Repository navigation
v0.1.33
⚠️ Isolation suites were not run for this tag
The Cilium packet-level suite and the Kata boot probe did not execute, so
tenant isolation is unverified and the control and guest image digests
were not verified by them. This release continues only because a
workflow_dispatch run set isolation_skip=true (acknowledgement none):
a tag push cannot take this path.
Set SANDBOX_ISOLATION_RUNNER and re-run the release without the skip
input. See docs/release-process.md#isolation-gate.
What's Changed
- docs: record the 0.1.32 isolation run and point the site at it by @ejackson-vp in #242
- config: fail closed on the insecure-dev flag, and stop retyping bounds by @ejackson-vp in #244
- refactor(protocol): declare every /internal answer once, in shared by @ejackson-vp in #246
- deploy: stop shipping a config the controller refuses to start by @ejackson-vp in #245
- test(contract): the contract suite needs no guest image by @ejackson-vp in #250
- fix(store): fence only the operation a session's gate owns by @ejackson-vp in #247
- docs(adr): record the capability key ring and supersede 0001 by @ejackson-vp in #248
- docs(runbook): what to do when a refund is stuck by @ejackson-vp in #249
- fix: jitter every retry loop, and read the Retry-After we already send by @ejackson-vp in #251
- refactor: make a swapped relay/proxy endpoint fail to compile by @ejackson-vp in #253
- lint: ratchet function length, not just file length by @ejackson-vp in #252
- refactor(controller): one module per payment route, and the tests that found three leaks by @ejackson-vp in #255
- fix(capacity): an allocated reservation names a session, or it has a deadline by @ejackson-vp in #256
- release: name the guest version, because it can lag by @ejackson-vp in #257
- release: pin 0.1.33 by @ejackson-vp in #258
Full Changelog: v0.1.32...v0.1.33
Images
Multi-arch (linux/amd64, linux/arm64), published to GHCR:
ghcr.io/ejackson-vp/sandbox-control:v0.1.33ghcr.io/ejackson-vp/sandbox-guest:v0.1.33
These tags were created only after both images passed the Trivy
gate. Each image carries an SBOM and SLSA build provenance as OCI
attestations:
docker buildx imagetools inspect ghcr.io/ejackson-vp/sandbox-control:v0.1.33Verifying provenance
The *.sigstore.json assets are keyless Sigstore bundles produced by
this workflow, covering exactly the digests above. Verify the image
itself:
gh attestation verify \
oci://ghcr.io/ejackson-vp/sandbox-control:v0.1.33 \
--repo ejackson-vp/sandboxAdd --bundle-from-oci to check the attestation stored in the
registry instead of the one on GitHub.
*.intoto.jsonl is the in-toto statement carried inside the bundle,
for SLSA tooling that consumes statements rather than bundles.
The guest image is what runs inside the microVM and must match the
cluster architecture. CVE results for this tag are in the Security
tab (trivy-control, trivy-guest).
Staging images from the build (staging-<run-id>-<sha>) are not
release artifacts; see docs/release-process.md for cleanup.
Isolation evidence for the promoted pair
The skip above is about CI: the preflight needs a KVM runner this cluster does not have. The suites were run afterwards against the digests this release promoted, on the real cluster:
| Control | 0.1.33 sha256:4804da2e… |
| Guest | 0.1.33 sha256:933a53e0… |
| Result | 23 passed, 1 skipped (the X-Forwarded-For case; this deployment sets TRUST_PROXY_HOPS=1) |
Recorded in docs/isolation-evidence.md, and the gate row in security-review/RELEASE_GATE.md names these digests.
What the tag alone deploys
The guest image stamps BUILD_COMMIT into a label, so the digest of a release cannot exist until after the commit that pins it. This tag therefore pins guest 0.1.32, and the commit that follows promotion moves the overlay to guest 0.1.33.
Deploying by tag alone gives you control 0.1.33 with the previous guest. Deploy the commit after it, or pin the guest digest explicitly, if you want the reconnect jitter this release added to the guest.