`uv` and `yq` were `latest`, and helm was declared at 3.17.3 while
ci.yaml installs v3.10.0 — so `mise run chart:lint` and the pipeline were
validating two different helm versions, and nothing said so.
Each tool is pinned on an explicit backend and checked by running it
rather than by trusting version resolution: Python 3.12.13, uv 0.13.1,
helm v3.17.3+ge4da497, yq v4.54.1.
`aqua:` for helm, yq and uv, matching the other repos in this workspace.
The `github:` backend cannot name the yq binary `yq` at all on the mise
version this repo pins, and it mis-extracts helm 3.17.x — writing the
833-byte .asc PGP signature to
installs/github-helm-helm/3.17.x/helm, with attestation and SLSA
verification both logging as passing.
helm stays on 3.17.3, matching what mise.toml already declared for the
chart. That this file and ci.yaml disagree about the version is now the
only remaining gap, and it is in the workflow, not here.
The three mise tasks parse unchanged.
- Replace in-process iperf3 C binding with subprocess execution
- Add active I/O stall detection and connect-timeout
- Expose TCP retransmissions, mean RTT (ms), and exporter cycle timestamp
- Upgrade Dockerfile to Python 3.12 running as non-root user (nobody)
- Fix testTimeout/testDuration environment variable mapping in Helm chart
- Remove dead TrueCharts prometheus-operator chart dependency
- Add unit tests for exporter metric parsing and stall handling
- Switch tooling from devbox to mise
Co-authored-by: Malarbase <malarbase@gmail.com>
