Best practices

One page per task, for the jobs that take more than one call: the requests in order, the in-guest work between them, and what goes wrong in the middle.

Facts — limits, field names, error codes — belong to the guide that owns the resource. A page here repeats one only where a step depends on it.

Every page exists twice, the markdown and an .html copy. Fetch the markdown and read it whole rather than through a summarizing fetch: a summary can drop a step.

• Recreate a VM — shrink a root disk, replace a machine, move off failed hardware
• Debug a VM that will not come up — what to read when there is no console
• Harden a VM — firewall and sshd changes that cannot lock you out
• Put a CDN in front — serve cacheable traffic near the reader, and cut egress charges
• Self-host storage — S3 or a shared filesystem on a VM you own
• Build a private mesh — private addressing between your own VMs
• Watch your credits and scale down — what to poll, what to cut or shrink when the spend is too high, and what happens at zero
• Watch metrics and scale up — an exporter on each VM, a metrics VM when you need history, and resizing before a resource runs out