Skip to content

Commit ab45f1f

Browse files
committed
docs(nri-imex): document real nvidia-imex-ctl domain scenario
Signed-off-by: Giulio Calzolari <gcalzolari@nvidia.com>
1 parent e7f67b1 commit ab45f1f

2 files changed

Lines changed: 19 additions & 1 deletion

File tree

CHANGELOG.md

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -13,6 +13,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
1313
NRI foundation, with a dedicated `nri-imex` demo and e2e scenario. (#437)
1414

1515
### Changed
16+
- The nri-imex demo now exercises the REAL `nvidia-imex-ctl` against a live
17+
NO-GPU IMEX domain formed across both workers (`-q` READY, `-N -j` UP with
18+
2 nodes READY/NO_GPU), replacing the previous best-effort domain-status
19+
check. (#437)
1620
- ComputeDomain simulation now runs the REAL `nvidia-imex` daemon in NO
1721
GPU mode (`--nogpu`) instead of the fake marker-file binaries: the new
1822
`imex-nogpu-shim` injects the flag around upstream's hard-coded argv,

docs/demo/nri-imex/README.md

Lines changed: 15 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -12,13 +12,27 @@ a single ComputeDomain spanning two Kind nodes (issue
1212
./docs/demo/nri-imex/run.sh
1313
```
1414

15+
## Prerequisites
16+
17+
- Docker, Kind, kubectl, Helm.
18+
- `jq` — Scenario 3 parses `nvidia-imex-ctl -N -j` JSON.
19+
20+
The demo builds a local overlay image (`nvml-mock:nri-imex-real-imex`) that
21+
layers the real `nvidia-imex` / `nvidia-imex-ctl` (NO GPU mode via
22+
`imex-nogpu-shim`) onto the mock stack. This image is **local build only** — it
23+
repackages the proprietary `nvidia-imex` binary and is never published (see
24+
`deployments/nvml-mock/Dockerfile.compute-domain-daemon`).
25+
1526
## What it proves
1627

1728
1. A plain workload annotated `nvml-mock.nvidia.com/imex-channels: "true"` sees
1829
`channel0..15` on **both** workers.
1930
2. `check-fabric` reports the same `clusterUuid` on both workers (consistent
2031
ComputeDomain identity, injected purely through NRI).
21-
3. The real `nvidia-imex` domain reports `READY`.
32+
3. Two real `nvidia-imex` NO-GPU daemons (started via `imex-nogpu-shim`) form a
33+
domain across both workers: `nvidia-imex-ctl -q` reports `READY` locally and
34+
`nvidia-imex-ctl -N -j` reports the domain `UP` with both nodes `READY` and
35+
version `NO_GPU`.
2236

2337
## Limitation: presence-only channels
2438

0 commit comments

Comments
 (0)