microvm-ctl playground
Dry run is on. Every mutating action (bootstrap, build, run, lease, suspend, resume, terminate, scale, drain, probe, call) records the exact request it would send in the API trace and sends nothing. Reads still hit AWS.

Overview

What this account can do right now: the applied quotas, the fleet by state, and the last few things the playground did.

Applied quotas

quotapublishedappliedplane throttles at

Fleet

Counts every microVM in the region. RUNNING and SUSPENDED both hold memory quota.

Configuration

Recent API calls open trace

opstatusms

Recent jobs

Images

An image is a snapshot: your Dockerfile built on the managed AL2023 base, warmed until /ready says 200, then frozen. Every RunMicrovm clones it.

Build an image

Expect ~2 minutes. Progress streams below; build logs land in CloudWatch.

Images in

namestateactive versioncreated

Fleet

Every microVM in the region, live. Launch, converge a fleet to N, park it, drain it. Every mutating call goes through the token bucket at 80% of the applied quota.

microVMs  

microvm idstateimageverageendpoint

Run microVMs

Scale a fleet

Scale-down terminates SUSPENDED members first (they still hold memory quota), then the youngest RUNNING. The job log shows the victims before anything is sent.

Lease a VM one task RunMicrovm with a lease in runHookPayload; the VM heartbeats and completes it itself

clientToken = sha256(kind:token), so a replayed launch returns the same VM. The policy becomes the VM's idle policy (max idle = budget, no auto-resume) and its duration cap (budget + slack). The job log shows the encoded payload and the exact RunMicrovm request before anything is sent.

Call a VM

An authenticated request through the execution plane: a port-scoped JWE token in X-aws-proxy-auth, 429 backoff, a patient 502 retry while a suspended VM auto-resumes, and a one-time re-mint on 403.

Request

Quick paths: GET /healthz · POST /execute · GET /state

Response

Nothing sent yet.

History

timerequeststatusms

Logs

CloudWatch tail of /aws/lambda/microvms/<image>. Build output and runtime stdout land here, one stream per VM.

Probe

A parameterised version of the benchmark harness. It launches real VMs, times every stage, checks that state survives suspend and resume, and cleans up after itself. It costs a few cents.

Plan

Launches run one at a time and each extra VM is terminated as soon as it is measured, so the probe fits an 8 GB memory quota with a 2 GB image.

Results

No probe run yet.

Cost

The CostModel arithmetic, computed server-side from the published us-east-1 rates, plus what the current fleet costs per hour if nothing changes.

Price a session shape

Fleet cost per hour, as it stands

API trace

Every AWS API call this process makes, captured from botocore's event hooks: operation, HTTP status, duration, and the redacted request and response. Dry-run entries are the requests that were not sent.

timeserviceoperationstatusmsdetail
No calls yet.

Settings

The PlaneConfig for this session. Values start from the MVM_* environment variables; changes here apply to this process only.

Credentials held in this process only; never written to disk

SSO profiles need aws sso login --profile NAME in your terminal first.

Plane configuration

Account:
Base image:

Bootstrap the account

Creates, idempotently, the artifact bucket (public access blocked) and two IAM roles: a build role that can read the bucket and write build logs, and an execution role for the running VM. Safe to re-run.