Treat tart delete exit code 2 as successful cleanup so deleting an
already-removed VM does not prevent reconciliation from continuing.
Preserve other command failures and retain the wrapped process error
so callers can inspect its exit status.
Cover repeated deletion, other exit codes, process startup failures,
and wrapped exit-status preservation.
Pin CI actions to their previously used commit SHAs so jobs can start
under the repository action-pinning policy.
Validation: vmmanager package tests and lint for changed lines pass;
CI action pins match the last successful run.
Share one Stop completion per run and retain Stopping until both the stop command and VM cleanup finish. Expose transition conditions to reconciliation and return after requesting Stop so it cannot apply a generation or restart the VM early.
Cover both completion orders, repeated and concurrent calls, restart, and generation acknowledgement with bounded fake-command regressions. Allow process startup under parallel test load while preserving explicit shutdown ordering.
Read the cloned VM state before applying resource overrides or randomizing
its MAC address and serial. Suspended clones must retain the configuration
used to save their state.
Cover suspended and stopped clones through the command runner, including
assigned resource precedence, optional image FQN lookup, and metadata
command and JSON errors.
Keep at most five rotated worker log files for up to fourteen days and
compress them. Retain the existing 100 MB rotation threshold so long-running
workers do not accumulate an unlimited number of log backups.
Co-authored-by: kroo-oai <kroo@openai.com>
Expose noClipboard in VM creation, CLI inspection, and the API schema.
Pass --no-clipboard to Tart when requested. Clipboard sharing remains
enabled by default for existing VM specifications.
tart run supports --no-audio to disable host audio pass-through, but
there was no way to request it through a VM's spec - Headless, Nested,
and Suspendable all have a flag/field but audio didn't. Useful on
headless hosts with no audio hardware, where audio pass-through can
make guest CoreAudio depend on host audio services that don't exist.
* Apply backpressure to live exec output subscribers
* Use integer range loops in exec backpressure test
* Preserve reconnectable exec subscriber isolation
---------
Co-authored-by: Yibo Zhuang <yzhuang@openai.com>
* Migrate CI to GitHub Actions
* Publish Orchard through the OpenAI Homebrew tap
* Scope release secrets to GoReleaser steps
* Separate release and snapshot trust paths
* Use GitHub-hosted macOS 26 runner
* Run macOS unit tests on hosted runner
* Update Homebrew install command
* Build Docker image with GoReleaser
* Use GitHub-hosted Linux runner
* Use smaller Vetu image for Linux tests
* Run Linux tests on hosted ARM runner
* Use host architecture for Linux integration tests
* Preinstall Cloud Hypervisor for Linux tests
* Select Linux test images by architecture
* Satisfy platform image lint rules
* Format platform image errors
* Enable KVM access for Linux tests
* Honor VM wait time in WebSocket requests
* Satisfy WebSocket client lint rules
* Allow hosted Linux integration tests more time
* Support Vetu virtualization on Linux in addition to Tart on macOS
* api(portForward): ensure that rendezvousConn is closed
* Re-try SSH connections in integration tests
Because a VM might be still booting.
* controller(api): new "GET /vms/{name}/exec" WebSocket-based endpoint
* Split SSH connection and execution to avoid standard input handoff
* execstream: make Exit field a pointer again
To support serializing exit codes equal to 0.
* Implement server-side filtering for VMs by worker
* Parse more than one filter but error out when more than one is provided
* Fix off-by-one
* No need to use "\n" in Debugf()
* Load testing: synthetic VMs, multiple worker support and Grafana k6 test
* echoserver: prevent fallthrough when Accept() fails
* Move default local-dev context logic to CreateDevController()
* Synthetic: add a random delay to startup script echoing
* Add pagination support for listing VM events
Introduced a paginated event listing API, added support for pagination parameters in the request, and included cursor-based navigation using headers. Relevant tests and Badger store implementations were updated to support the new logic.
* Remove support for ordering VM events
Dropped `ListOrder` type, `order` query parameter, and related logic for ordering VM events. Updated tests, API schema, and Badger store to reflect the removal.
* Remove invalid VM events ordering test
Deleted a test case for invalid VM events ordering since the `order` query parameter and related functionality have been removed.
* Add support for ordering VM events
Implemented `order` query parameter for specifying sort order (ascending/descending) of VM events. Updated API schema, Badger store, and added related tests.
* Add support for limiting and ordering VM logs
Introduced `--limit` and `--order` flags for controlling the number of log lines and their sort order (ascending/descending). Updated API client to handle new options.
* Update internal/controller/store/badger/badger_events.go
Co-authored-by: Nikolay Edigaryev <edigaryev@gmail.com>
* fix: address PR review feedback
- switch logs CLI to --tail with desc ordering
- reuse ParseLogsOrder in controller with helpful errors
- always use ListEventsPage and scope event cursors
- move events pagination coverage to integration test
🤖 Generated with [Codex](https://chatgpt.com/codex)
Co-Authored-By: Codex <codex@openai.com>
* refactor: simplify prefix trimming and improve error formatting
- Replaced manual prefix check with `bytes.TrimPrefix` in Badger store.
- Enhanced error message formatting in VM logs controller.
* fix: address PR review feedback
- use suggested reverse seek in badger events pagination
- add events pagination client helper and use it in integration test
🤖 Generated with [Codex](https://chatgpt.com/codex)
Co-Authored-By: Codex <codex@openai.com>
---------
Co-authored-by: Nikolay Edigaryev <edigaryev@gmail.com>
Co-authored-by: Codex <codex@openai.com>
* Ability to set VM's power state and retrieve backing Tart VM's name
* Validate user-provided "powerState" field
* Introduce TestSpecUpdatePowerStateSuspend
* Introduce TestSpecUpdatePowerStateStopped
* OpenAPI specification: add note about suspended VMs to "tartName" desc.
* Sometimes we need to wait more than 30 seconds
edi-oai