Skip to content

test(ats): the platform Harness goes with the kagent release from 1.8.1 #924

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
teemow merged 1 commit into from
Oct 11, 2026
Merged

test(ats): the platform Harness goes with the kagent release from 1.8.1 #924

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
33 changes: 16 additions & 17 deletions tests/ats/test_smoke.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,8 +52,8 @@
the kagent CRDs survive it, with the Agents and
RemoteMCPServer (the line's keep policy), the substrate-crds release is
uninstalled and the three ate.dev CRDs survive it (the Substrate line's
keep policy), the platform Harness (the kagent release's keep policy) —
and nothing else does: no ModelConfig, WorkerPool, worker or controller
keep policy) — and nothing else does: no platform Harness (the kagent
release's object since kagent 1.8.1), no ModelConfig, WorkerPool, worker or controller
in the kept kagent namespace, no SandboxConfig, Substrate's control plane
gone.

Expand Down Expand Up @@ -81,7 +81,6 @@
AGENT_CHART_SEMVER,
AGENT_CHART_URL,
ATE_NAMESPACE,
CONNECTIVITY,
CROSS_CLIENT_AUDIENCE,
FLUX_CRD_SUFFIX,
HARNESS,
Expand Down Expand Up @@ -582,7 +581,9 @@ def recreated() -> Any:
TIMINGS.record(f"deleted Harness {HARNESS} back (drift correction on a requested reconcile, no forceAt)", time.monotonic() - started)
annotations = harness["metadata"].get("annotations") or {}
assert annotations.get("meta.helm.sh/release-name") == "kagent", f"the recreated Harness is not the kagent release's: {annotations}"
assert annotations.get("helm.sh/resource-policy") == "keep", annotations
# kagent 1.8.1 renders the Harness without helm.sh/resource-policy:
# keep (giantswarm/kagent-upstream#371): it is the release's object.
assert annotations.get("helm.sh/resource-policy") is None, annotations
assert harness["spec"] == before["spec"], f"the recreated Harness differs from the deleted one:\n{harness['spec']}\n{before['spec']}"
for name in (DECLARATIVE_AGENT, MANAGED_AGENT):
wait_for_agent_ready(kube, name, timeout=300)
Expand Down Expand Up @@ -728,11 +729,13 @@ def test_uninstall_is_the_ordered_teardown(kube: Kube, helm: Helm, app_deploymen
# helm.sh/resource-policy: keep — survived it, so every Agent and
# RemoteMCPServer is still there in the kept kagent namespace (the
# agents' HelmRelease objects went with the Flux CRDs; the declarative Agent
# was never Helm's). The platform Harness stays too — since 4.8.0 the kagent
# release renders it (harness.create) with helm.sh/resource-policy: keep,
# the runtime of every Agent that names it (through 4.7.19 the
# connectivity release rendered it with the same keep, and the 4.8.0
# upgrade adopted it in place; giantswarm/agent-platform#406). Nothing else of the agent
# was never Helm's). The platform Harness goes with the kagent release:
# from kagent 1.8.1 the release renders it (harness.create) without
# helm.sh/resource-policy: keep, so an entry removed from the values is
# deleted on the next upgrade (giantswarm/kagent-upstream#371). The keep
# of giantswarm/agent-platform#406 served the 4.8.0 hand-over from the
# connectivity release, long done; a reinstall renders the Harness again
# and the kept Agents name it by spec.harnessRef. Nothing else of the agent
# runtime survives: the ModelConfig, the WorkerPool and its workers, the
# controller and its Postgres (the kagent release's), the
# substrate release's SandboxConfig and Substrate's control plane go. The
Expand All @@ -755,11 +758,7 @@ def test_uninstall_is_the_ordered_teardown(kube: Kube, helm: Helm, app_deploymen
