Skip to content

Record v0.14, the release that spent nothing - #57

Merged
sethbergman merged 1 commit into
mainfrom
docs/v0.14-roadmap
Aug 30, 2026
Merged

Record v0.14, the release that spent nothing#57
sethbergman merged 1 commit into
mainfrom
docs/v0.14-roadmap

Conversation

@sethbergman

Copy link
Copy Markdown
Owner

Fifteen PRs since v0.13, with one theme: settle everything about a cloud apply that can be settled without performing one.

Five real defects

None of which the twenty-seven checks could see:

NLB client-IP preservation the node SG admitted the load balancer, but instance targets preserve the client address — so it matched health checks and nothing else. Healthy targets, no reachable cluster.
Missing kms:GenerateDataKey every hourly snapshot upload denied by KMS while every S3 permission looked right
leader_tls_servername verified the leader against a name no certificate carried, so no peer could ever have joined
ASG replacement loop ELB health before Ansible had issued certificates: a bare apply terminated and relaunched forever, billing throughout
Shared KMS key root volume and seal on one key, so destroy scheduled the key every archived snapshot depended on

Two guards

  • tests/preflight-static — the cross-layer agreements no single layer can see: Terraform against the cloud-init it renders, that against the Ansible layer that finishes the node
  • tests/cloud-apply-emulated — a real terraform apply, through the real AWS provider, against an implementation of the AWS API

The pattern worth noting

Three of the five were held in place by passing assertions — a test requiring the value that made the apply loop, one forbidding the rule the cluster needed to serve traffic, and one of mine comparing arns a mock had flattened. Written up in the posts now linked from the docs.

Once this is in, v0.14 can be tagged and the release cut.

🤖 Generated with Claude Code

Fifteen pull requests since v0.13, with one theme: settle everything
about a cloud apply that can be settled without performing one.

Five real defects, none of which the twenty-seven checks could see:

  - the node security group admitted the load balancer, but instance
    targets preserve the client address, so it matched health checks and
    nothing else -- healthy targets, no reachable cluster
  - snapshot uploads had no kms:GenerateDataKey, so every hourly upload
    would have been denied by KMS while every S3 permission looked right
  - retry_join verified the leader against a name no certificate carried,
    so no peer could ever have joined
  - the autoscaling group used ELB health before Ansible had issued
    certificates, so a bare apply terminated and relaunched forever
  - the root volume and the seal shared one KMS key, so destroy
    scheduled the key every archived snapshot depended on

And two guards so the class stops needing an apply to find:
tests/preflight-static checks the agreements no single layer can see, and
tests/cloud-apply-emulated runs a real terraform apply against an
implementation of the AWS API.

Three of the five were held in place by passing assertions. That is
written up in the posts now linked from the docs.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@sethbergman
sethbergman merged commit 5ae0d7e into main Aug 30, 2026
27 checks passed
@sethbergman
sethbergman deleted the docs/v0.14-roadmap branch August 30, 2026 01:39
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant