Issues

Scope the OpenBao init-cluster Job Secret RBAC to the Secrets it manages
NVIDIA/nvcf#2316 · opened on Oct 6, 2026
fix(pylon): canary flags capped reasoning responses as runaway generation
NVIDIA/nvcf#2314 · opened on Oct 6, 2026
OpenBao init Job fails: manual raft join conflicts with retry_join and cannot recover on retry
NVIDIA/nvcf#2313 · opened on Oct 6, 2026
Support auto-unseal (KMS/HSM seals) in the self-managed OpenBao chart
NVIDIA/nvcf#2309 · opened on Oct 6, 2026
Exclude inline base64 from Chat Completions and Messages admission estimates
NVIDIA/nvcf#2303 · opened on Oct 6, 2026
Enforce API key scopes in event-ledger for self-managed deployments
Task
NVIDIA/nvcf#2301 · opened on Oct 6, 2026
nvsnap: import the GPU shared-memory checkpoint library for multi-GPU checkpoint/restore
NVIDIA/nvcf#2299 · opened on Oct 6, 2026
Document a three-Spark Kubernetes cluster setup
NVIDIA/nvcf#2298 · opened on Oct 6, 2026
Pylon fallback input TPS counts cached prompt tokens as prefill throughput
NVIDIA/nvcf#2294 · opened on Oct 5, 2026
1
modelUpdates silently ignores llmConfig.uris and llmConfig.tokenizer
NVIDIA/nvcf#2291 · opened on Oct 5, 2026
feat(stargate,llm-api-gateway): route returning sessions to their last cluster in wait-and-widen and pulsar-wait-and-widen
NVIDIA/nvcf#2289 · opened on Oct 5, 2026
pulsar-wait-and-widen: wait for affinity and widen gradually with an overflow guard
NVIDIA/nvcf#2287 · opened on Oct 5, 2026
MockDynamo: model batched multi-GPU Dynamo deployments
NVIDIA/nvcf#2285 · opened on Oct 5, 2026
fix(stargate): expire in-flight routing reservations after one backend RTT instead of on Pylon updates
NVIDIA/nvcf#2284 · opened on Oct 5, 2026
1
nvca-operator: transient RBAC failure on first allowedExtraKubernetesTypes entry stalls agent rollout
NVIDIA/nvcf#2282 · opened on Oct 5, 2026
grpc-proxy: MaxConcurrentStreams=MaxInt32 disables Go's HTTP/2 rapid-reset protection
NVIDIA/nvcf#2280 · opened on Oct 5, 2026
1
Self-managed: no routing path for pod-addressed worker callbacks, so invocation-service and grpc-proxy are limited to one replica
NVIDIA/nvcf#2276 · opened on Oct 5, 2026
1
self-managed: cassandra-autoscaler published-compatibility test asserts 0.21.3 while the Helmfile pins Cassandra 0.22.0
NVIDIA/nvcf#2264 · opened on Oct 5, 2026
docs: self-managed 1.0.1 pages link to sample files at a path that no longer exists
NVIDIA/nvcf#2262 · opened on Oct 5, 2026
fix(worker): recover from NATS consumer timeout without restarting utils
NVIDIA/nvcf#2260 · opened on Oct 4, 2026
fix(release): align published self-managed bundle and receipt versions
NVIDIA/nvcf#2258 · opened on Oct 4, 2026
docs: resolve live link audit findings
NVIDIA/nvcf#2254 · opened on Oct 3, 2026
nvca: shared-claim model cache writer cannot write a fresh ReadWriteMany volume (fsGroup not applied), caching silently disabled
NVIDIA/nvcf#2248 · opened on Oct 3, 2026
[nvca-operator] BackendK8sCache ignores --system-namespace and always uses nvca-operator
NVIDIA/nvcf#2244 · opened on Oct 3, 2026
1
nvca-operator chart pins the credential-helper and collector images to the public org regardless of image.repository
NVIDIA/nvcf#2237 · opened on Oct 2, 2026