* feat: support usb devices for vm; fixes#665
Signed-off-by: Daniel Muehlbachler-Pietrzykowski <daniel@muehlbachler.io>
* chore: fix linter errors
Signed-off-by: Pavel Boldyrev <627562+bpg@users.noreply.github.com>
---------
Signed-off-by: Daniel Muehlbachler-Pietrzykowski <daniel@muehlbachler.io>
Signed-off-by: Pavel Boldyrev <627562+bpg@users.noreply.github.com>
Co-authored-by: Pavel Boldyrev <627562+bpg@users.noreply.github.com>
Disks imported using `file_id` argument do not have FileVolume set,
that makes PathInDatastore return an empty string, which makes
IsOwnedBy return fales for any vm id.
The end result is the inability to resize imported disks,
this fixes it.
Signed-off-by: Oto Petřík <oto.petrik@gmail.com>
Co-authored-by: Pavel Boldyrev <627562+bpg@users.noreply.github.com>
* feat(vm): add 'path_in_datastore' disk argument
Provide access to actual in-datastore path to disk image,
and experimental support for attaching other VM's disks or host devices.
Signed-off-by: Oto Petřík <oto.petrik@gmail.com>
* chore: added to `/example` for acceptance testing
Signed-off-by: Pavel Boldyrev <627562+bpg@users.noreply.github.com>
---------
Signed-off-by: Oto Petřík <oto.petrik@gmail.com>
Signed-off-by: Pavel Boldyrev <627562+bpg@users.noreply.github.com>
Co-authored-by: Pavel Boldyrev <627562+bpg@users.noreply.github.com>
The CloudInit interface can be left empty in order to allow
autodetection of the drive being used. However, it would seem that this
value was causing problems (see #539).
This commit adds an additional validator for CloudInit interfaces which
allows the `""` value.
move `types` back from `internal` to `proxmox` and adjust a few other types, to make sure `proxmox` package is not dependent on anything else, and therefore can be extracted to a separate repo (#423)
* fix(vm): wait for VMs to actually stop when sending a shutdown command
Due to how a Proxmox cluster reacts to a VM shutdown command when
running in HA mode, the VM might still be running when the shutdown API
calls returns. This commit adds a loop that actively waits for the VM's
status to change to "stopped" (while also accounting for the shutdown
timeout) after the call's return.
* chore(refactoring): extracted VM state change wait loop into a separate function
* fix(vm): wait for VMs to actually start after requesting it from the cluster
This commit forces the plugin to wait for a VM to actually run after
requesting it to be started. This avoids problems with Proxmox's High
Availability mode, where a start request may not be immediately honoured
by the cluster.
* fix: linter errors
* fix: use `vmAPI.WaitForVMState`
---------
Co-authored-by: Pavel Boldyrev <627562+bpg@users.noreply.github.com>
The datastore update support introduced in #486 only worked if the
CloudInit interface was also changed at the same time. This commit
fixes the problem.
Co-authored-by: Pavel Boldyrev <627562+bpg@users.noreply.github.com>
* feat(vm): support for migration when the node name is modified
* Added a `migrate` VM flag which changes the provider's behaviour
when the VM's `node_name` is updated. If `true`, the VM will be
migrated to the specified node instead of being re-created.
* Added a `timeout_migrate` setting to control the timeout for VM
migration.
* Fixed a bug in the API's migration data structure that prevented
the online migration flag to be set.
* fix: update description
---------
Co-authored-by: Pavel Boldyrev <627562+bpg@users.noreply.github.com>
* feat(vm): pool update support
This commit removed the ForceNew flag from the VM resource's `pool_id`
argument and implements pool update:
* if the VM was part of a pool, it is removed from it,
* if the new `pool_id` value is non-empty, the VM is added to that new
pool.
* fix: use `types.CustomCommaSeparatedList` in `PoolUpdateRequestBody` datatype, minor error fix
---------
Co-authored-by: Pavel Boldyrev <627562+bpg@users.noreply.github.com>
fix: linter error in ambush
* This commit fixes a linter error that somehow doesn't manifest
unless some other, unrelated changes trigger it (see #501 and
#505).
* In addition it fixes a similar issue that had so far gone undetected
by the linter.
* Refactored the code in question into a function, since it was mostly
duplicated.
* Simplified a pair of conditionals that had the same code in both
branches.
* fix(vm): Don't add an extra efi_disk entry
* fix(vm): Don't add an extra efi_disk entry
---------
Co-authored-by: Pavel Boldyrev <627562+bpg@users.noreply.github.com>
* add import support for a lot of resources
* fix lints
* set user_id after importing
(cherry picked from commit c3d09ed00f6e1d7b0bb3ab01a1fc5c81510aa2e1)
* add tests, fix empty ID
* add import docs
* fix lint
* (fix) search for vm in cluster resources before calling node api
* (lint) search for vm in cluster resources before calling node api
* cleanup redundant error checks
---------
Co-authored-by: michalg91 <michal.gawrys@g2a.com>
Co-authored-by: Pavel Boldyrev <627562+bpg@users.noreply.github.com>
Token logins using root@pam!sometoken=uuid are not considered
by PVE as 'root' logins, and fail to change VM's arch.
Make sure the provider does not try to set/change VM's arch.
Co-authored-by: Pavel Boldyrev <627562+bpg@users.noreply.github.com>
* efi disks support
* make size a string
* add cpu numa support
* add docs
* fix description
* string size fixes
(cherry picked from commit 254724d1d10c5191d294f0377adac49ad41dc880)
* fix lints
* fix(vm): replace `size` with `type` for efi disk
also
- add support for `pre-enrolled-keys`
- add example
- fix handling of `numa` flag when it's not in the template
- cleanup linter errors
---------
Co-authored-by: Pavel Boldyrev <627562+bpg@users.noreply.github.com>
* fix(vm): Make vm_id computed, fix#364
Defaulting vm_id to -1 prevents resources depending on vm_id value.
Make vm_id computed, also update existing vm_id = -1 with correct vm_id.
* update examples to use auto-generated `vm_id`s
---------
Co-authored-by: Pavel Boldyrev <627562+bpg@users.noreply.github.com>
Fields `network_interface_names`, `ipv4_addresses` and `ipv6_addresses`
are sometimes be marked `computed` again, using `CustomizeDiff`.
Changes in `VMStarted` or `VMNetworkDevice` attributes invalidate
those values. Marking fields `computed` again avoids stale values.
`network_device` configuration block without `mac_address` results
in MAC address set to its default value (an empty string).
Terraform state expects `mac_address` to be an empty string, server
provides the actual (random) MAC address of the network device.
Terraform detects there are changes to be made, but because of
`DiffSuppressFunc` on `mac_address` they are hidden from the user,
but not from d.HasChange(mkResourceVirtualEnvironmentVMNetworkDevice)
By making `mac_address` also `computed`, the server-generated MAC
address is stored locally, avoiding spurious changes that trigger
re-computation of `network_interface_names`, `ipv4_addresses` and
`ipv6_addresses`.
* cleanup 1
* continue refactoring
* more refactoring
* move VMs under nodes
* move container and other apis under nodes
* cleanups
* enabled revive.exported linter & add comments to exported stuff
* enable godot linter
* enable wsl linter
* enable thelper linter
* enable govet linter
* cleanup after rebase
* cleanup after rebase
* extract SSH ops into a separate interface
* fix linter error
* move ssh code to its own package
* cleaning up VirtualEnvironmentClient receivers
* on the finish line
* not sure what else I forgot... 🤔
* fix ssh connection and upload
* renaming client interfaces
* final cleanups
