Release v0.6.0

Per-VM Execution Queues: Each VM executes tasks serially through its own queue of 64 pending tasks, allowing different VMs to run concurrently. Apply, checkpoint, restore and initialization share the instance's serialized execution lifecycle.
Queue Admission Check: Messages are rejected with err_process_busy when the target VM has at least 48 pending tasks. This is an early admission check, not a capacity reservation; concurrent requests and the Node input backlog can still exceed the threshold.
Optional VM Checkpoints: Shutdown and single-VM stop skip VM checkpoints by default. Use hymx stop --checkpoint, CloseWithCheckpoint(), StopWithCheckpoint(pid) or POST /admin/vms/stopWithCheckpoint when a checkpoint is required.
Redis Outbox Queues: Outbox queues are persisted in Redis independently of VM checkpoints.

Drain accepted VM tasks and wait for local result/outbox persistence before closing the node's backing services.
Stop all VM instances before waiting in KillAll, coordinate queue submissions with shutdown, and remove instances even if their underlying Close returns an error.
Copy mutable checkpoint state to avoid sharing it with later VM execution.

Shorten global VMM lock usage so one VM's execution or checkpoint does not hold the lock needed by other VMs.
Remove unused Node registry wait state and expand lifecycle, queue admission, shutdown and isolated Redis integration coverage.

Drain pending messages before upgrading from v0.5.0. Old in-memory outbox queues and outbox data in legacy VM checkpoints are not automatically imported into Redis; legacy checkpoint outbox fields are ignored.
Restarting retains messages already written to Redis but does not automatically resend them. Call SDK TrySend(pid, target) / POST /trysend, or submit a new message to the same queue, to resume delivery. Redis durability depends on its persistence and eviction configuration.
Skipping VM checkpoints can increase recovery time and requires database history to remain available. Payment state persistence is unchanged.
hymx stop --checkpoint uses SIGUSR1 and must only target a daemon running this version. The stop command confirms signal delivery, not shutdown or checkpoint completion.
This release provides Linux and macOS binaries for amd64 and arm64. Windows binaries are not included.
Graceful shutdown drains accepted tasks and local outputs; it does not guarantee remote outbox delivery or forcibly interrupt a VM that never returns.

Updated README shutdown/recovery instructions and admin API documentation.
Added an isolated local network verification script for checkpoint and recovery behavior.

Merged changes: #71, #73.

Full Changelog: v0.5.0...v0.6.0

Navigation Menu