Workspace Lifecycle¶

This guide covers how Prism manages the lifecycle of your workspaces — from auto-shutdown with time limits, to DNS hostnames, to idle detection that saves costs when you're not working.

Time-to-Live (TTL)¶

A TTL is a hard time limit on a workspace. Once the TTL expires, the workspace is stopped automatically.

Setting a TTL at launch¶

prism workspace launch python-ml my-project --ttl 8h
prism workspace launch r-research paper-analysis --ttl 24h
prism workspace launch genomics long-run --ttl 48h

Duration format: h (hours), m (minutes), s (seconds). Examples: 8h, 30m, 1h30m.

What happens when the TTL expires¶

The spored daemon on the instance checks the TTL continuously.
5 minutes before expiration: wall broadcast to all logged-in users — "Workspace will stop in 5 minutes".
At expiration: spored stops (or terminates, depending on --on-complete) the workspace.
Safety valve: the Prism daemon also monitors ExpiresAt every 10 seconds and stops the workspace if spored fails to act.

Extending the time limit¶

In the GUI, click the Actions dropdown on the workspace row and select Extend Time (+4h). This adds 4 hours to the current expiry and updates the prism:ttl EC2 tag so spored picks up the change.

Via the daemon API:

curl -X POST http://localhost:8947/api/v1/instances/my-project/extend \
  -H "X-API-Key: $(cat ~/.prism/state.json | jq -r .Config.APIKey)" \
  -d '{"hours": 4}'

Viewing time remaining¶

The instance table in the GUI shows a Time Remaining column with color-coded status: - 🟢 Green — more than 2 hours - 🟡 Yellow — less than 2 hours - 🔴 Red — less than 1 hour or expired

DNS Hostnames (prismcloud.host)¶

Every Prism workspace automatically registers a DNS hostname at launch.

Hostname format¶

{name}.{account-base36}.prismcloud.host

For example, if your workspace is named my-project and your AWS account ID encodes to abc123:

my-project.abc123.prismcloud.host

The account base36 encoding is a compact representation of your AWS account ID — it scopes hostnames so different AWS accounts never collide.

Custom DNS names¶

Use --dns to specify a different record name:

prism workspace launch python-ml my-project --dns ml-ws
# → Registers: ml-ws.abc123.prismcloud.host

If --dns is not specified, the workspace name is sanitized (lowercased, non-alphanumeric characters replaced with hyphens) and used as the DNS name.

Using the hostname¶

The hostname appears in the Hostname column of the instance table and is used automatically in the SSH connect command:

ssh ubuntu@ml-ws.abc123.prismcloud.host

DNS registration happens automatically via the spored daemon when the workspace boots. The A record is deleted when the workspace stops.

Note: DNS propagation takes up to 60 seconds after workspace boot.

Idle Detection¶

Prism automatically detects when a workspace is idle and can hibernate or stop it to save costs. Idle detection runs on the instance via the spored daemon — it uses 7 independent signals that are more accurate than polling CloudWatch.

How idle detection works¶

The spored daemon checks the following signals:

Signal	Default threshold
CPU utilization	< 5% average
Network throughput	< 10 KB/min
Disk I/O	< 100 KB/min
GPU utilization (if present)	< 5%
Active terminal sessions (`/dev/pts`)	None active
Logged-in users (`who`)	None
Recent user activity (`last`)	No activity in threshold window

All signals must be below threshold simultaneously for the idle timeout to trigger.

Configuring idle detection¶

Idle detection is configured via your idle policy. Apply a policy when launching:

prism workspace launch python-ml my-project --idle-policy

Or override the threshold directly:

prism workspace launch python-ml my-project --idle-timeout 2h

This sets prism:idle-timeout on the EC2 instance tags, which spored reads at startup.

Built-in idle policies¶

Policy	Idle threshold	Action	Best for
`aggressive`	15 minutes	Hibernate	Dev/test
`balanced`	60 minutes	Hibernate	General use
`conservative`	4 hours	Stop	Long-running jobs
`research`	2 hours	Hibernate	ML/data science

Manage policies in the GUI via Settings → Idle Detection, or via:

prism admin idle-policy list
prism admin idle-policy apply <workspace> --policy balanced

Workspace States¶

State	Cost	Description
`running`	Compute + storage	Instance is running and accessible
`stopping`	Compute (brief)	Shutdown in progress
`stopped`	Storage only	Instance stopped; data preserved
`hibernated`	Storage only	RAM state saved to EBS; resumes in seconds
`pending`	Compute	Starting up
`terminated`	None	Permanently deleted

Cost comparison (example m5.xlarge, us-west-2)¶

State	Hourly cost
Running	~$0.192/hr
Stopped	~$0.008/hr (EBS only)
Hibernated	~$0.008/hr (EBS only)
Terminated	$0.00

Hibernation and stopping save ~96% on compute costs while preserving your work.

Spot Interruption Handling¶

If a workspace is launched with --spot, spored monitors the EC2 Instance Metadata Service (IMDS) every 5 seconds for a spot interruption notice. When a 2-minute warning is received:

spored broadcasts a wall warning to all users
Pre-stop hook is executed (if configured via prism:pre-stop tag)
DNS record is deleted
Instance terminates (or stops, per prism:on-complete tag)

The 2-minute window gives you time to save work if you're actively using the workspace.