Lifecycle and console

The power actions are on both the Instances list and the machine’s own page. The page adds the ones that need room.

The actions

Action	What it does
Boot	Start a created or stopped machine
Shutdown	Ask the guest to shut down cleanly
Reboot	Restart the guest
Pause / Resume	Freeze the machine in memory and thaw it again. The console offers it for classic VMs only
Force stop	Stop the machine without asking the guest. Nothing gets a chance to flush
Reinstall OS	Rebuild the machine’s disk from an image, keeping its identity
Create image	Turn this machine’s root disk into a reusable image. Powered off only
Export transfer package	Download the machine and every disk as one file. Powered off only
Delete	Remove the machine and the disks the host owns for it

Shutdown sends the guest the same signal as pressing its power button and then waits. A guest that does not bring itself down within that grace period is stopped by the host instead, so the machine does not sit half off. Reaching the guest at all needs the machine to answer, which is the one thing an unresponsive machine is not doing: use Force stop for that.

Pause, Force stop, Create image, Export transfer package and Delete live under More on the machine’s page, and Reinstall is on its Hardware tab. Create image and export both need the machine completely stopped, because neither can copy a disk the host still has open. See Images and templates and Move a machine to another host.

Reinstall and delete destroy the disk

Reinstall replaces the system disk with a fresh copy of the image, and every snapshot of the machine goes with it. Delete removes the machine and every disk the host made for it.

Whether a data volume survives depends on how it was made, and the difference is easy to miss. A volume you created on the machine’s create form is private to that machine and is deleted with it. A volume you created on the Volumes page and then attached is a separate object: detaching or deleting the machine removes the attachment and leaves the volume and its data alone.

States you will see

State	Meaning
Created	Built but never booted
Booting	Started, not yet seen on the network
Running	Started and observed
Paused	Frozen. Its processors are stopped and its memory is held where it was
Powered off	Not running, with no host processes left for it. The console says “powered off” rather than “stopped” so that a clean shutdown reads differently from Paused, where the machine is frozen but its memory is still held
Failed	Something went wrong; the record is kept so you can read why

Three more labels appear while a machine is being made or could not be made: Creating, Downloading, and Provision failed. The last one is distinct from Failed: it means the machine was never built, rather than that it was built and then went wrong.

Booting ends when the host sees this start of the machine: it takes an address, its guest agent answers, or a probe reaches it. A machine built with cloud-init that is never seen is shown as Failed after two minutes with a boot-timeout reason. That is a reading of the situation on screen, not a teardown: the host has left the machine alone, and it may still be booting slowly. Open its console and look before you decide.

A failed machine is deliberately left in place. Open it and read the failure reason before deleting anything. The reason also says which way out applies. If the host proved it had taken the machine’s runtime down cleanly when it failed, boot it again; its disk is untouched. Otherwise use Reset to stopped, which clears anything the host still holds for the machine and returns it to Powered off so you can boot it. Its disk is untouched there too. Reinstall, which erases the disk, is a separate decision you make deliberately.

When a machine counts as dead

The host keeps watching every running machine by asking it whether it is still there, and a machine has to miss several answers in a row before anything is concluded. What is concluded then depends on what the host can actually see.

What the host sees	What it concludes
The machine’s process has exited, was killed by a signal, or has closed the channel it holds open while running	The machine is gone. It is marked Failed and the reason is recorded
The machine stopped answering while its own process is still alive	It is hanging, not dead. It keeps running
The machine keeps restarting without ever coming up	The host stops it and marks it Failed as a boot loop

The middle row is the one that matters when a host is under pressure. A machine that stops answering but is still alive is left running on purpose, and the host records a vmm_unresponsive entry against it, which you read under Recent failures on the Logs and diagnostics page. Treating a missed answer as death is not a safe guess: a host that is briefly overwhelmed stops answering for every machine on it at once, while the guests themselves are healthy. Destroying them cannot be undone and loses the memory they were holding, so the host keeps the machine and tells you instead. Deciding what to do with a machine that is not answering stays your call.

Two cases skip that check

A stop you asked for is the stop you asked for. A machine caught in a boot loop is stopped on purpose, because it is burning host resources; that is a handling decision, not a conclusion that its process died.

A boot loop is judged in two ways:

• A machine that restarts without ever having been seen up on this start is stopped after one such restart inside five minutes. That is the shape of a boot that never gets going.
• A machine that was seen up and only then starts restarting is stopped at five restarts inside five minutes. Restarting once or twice is normal work, after a kernel update for example, and is not a loop.
• A machine the host has no way to observe at all is judged on the count alone. Silence is not evidence that it never came up.

If a machine is running but not answering, work outside in:

1. Look at the host first. Several machines going unresponsive in the same window points at the host rather than at the machines. See Logs and diagnostics.
2. Open the machine’s console and read what the guest is doing.
3. Then decide. Force stop under More ends the machine without needing it to cooperate.

Console

The console gives you the machine’s serial output and a login prompt, which is the right tool when the network is exactly what is broken. Use it to watch a boot, read cloud-init output, or log in when SSH will not connect.

Memory reclaim

Machines are created with memory ballooning enabled. The host automatically reclaims pages the guest has freed, so host memory reflects what guests are actually using rather than the highest they ever reached, and a guest under memory pressure gets its memory back automatically.

You can also reclaim memory from a machine by hand on the VM detail page. Virtainer Free does not oversubscribe memory on its own: reclaim is an action you take, not a policy that runs behind your back.

A machine that has been given a passed-through device is the exception: it gets no balloon device at all. The hypervisor cannot coordinate a balloon with memory pinned for direct device access, so the two are mutually exclusive and the host refuses the combination rather than choosing one for you. That means a passthrough machine’s memory is fixed at its configured size, and manual reclaim does not apply to it. See PCI passthrough.

Metrics

Each machine keeps its own history of CPU, memory, disk, and network use, over the last hour or the last day. It is stored on the host, so it survives a restart of Virtainer Free and needs no external monitoring stack to be useful.