev/null 2>&1 || true; fi"
+      # A freshly cloned VM is still settling when SSH first answers, and two independent
+      # subsystems contend for the dpkg lock: cloud-init (which itself drives apt) and the
+      # apt-daily / unattended-upgrades systemd timers, which fire on their own schedule even
+      # after cloud-init reports done. Installing the client — or the first client run's package
+      # resources — then races them for the lock and fails non-deterministically. Clear the field
+      # before the omnibus install, in order:
+      #   1. wait for cloud-init to finish;
+      #   2. stop the apt-daily / unattended-upgrades timers and any in-flight run, so nothing
+      #      grabs the lock again for the rest of this boot (`stop`, not `disable`/`mask`: the
+      #      services return on the next reboot, leaving the image unchanged);
+      #   3. block (bounded) until the dpkg lock a stopped run may still hold is released.
+      # Every step is guarded on the tool existing and `|| true`, so non-cloud-init / non-systemd
+      # / non-apt images and degraded states still let the bootstrap proceed.
+      PREINSTALL_WAIT_COMMAND = <<~SH
+        if command -v cloud-init >/dev/null 2>&1; then cloud-init status --wait >/dev/null 2>&1 || true; fi
+        if command -v systemctl >/dev/null 2>&1; then
+          systemctl stop apt-daily.timer apt-daily-upgrade.timer apt-daily.service apt-daily-upgrade.service unattended-upgrades.service >/dev/null 2>&1 || true
+        fi
+        if command -v flock >/dev/null 2>&1; then
+          for lock in /var/lib/dpkg/lock-frontend /var/lib/dpkg/lock; do
+            [ -e "$lock" ] && flock -w 300 "$lock" true >/dev/null 2>&1 || true
+          done
+        fi
+      SH
 
       deps do
         require "chef/knife/bootstrap"
@@ -60,8 +77,8 @@ class Chef
         # TOFU: a freshly cloned VM has no entry in known_hosts. Accept its key on
         # first connect rather than failing the bootstrap or disabling verification.
         config[:ssh_verify_host_key] ||= :accept_new
-        # Wait for cloud-init to finish before the bootstrap installs the client (see constant).
-        config[:bootstrap_preinstall_command] ||= CLOUD_INIT_WAIT_COMMAND
+        # Drain cloud-init and the dpkg lock before the bootstrap installs the client (see constant).
+        config[:bootstrap_preinstall_command] ||= PREINSTALL_WAIT_COMMAND
       end
 
       # The bootstrap target host does not exist yet — it is resolved to the VM's IP

data/lib/knife-proxmox-ve/api.rb

@@ -76,6 +76,15 @@ module Knife
         @client.post("/nodes/#{enc(node)}/qemu/#{enc(vmid)}/config", params)
       end
 
+      # Grow a virtual disk. +size+ uses Proxmox's own format: "+<n>[KMGT]" adds to the
+      # current size, a bare "<n>[KMGT]" sets the new absolute size (Proxmox refuses to
+      # shrink). PVE 8 runs the resize in a worker and returns a UPID; earlier releases do
+      # the work inline and return nil — the caller decides whether there is a task to wait
+      # on, since sequencing is not this class's job.
+      def resize_disk(node:, vmid:, disk:, size:)
+        @client.put("/nodes/#{enc(node)}/qemu/#{enc(vmid)}/resize", { disk:, size: })
+      end
+
       # Start a VM. Returns the UPID of the async start task.
       def start(node:, vmid:)
         @client.post("/nodes/#{enc(node)}/qemu/#{enc(vmid)}/status/start")

data/lib/knife-proxmox-ve/client.rb

@@ -62,6 +62,12 @@ module Knife
         send_with_body(Net::HTTP::Post, path, body)
       end
 
+      # PUT +path+ with a form-encoded body. Proxmox reserves PUT for in-place mutations of
+      # an existing object (e.g. the qemu disk resize endpoint).
+      def put(path, body = {})
+        send_with_body(Net::HTTP::Put, path, body)
+      end
+
       # Redact secret-bearing values from arbitrary text (e.g. a Proxmox task log, which is
       # returned with HTTP 200 and so does not pass through the error-body scrub path).
       def scrub_text(str)

data/lib/knife-proxmox-ve/provisioner.rb

@@ -31,6 +31,18 @@ module Knife
       CLOUDINIT_DRIVE_KEY = /\A(?:ide|scsi|sata)\d+\z/
       CLOUDINIT_DRIVE_VALUE = /cloudinit/
 
+      # Config keys that can carry a resizable virtual disk.
+      DISK_KEY = /\A(?:scsi|virtio|sata|ide)\d+\z/
+
+      # Drives that are not resizable storage: the cloud-init drive and any CD-ROM.
+      NON_DISK_VALUE = /cloudinit|media=cdrom/
+
+      # Modern boot config: "boot: order=scsi0;ide2;net0".
+      BOOT_ORDER = /\border=([^,]+)/
+
+      # A Proxmox worker id, as returned by endpoints that offload their work to a task.
+      UPID_PREFIX = "UPID:"
+
       # Proxmox reports a VMID collision as a non-2xx body mentioning the existing id.
       DUPLICATE_ID = /already exists|config file already exists|VM \d+ already/i
 
