Compose, Quadlet and Kubernetes renderings of the robot topology - #463
Merged
Minipada merged 5 commits intoSep 1, 2026
Merged
Conversation
…obot topology Adds deploy/robot/, describing the three-container robot tier (dc-ros, vector, dc-uploader) three ways so a site deploys it with whatever it already runs: compose.yaml, quadlet/*.container+*.volume+*.network, and kubernetes/robot-pod.yaml — a single Pod, which is what lets `podman kube play` run it directly with no cluster. kubeconform (tools/ci/pre-commit/kubeconform.sh) validates the Kubernetes manifest, wired into pre-commit. deploy/robot/scripts/verify_kube_play.sh proves the Pod reaches ready the same way tools/release/scripts/verify_published_images.sh already does (the "dc_bridge reports ready" log line). deploy/robot/scripts/verify_network_isolation.sh proves the routing claim with plain Podman networks: no internet route for the robot tier, and only outbound robot-to-edge connectivity — a routing- level proof, not the CNI-enforced one #452 does with kind + NetworkPolicy. Both run as a new CI job (verify-robot-manifests) against the same published images verify-published-images already exercises. Unlike tools/e2e/compose.split.yaml (the *test* rendering, with Postgres/ RustFS standing in for reachable Destinations), these renderings are the deployable robot tier for the fleet shape (epic #440 scenario 3): Records go out through the blessed `vector` Destination type to an edge aggregator, and Files go to edge-local object storage — both placeholder hostnames, since that edge infrastructure is outside this repository. No new build or publish path: all three renderings pull the same ghcr.io/minipada/ros2_data_collection/dc-ros and dc-uploader images .github/workflows/ci.yaml already builds and pushes. Closes #450 Signed-off-by: David Bensoussan <d.bensoussan@proton.me>
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## jazzy #463 +/- ##
==========================================
- Coverage 71.46% 71.44% -0.01%
==========================================
Files 124 124
Lines 7563 7563
==========================================
- Hits 5404 5403 -1
- Misses 2159 2160 +1
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
Adds deploy/robot/docs/{single-robot-quick-setup,containers-single-machine,
fleet-topology}.html: standalone, interactive comparisons of scenario 1
(all-in-one container), the Compose/Quadlet rendering this directory adds
(#450), and the target fleet shape (epic #440 scenario 3). Each calls out
the weaknesses of a direct-to-destination shape versus the fleet tier,
grounded in "A Reference Architecture for Robot Fleets" (security exposure,
no edge buffer, sites that won't allow it on their network).
REUSE.toml annotates them instead of an inline header — a leading comment
before <!DOCTYPE html> risks quirks mode in some engines.
Signed-off-by: David Bensoussan <d.bensoussan@proton.me>
build-dc-ros-image/build-dc-uploader-image now build and push a :<sha> ref on every run, PR included — the same CI-internal-reuse pattern already used for dc-workspace/dc-e2e — instead of being gated to `push` on jazzy entirely. That lets verify-published-images/verify-robot-manifests pull and run the actual image a PR just built, catching a broken image or a broken robot-manifest rendering before merge instead of only after. Refs are computed by .github/actions/image-ref, ported from ~/dev/monorepo's action of the same name: it also derives a sluggified branch ref. Both that branch ref and a fixed :latest tag (the ref a fleet robot would actually pin to, CLAUDE.md's convention) are pushed only on `push` to jazzy — mirroring monorepo's own podman-push, which withholds branch_ref on pull_request the same way. A not-yet-merged PR must never move either floating ref. verify_kube_play.sh now accepts DC_ROS_IMAGE/DC_UPLOADER_IMAGE overrides and substitutes them into a temp copy of kubernetes/robot-pod.yaml before `podman kube play` — the committed manifest hardcodes :latest, which would otherwise make verify-robot-manifests test a stale image on every PR instead of the one that PR just built. Signed-off-by: David Bensoussan <d.bensoussan@proton.me>
The Format job failed on push: codespell's -w autofix rewrote "logicalY" to "logically" inside deploy/robot/docs/*.html's bundled JS (a pan/zoom helper's identifier, not a real misspelling), corrupting a delivered, checksummed Archify artifact that should never be hand- or tool-edited after delivery. Adds those files to the hook's -S skip list, the same mechanism already used for other generated/binary content in this repo. Verified the already-committed diagrams are unaffected (git history never saw the bad rewrite — only codespell's own scratch checkout in CI did). Signed-off-by: David Bensoussan <d.bensoussan@proton.me>
Both latent, surfaced only now that verify-robot-manifests/
verify-published-images actually run on a PR (previous commit) instead of
being skipped entirely.
