Known Boundaries¶

Release binding: v0.1.0-enterprise-ready · AMI ami-010806d4d3445660e · 2026-05-19

Stated limits of the shipped release. Reviewers and buyers should read this page alongside Performance Evidence.

1. g5.xlarge is not enterprise-ready (blocking)¶

On g5.xlarge, hx-engine and hx-gate exhibit instability under load (out-of-memory conditions, service restarts). That instance type is below the enterprise floor. The minimum enterprise instance is g5.2xlarge (A10G, 32 GB RAM). Do not deploy production workloads on g5.xlarge.

2. hx-gate does not auto-recover after a Redis restart (P3)¶

If Redis restarts, hx-gate does not automatically reconnect. Workaround:

sudo systemctl restart hx-gate

This is a known gap, planned for resolution in a future release with a health-check-based reconnection loop.

3. Top-K is not bit-stable across a reboot (informational)¶

After a reboot, top-K results overlap approximately 9 of 10 with pre-reboot results for the same query. Scores are preserved; ordering of tied or near-tied scores may shift with GPU index rebuild order. Data durability is unaffected. Applications relying on exact ordering across reboots should filter by score rather than by position.

4. No contractual benchmark or recall claims (informational)¶

The benchmark proof stage was not produced for this release, so no recall, latency, or throughput number is published as a contractual claim for v0.1.0-enterprise-ready. Measured engineering evidence, with operating points, is on the Performance Evidence page. Customers validate on representative workloads with the shipped hx-sdp validate and hx-sdp sizing tools.

5. The KMS-encrypted AMI requires explicit grants (blocking if missing)¶

The AMI in the archived AWS image reference is encrypted with a KMS CMK. This requirement applies only when that image is explicitly supplied for your agreed deployment. Launching principals, and any attached instance profile role, need kms:CreateGrant on:

arn:aws:kms:us-east-1:487916112059:key/e1d9cc91-4aa3-4e1d-9cf3-d6b110694a80

Without the grant, the instance fails to boot with an EBS volume decryption error. Policy fix:

{
  "Effect": "Allow",
  "Action": ["kms:CreateGrant"],
  "Resource": "arn:aws:kms:us-east-1:487916112059:key/e1d9cc91-4aa3-4e1d-9cf3-d6b110694a80",
  "Condition": {
    "Bool": {"kms:GrantIsForAWSResource": "true"}
  }
}

Deployment target boundary¶

Current access is through a Private Appliance or scoped pilot. The Private Appliance target documentation covers VMware OVA on supported NVIDIA GPU hosts; KVM, Proxmox, and OpenStack QCOW2 on supported NVIDIA GPU hosts; and the air-gapped OCI Runtime Kit on supported NVIDIA GPU Kubernetes nodes. Package availability and target validation are agreed for the supplied release.

AWS and Google Cloud image pages are archived technical references, not current marketplace offers. A cloud target requires an explicitly agreed pilot and authorized deployment package.

Unsupported: unmanaged consumer Windows hosts, non-NVIDIA GPU paths, public unauthenticated self-service, customer-modified base images, and unvalidated hypervisor or container runtime combinations. OpenStack QCOW2 may remain gated until OpenStack GPU validation is present in the release dossier.

Performance boundary¶

Latency and capacity guidance is scoped to the validated hardware envelope and documented estimates. Broad max-VRAM latency claims are not part of the release unless target evidence explicitly supports them.

Summary¶

Item	Severity
g5.xlarge instability	Blocking: do not deploy
hx-gate Redis reconnect	P3, workaround available
Top-K ordering drift across reboot	Informational
No contractual benchmark claims	Informational
KMS grant requirement	Blocking if missing