noisypigeon.garden

scaleway/compute-instance v8.0.0

By noisypigeon.garden

compute-instance v8.0.0 lets a workload's scaleway.instance leaf (LIB-0043 §7) stand on its own. It places the server in a project, creates the server's public IPv4 when asked, and attaches it to a Private Network, so no caller-side resources are needed.

What's new

  • project_id is required. The server, and any IP the module creates, live in it. Before this, the instance landed in the provider's default project, which for a workload leaf isn't the workload's. This is the breaking change: every caller has to pass project_id when it moves to v8.
  • public_ipv4 (default false) creates a routed_ipv4 scaleway_instance_ip in project_id and attaches it alongside any ip_ids. It reverses DEV's ADR-0153 for the IP. One module can create an IP and the server that references it, and the IP goes with the server, as self-delete already has it. The IP isn't gated by enabled, so an enabled = false cycle keeps its address. While a Public Gateway pushes its default route on the instance's network, the IP works outbound only (ADR-0146).
  • enable_private_network and private_network_id (default false and null) attach the server through a standalone scaleway_instance_private_nic, so attaching or detaching never replaces the server (ADR-0145). The toggle is an explicit bool for v5.6.0's reason: count can't depend on an apply-time-unknown ID. Setting enable_private_network without an ID fails validation. In a workload, the NIC has to land in the same apply as the server, or a gateway-only instance's cloud-init would run with no route out.
  • private_ipv4 outputs the server's IPv4 on its Private Network, with the dual-stack IPv6 address filtered out, for a PAT rule to target.

Fixed

cloud_init = null, the default, no longer fails the plan. The module hashed coalesce(var.cloud_init, ""), and coalesce refuses when every argument is null or empty. The hash is unchanged for every value, so no server is replaced by the fix.

Upgrading

Add project_id. A caller that creates its own scaleway_instance_ip and NIC, as compute-cluster does, can keep doing so through ip_ids, and leave the new inputs at their defaults. compute-cluster pins v7.0.0 and doesn't move.

Verified with tofu validate, tofu fmt -check, and tofu test with a mocked provider: one IP and one NIC with both on, none with both off, and enable_private_network without an ID refused. No plan was run against a live project.

🤖 Generated with Claude Code

PR: https://github.com/noisypigeon/library/pull/221