@@ -49,8 +61,12 @@ module Knife
         node = spec[:target_node] || template[:node]
 
         params = build_config(spec.fetch(:config, {}) || {})
-        guard_cloud_init_drive!(node, newid, spec[:name]) if cloud_init?(params)
+        # One read serves both the cloud-init drive guard and disk discovery; a hardware-only
+        # provision needs neither and never reads the config back.
+        vm_config = read_vm_config(node, newid, params, spec)
+        guard_cloud_init_drive!(vm_config, spec[:name]) if cloud_init?(params)
         @api.update_config(node:, vmid: newid, **params) unless params.empty?
+        resize_disk!(node, newid, vm_config, spec) if spec[:disk_size]
 
         boot(node, newid)
         ip = wait_for_ip(spec, node, newid)
@@ -154,10 +170,16 @@ module Knife
           params.key?(:ipconfig0) || params.key?(:nameserver) || params.key?(:searchdomain)
       end
 
+      # Read the cloned VM's config only when something downstream actually needs it.
+      def read_vm_config(node, vmid, params, spec)
+        return nil unless cloud_init?(params) || spec[:disk_size]
+
+        @api.vm_config(node:, vmid:)
+      end
+
       # Fail fast before a doomed bootstrap: if the (template-derived) VM has no cloud-init
       # drive, the keys we are about to set would be silently ignored by the guest.
-      def guard_cloud_init_drive!(node, vmid, name)
-        config = @api.vm_config(node:, vmid:)
+      def guard_cloud_init_drive!(config, name)
         present = config.any? do |key, value|
           CLOUDINIT_DRIVE_KEY.match?(key.to_s) && CLOUDINIT_DRIVE_VALUE.match?(value.to_s)
         end
@@ -168,6 +190,61 @@ module Knife
           "cloud-init settings would be silently ignored"
       end
 
+      # Grow a disk of the fresh clone, BEFORE its first start: cloud-init's growpart/resizefs
+      # only run on boot, so a disk grown now ends up usable inside the guest, while a resize
+      # after boot leaves the extra space unclaimed until something in the guest is told to
+      # take it.
+      def resize_disk!(node, vmid, vm_config, spec)
+        disk = resolve_disk(vm_config, spec[:disk])
+        size = spec.fetch(:disk_size)
+        @ui&.info("Resizing #{disk} of VM #{vmid} to #{size}...")
+        upid = @api.resize_disk(node:, vmid:, disk:, size:)
+        # PVE 8 offloads the resize to a worker and hands back its UPID; older releases resize
+        # inline and return nothing, so there is nothing to wait on.
+        @api.wait_task(upid:) if task_upid?(upid)
+      end
+
+      # An explicitly named disk must exist on the VM — resizing a key Proxmox does not know
+      # is a silent no-op otherwise. Without one, fall back to the detected boot disk.
+      def resolve_disk(vm_config, disk)
+        return detect_disk(vm_config) if disk.nil?
+        return disk.to_s if vm_config.key?(disk.to_s)
+
+        raise Knife::Proxmox::Error,
+          "VM has no disk #{disk} (found: #{disk_keys(vm_config).join(", ")})"
+      end
+
+      # The disk to grow when the caller named none: the first entry of the VM's boot order
+      # that is a real disk, else the lowest-sorting disk key. cloud-init drives and CD-ROMs
+      # are never candidates.
+      def detect_disk(vm_config)
+        candidates = disk_keys(vm_config)
+        # Array#& keeps the receiver's order, so the boot order wins where it says anything.
+        disk = (boot_order(vm_config) & candidates).first || candidates.min
+        return disk if disk
+
+        raise Knife::Proxmox::Error,
+          "cannot tell which disk to resize (no disk found on the VM); pass --disk scsi0"
+      end
+
+      def disk_keys(vm_config)
+        vm_config.select do |key, value|
+          DISK_KEY.match?(key.to_s) && !NON_DISK_VALUE.match?(value.to_s)
+        end.keys.map(&:to_s)
+      end
+
+      def boot_order(vm_config)
+        match = BOOT_ORDER.match(vm_config["boot"].to_s)
+        return match[1].split(";") if match
+
+        # Pre-6.4 layout: "boot: cdn" plus a separate "bootdisk: scsi0".
+        [vm_config["bootdisk"]].compact.map(&:to_s)
+      end
+
+      def task_upid?(value)
+        value.is_a?(String) && value.start_with?(UPID_PREFIX)
+      end
+
       def boot(node, vmid)
         @ui&.info("Starting VM #{vmid}...")
         upid = @api.start(node:, vmid:)

data/lib/knife-proxmox-ve/version.rb

@@ -2,6 +2,6 @@
 
 module Knife
   module Proxmox
-    VERSION = "0.1.3"
+    VERSION = "0.1.5"
   end
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: knife-proxmox-ve
 version: !ruby/object:Gem::Version
-  version: 0.1.3
+  version: 0.1.5
 platform: ruby
 authors:
 - Piotr Wojcieszonek
@@ -77,7 +77,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
     - !ruby/object:Gem::Version
       version: '0'
 requirements: []
-rubygems_version: 4.0.10
+rubygems_version: 4.0.16
 specification_version: 4
 summary: Knife plugin to create and automatically bootstrap Proxmox VE VMs with Chef/CINC.
 test_files: []