deploy/robot/kubernetes/robot-pod.yaml: dc-ros's robot-params volumeMount
combined a hostPath already pointing at the exact file with `subPath:
robot_params.yaml`, which appends that path again — podman tried to open
.../robot_params.yaml/robot_params.yaml and failed ("not a directory").
Verified locally: podman kube play now runs the Pod to Ready end-to-end
(dc-ros, dc-uploader and vector all Up, "dc_bridge reports ready" logged).
tools/release/scripts/verify_published_images.sh and five e2e scripts
(run_split.sh and friends) extracted VECTOR_VERSION with a bare `grep -oP
'version: v...'` across all of ros2_data_collection.repos, which also
matches aws_sdk_vendor's own `version: v1.11.600` line — collapsing into
"0.57.0\n1.11.600" and breaking podman's image ref parsing
("docker.io/timberio/vector:0.57.0\n1.11.600-debian: invalid reference
format"). Scopes the grep to the vector_vendor block (`grep -A3
'vector_vendor:' | grep -oP ...`) in all six places that shared this
pattern, so the apt and container paths agree on one version everywhere,
consistent with the comments already claiming they do.
Signed-off-by: David Bensoussan <d.bensoussan@proton.me>
github-actions
Bot
deleted the
feature/450-compose-quadlet-and-kubernetes-rendering
branch
September 1, 2026 12:46
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
deploy/robot/: the three-container robot tier (dc-ros,vector,dc-uploader) described three ways —compose.yaml, Podmanquadlet/*.container+*.volume+*.networkunits, and a single Kuberneteskubernetes/robot-pod.yamlPod — so a site deploys it with whatever it already runs, without the topology becoming three different decisions (docs/adr/0015-split-deployment-topology.md).tools/ci/pre-commit/kubeconform.shschema-validates the Kubernetes manifest via Podman, wired into.pre-commit-config.yaml.deploy/robot/scripts/verify_kube_play.sh:podman kube playruns the Pod manifest with no cluster and asserts it reaches ready (samedc_bridge reports readysignaltools/release/scripts/verify_published_images.shalready uses).deploy/robot/scripts/verify_network_isolation.sh: plain Podman networks prove the robot tier's routing claim — no internet route, and only outbound robot-to-edge connectivity permitted. Verified locally end-to-end (all three checks pass, cleanup confirmed).verify-robot-manifests(.github/workflows/ci.yaml) runs both checks on push, against the same publisheddc-ros/dc-uploaderimagesverify-published-imagesalready exercises.Unlike
tools/e2e/compose.split.yaml(the test rendering, with Postgres/RustFS standing in for reachable Destinations), these renderings are the deployable robot tier for the fleet shape (epic #440 scenario 3): Records go out through the blessedvectorDestination type to an edge aggregator, and Files go to edge-local object storage — both placeholder hostnames (edge.site.example), since that edge infrastructure is outside this repository. No new build or publish path — building and shipping still uses Podman, unchanged.Closes #450
Test plan
prek run --files <changed> --skip build-doc— all hooks pass, including the newkubeconformhookkubeconformvalidateskubernetes/robot-pod.yamllocally (Valid: 1, Invalid: 0, Errors: 0)deploy/robot/scripts/verify_network_isolation.shrun locally end-to-end — all three checks pass, no leftover containers/networks after cleanupdeploy/robot/scripts/verify_kube_play.sh— needs the publishedghcr.io/minipada/ros2_data_collection/dc-ros/dc-uploaderimages (private registry, not pullable without CI credentials from this environment); exercised by the newverify-robot-manifestsCI job on pushverify-robot-manifests,Formatting (prek), existing jobs unaffected)🤖 Generated with Claude Code
https://claude.ai/code/session_018bkW24nSMdaGbCJyxhBU23