Skip to content

Commit 1bf9605

Browse files
committed
docs: clarify multi-gateway is not multi-tenancy, link upstream issues
Clarify in ADR-0004 that running multiple independent gateway Helm releases in one namespace is not multi-tenancy and does not depend on upstream OpenShell work (NVIDIA/OpenShell#1719, #1722, #1795). Add upstream issue references to the multi-tenancy limitations section. Update AgentPlaybookRun in CONTEXT.md to reference gateway selection instead of stale model selections. Signed-off-by: David Zager <[email protected]>
1 parent c771cd7 commit 1bf9605

2 files changed

Lines changed: 28 additions & 9 deletions

File tree

‎CONTEXT.md‎

Lines changed: 6 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -68,11 +68,12 @@ params (APP_ID, HUB_BASE_URL, etc.) for migration use cases.
6868

6969
**AgentPlaybookRun** — A request to execute an AgentPlaybook. References an
7070
AgentPlaybook (or inlines the spec) and carries generic parameters,
71-
model selections, and env/envFrom. The controller orchestrates the
72-
execution: creates an AgentRun per stage sequentially, all sharing the
73-
same target branch. Cross-stage continuity comes from committed handoff
74-
files (e.g. `.konveyor/handoff.md`). Tracks per-stage status (Pending,
75-
Running, Succeeded, Failed).
71+
a gateway selection, and env/envFrom. The controller orchestrates the
72+
execution: creates an AgentRun per stage sequentially (each using the
73+
selected gateway), all sharing the same target branch. Cross-stage
74+
continuity comes from committed handoff files (e.g.
75+
`.konveyor/handoff.md`). Tracks per-stage status (Pending, Running,
76+
Succeeded, Failed).
7677

7778
## Personas
7879

‎docs/adr/0004-openshell-as-execution-interface.md‎

Lines changed: 22 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -77,14 +77,23 @@ helm install openai-gw oci://ghcr.io/.../helm-chart -n konveyor
7777
```
7878

7979
Each produces a separate Service (`anthropic-gw`, `openai-gw`) that
80-
the controller can reach without cross-namespace access.
80+
the controller can reach without cross-namespace access. Each gateway
81+
is fully independent — its own database, TLS material, provider
82+
config, and inference route. They do not share state.
83+
84+
**This is not multi-tenancy.** OpenShell has open upstream work on
85+
multi-tenant deployments (NVIDIA/OpenShell#1722), a Kubernetes
86+
operator (NVIDIA/OpenShell#1719), and cross-namespace sandbox
87+
creation (NVIDIA/OpenShell#1795). Our design does not depend on any
88+
of these. We are running multiple independent single-user gateway
89+
instances in one namespace — this is supported today via Helm and
90+
does not require upstream multi-tenancy features.
8191

8292
OpenShell is explicit about its current scope: "Alpha software —
8393
single-player mode. One developer, one environment, one gateway. We
8494
are building toward multi-tenant enterprise deployments." There are
8595
no Kubernetes CRDs for providers, inference routes, or policies —
8696
configuration is imperative (CLI/gRPC against the gateway).
87-
Multi-tenancy and multi-namespace support are not yet available.
8897

8998
### UX impact
9099

@@ -111,14 +120,23 @@ not in scope for this decision.
111120
OpenShell's namespace-scoped gateway model means all gateways, all
112121
sandbox pods, and all our CRs live in one namespace today. This works
113122
for our current single-namespace deployment but does not scale to
114-
multi-tenant scenarios where teams need isolated gateway
115-
configurations. Multi-namespace support would require either:
123+
multi-tenant scenarios where teams need namespace-level isolation
124+
between gateway configurations. Relevant upstream work:
125+
126+
- NVIDIA/OpenShell#1719 — Kubernetes operator (status: Idea)
127+
- NVIDIA/OpenShell#1722 — Multi-tenant deployments (milestone:
128+
OpenShell Beta, status: Idea)
129+
- NVIDIA/OpenShell#1795 — Cross-namespace sandbox CRs (open)
130+
131+
Multi-namespace support would require either:
116132

117133
- OpenShell adding cross-namespace or cluster-scoped gateway support
118134
- Us contributing upstream to OpenShell to expand their deployment model
119135
- Running separate controller + OpenShell stacks per namespace
120136

121137
This is a known constraint, not a blocker for the current phase.
138+
Our single-namespace, multi-gateway design works today without any
139+
upstream changes.
122140

123141
## Considered Options
124142

0 commit comments

Comments
 (0)