feat: field-test fixes — vDNS flag, partial updates, kubeconfig envelope + generate tool, force delete#38
Merged
Merged
Conversation
added 2 commits
July 14, 2026 11:20
…ope + generate tool, force delete
Nine items from a field-testing session, each verified against the live API:
- list_vpcs: VpcItem carries enabled_dns (vServer dnsStatus) — clusters
with azStrategy=MULTI can only use a vDNS-enabled VPC; the creation
guide filters accordingly and stops with console guidance when none
qualify.
- update_cluster: the API no longer requires version/whitelistNodeCIDRs
— UpdateClusterDto is now a true partial update (all fields optional,
empty bodies rejected at the handler with the valid-field list).
- configure_auto_upgrade crashed with "Expecting value: line 1 column 0"
— VKS answers several writes with 202 and an EMPTY body while
mcp_core BaseClient called resp.json() unconditionally. Empty success
bodies now return None; five handler messages no longer trail
"None"/"null".
- list_k8s_resources / get_cluster_kubeconfig broke ("Expected key
current-context") — the kubeconfig endpoint now wraps the YAML in a
JSON envelope {kubeConfig, status, ...}. New kubeconfig.py extracts
the YAML (legacy bare-YAML accepted), wired into both K8sClientCache
and the tool; a not-ready cluster gets an actionable error.
- generate_kubeconfig (new write tool, CLI generate-kubeconfig parity):
a new cluster has NO kubeconfig until one is minted — POST
.../kubeconfig {expirationDays, default 30}, asynchronous; the
get-before-generate error teaches the flow. Conventions: generate_
prefix is read-only elsewhere, so this tool is an explicit EXTRA_WRITE
exception (named after the CLI command).
- delete_nodegroup: force_delete param -> forceDelete=true (CLI
--force-delete parity), documented as an escalation after a stuck or
failed normal delete, user-confirmed. The CLI's client-side --force
(skip prompt) is deliberately NOT mapped — that role belongs to the
confirm gate.
- enabledServiceEndpoint is private-cluster-only, default TRUE there:
the guide asked about the wrong field name (enabledServiceEndpointPlugin)
with the wrong default; the DTO field is now Optional and omitted from
the wire body unless set (public clusters never carry it).
- delete_cluster_dryrun listed at most one page of node groups (VKS
enforces pageSize=10) — the deletion preview now pages through, so it
can never under-report what will be destroyed.
- Docs synced: 40 tools, cluster 13, partial update, enabled_dns,
force_delete, kubeconfig-envelope + empty-202 quirks, kubeconfig.py.
The kubeconfig carries a cluster-admin certificate + private key, yet the tool was a plain read — any read-only session could exfiltrate full-admin credentials into the conversation. Same gate as Kubernetes Secrets/logs now (K8sHandler pattern: registered, runtime-denied with an actionable error). - ClusterHandler takes allow_sensitive_data_access; main() passes the CLI flag through. - Docstring ## Requirements, SERVER_INSTRUCTIONS tool line, runtime-mode addendum, getting-started prompt, and README all state the gate. - Note: the K8s tools are unaffected — K8sClientCache uses the kubeconfig internally without exposing it to the conversation.
Contributor
Author
|
Added a second commit: security fix (High) — |
…F-02) The empty-body guard in mcp_core.http (earlier commit in this PR) already fixes both auto-upgrade tools; F-02 reproduced on v0.8.0, which predates it. This adds the missing regression test for the DELETE path so both tools named in the report are covered.
Contributor
Author
|
Bug report F-02 (configure/delete_auto_upgrade error despite success) is fixed by this PR — the report reproduces on v0.8.0, which predates the fix.
|
…s match on
Model-eval finding: "bật auto-healing cho cluster X" sent 3/6 models to
get_cluster first — the one-line description ("Configure auto-healing
for a VKS cluster") never said the tool both ENABLES and DISABLES
healing, nor what it tunes.
- configure_auto_healing: "Enable, disable, or tune..." + the three
tuning knobs (maxUnhealthy, unhealthyRange, timeoutUnhealthy) + a
read-vs-write pointer (get_cluster to only READ the config).
- configure_auto_upgrade: enabling = setting the weekly schedule;
turning OFF is delete_auto_upgrade (cross-ref); reading is
get_cluster.
- delete_auto_upgrade: framed as "turn OFF auto-upgrade", counterpart
of configure_auto_upgrade.
A conventions test pins the intent words, tuning params, and cross-refs.
This was referenced Jul 14, 2026
Merged
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.
What
Nine items from a field-testing session, fixed one by one on this branch, each verified against the live API.
API-change fixes (things that broke)
configure_auto_upgradecrash — VKS answers several writes with 202 + empty body;mcp_core.http.BaseClientcalledresp.json()unconditionally →Expecting value: line 1 column 1. Empty success bodies now returnNone(mcp-core), and five handler messages no longer trail "None"/"null". (touches mcp-core → dual release)list_k8s_resources/get_cluster_kubeconfigbroke (Expected key current-context) — the kubeconfig endpoint now wraps the YAML in a JSON envelope{kubeConfig, status, ...}. Newkubeconfig.pyextracts it (legacy bare YAML still accepted), wired into bothK8sClientCacheand the tool — which had been returning the raw envelope to users.update_cluster— the API no longer requiresversion/whitelistNodeCIDRs: the DTO is now a true partial update (all optional; empty body rejected at the handler with the valid-field list).New capabilities (CLI parity)
generate_kubeconfig(write): a new cluster has no kubeconfig until one is minted —POST .../kubeconfig {expirationDays}(default 30), asynchronous; the get-before-generate error teaches the flow.generate_is read-only by convention elsewhere, so this is an explicitEXTRA_WRITEexception (named after the CLI command).delete_nodegroup.force_delete→forceDelete=true(CLI--force-delete), documented as an escalation after a stuck/failed delete, user-confirmed first. The CLI's client-side--force(skip prompt) is deliberately not mapped — that role belongs to the confirm gate.Product-rule fixes in discovery/guides
list_vpcs.enabled_dns(vServerdnsStatus) —azStrategy=MULTIclusters require a vDNS-enabled VPC; the creation guide filters and stops with console guidance when none qualify.enabledServiceEndpointis private-cluster-only with default true — the guide had the wrong field name (enabledServiceEndpointPlugin) and wrong default; the DTO field is now Optional and omitted from the wire body unless set.delete_cluster_dryrununder-reported — it fetched one page of node groups (VKS enforcespageSize=10), so a cluster with >10 node groups showed an incomplete deletion preview. It now pages through.Docs
READMEs/CLAUDE.md synced: 40 tools, cluster 13, partial update,
enabled_dns,force_delete, kubeconfig-envelope + empty-202 quirks.Tests
vks-mcp-server 264, mcp-core 15, repo conventions green; 18 files, +425/−60. Live-verified: vDNS flags match the console, auto-upgrade configure/delete round-trip, kubeconfig YAML + 9 pods listed in kube-system, generate flow on a CREATING cluster (actionable errors).