assert agents == agents_before, f"Agents after the uninstall {agents} != before {agents_before}"
servers = sorted(s["metadata"]["name"] for s in kube.items("remotemcpservers.api.kagent.dev", namespace=KAGENT_NAMESPACE))
assert servers == [MANAGED_AGENT], f"RemoteMCPServers after the uninstall {servers}: the agent's survives (keep policy), the kagent release's {KAGENT_API_SERVER} goes with the release"
harness = kube.get("harnesses.api.kagent.dev", HARNESS, namespace=KAGENT_NAMESPACE)
assert harness, f"the platform Harness {HARNESS} went with the uninstall of the kagent release; it must be kept (helm.sh/resource-policy: keep, giantswarm/agent-platform#406)"
harness_annotations = harness["metadata"].get("annotations") or {}
assert harness_annotations.get("helm.sh/resource-policy") == "keep", harness_annotations
assert harness_annotations.get("meta.helm.sh/release-name") == "kagent", f"the kept Harness is not the kagent release's (components.kagent.chart = the releaseName): {harness_annotations}"
assert kube.get("harnesses.api.kagent.dev", HARNESS, namespace=KAGENT_NAMESPACE) is None, f"the platform Harness {HARNESS} survived the uninstall of the kagent release, whose object it is since kagent 1.8.1 (giantswarm/kagent-upstream#371)"
assert not kube.items("modelconfigs.api.kagent.dev", namespace=KAGENT_NAMESPACE), "ModelConfigs survived the uninstall of the kagent release"
assert kube.get("sandboxconfigs.ate.dev", SANDBOX_CONFIG) is None, "the substrate release's SandboxConfig survived its uninstall"
assert not kube.items("workerpools.ate.dev", all_namespaces=True), "a WorkerPool survived the kagent release's uninstall"
Expand All @@ -777,11 +776,11 @@ def test_uninstall_is_the_ordered_teardown(kube: Kube, helm: Helm, app_deploymen
bundles = substrate_trust_bundles(kube)
assert bundles == sorted(s.replace("/", ":") + ":primary-bundle" for s in PODCERT_SIGNERS), f"the podcert signers' ClusterTrustBundles after the uninstall: {bundles}"
assert_substrate_trust_chain(kube)
logger.info("kept after the uninstall: CRDs %s; in %s Agents %s, RemoteMCPServers %s (their HelmReleases %s are gone with the Flux CRDs), the Harness %s (keep policy, the %s release's); of Substrate: %s with its pools, %s with its two CA pools (keep policy), the ClusterTrustBundles %s carrying the pools' roots",
sorted(KEPT_CRDS), KAGENT_NAMESPACE, agents, servers, agent_hrs_before, HARNESS, CONNECTIVITY, ATE_NAMESPACE, PODCERT_NAMESPACE, bundles)
logger.info("kept after the uninstall: CRDs %s; in %s Agents %s, RemoteMCPServers %s (their HelmReleases %s are gone with the Flux CRDs), the Harness %s gone with the kagent release; of Substrate: %s with its pools, %s with its two CA pools (keep policy), the ClusterTrustBundles %s carrying the pools' roots",
sorted(KEPT_CRDS), KAGENT_NAMESPACE, agents, servers, agent_hrs_before, HARNESS, ATE_NAMESPACE, PODCERT_NAMESPACE, bundles)
assert elapsed < UNINSTALL_BUDGET_S, f"helm uninstall --wait took {elapsed:.0f}s (budget {UNINSTALL_BUDGET_S}s)"
logger.info("uninstall clean in %.0f s: no Flux CRD, operator CRDs kept, no controller, no Job, no release", elapsed)
# Leave the next scenario a cluster without the kept Agents and Harness
# Leave the next scenario a cluster without the kept Agents
# (its own kagent runs there; the kept CRDs it adopts); the namespace's termination
# completes in the background. Substrate's leftovers stay: the own-Flux
# scenario is the reinstall onto them (giantswarm/agent-platform#384).
Expand Down
Loading