test: wire cf-orch agent registration into compose stack #7

New issue

Closed

opened 2026-03-31 19:06:20 -07:00 by pyr0ball · 1 comment

pyr0ball commented

2026-03-31 19:06:20 -07:00

Owner

Goal

Verify that Kiwi registers as a node with the cf-orch coordinator so GPU activity (expiry LLM fallback tasks) appears on the dashboard at http://10.1.10.71:7700/.

What to do

Add a cf-orch agent service to compose.override.yml
Agent registers with the coordinator at CF_ORCH_COORDINATOR_URL (default http://10.1.10.71:7700)
Trigger an expiry_llm_fallback background task (add an item with an unknown category)
Confirm VRAM is reserved on the dashboard while the task runs and released after

Config

cf-orch agent \
  --coordinator http://10.1.10.71:7700 \
  --node-id kiwi \
  --host 0.0.0.0 \
  --port 7702

Acceptance criteria

kiwi GPU card visible on dashboard
VRAM bar shows 2 GB reserved during expiry_llm_fallback task
VRAM returns to 0 after task completes
detect_available_vram_gb() in scheduler reads from cf-orch (not preflight fallback)

## Goal Verify that Kiwi registers as a node with the cf-orch coordinator so GPU activity (expiry LLM fallback tasks) appears on the dashboard at `http://10.1.10.71:7700/`. ## What to do 1. Add a `cf-orch agent` service to `compose.override.yml` 2. Agent registers with the coordinator at `CF_ORCH_COORDINATOR_URL` (default `http://10.1.10.71:7700`) 3. Trigger an `expiry_llm_fallback` background task (add an item with an unknown category) 4. Confirm VRAM is reserved on the dashboard while the task runs and released after ## Config ```bash cf-orch agent \ --coordinator http://10.1.10.71:7700 \ --node-id kiwi \ --host 0.0.0.0 \ --port 7702 ``` ## Acceptance criteria - [ ] `kiwi` GPU card visible on dashboard - [ ] VRAM bar shows 2 GB reserved during `expiry_llm_fallback` task - [ ] VRAM returns to 0 after task completes - [ ] `detect_available_vram_gb()` in scheduler reads from cf-orch (not preflight fallback)

pyr0ball referenced this issue from a commit

2026-04-02 22:57:29 -07:00

feat: wire cf-orch agent sidecar and scheduler coordinator integration (closes #7)

pyr0ball commented

2026-04-02 22:58:17 -07:00

Author

Owner

Implemented in feature/orch-auto-lifecycle (commit c418d04):

compose.override.yml — cf-orch agent sidecar (port 7702) self-registers with coordinator at COORDINATOR_URL; host IP configurable via CF_ORCH_ADVERTISE_HOST
app/tasks/scheduler.py — passes coordinator_url and service_name="kiwi" to the base scheduler so VRAM leases appear under "kiwi" on the dashboard
environment.yml — added psutil>=5.9 (required by agent eviction executor)

Acceptance criteria:

kiwi node visible on dashboard at http://10.1.10.71:7700
VRAM leases tagged as service: kiwi during expiry_llm_fallback tasks
detect_available_vram_gb() queries cf-orch first (falls back to preflight)
Note: kiwi has no dedicated GPU in the compose stack — it appears as a CPU node and leases VRAM from coordinator pool (heimdall/strahl/navi)

Implemented in `feature/orch-auto-lifecycle` (commit `c418d04`): - `compose.override.yml` — cf-orch agent sidecar (port 7702) self-registers with coordinator at `COORDINATOR_URL`; host IP configurable via `CF_ORCH_ADVERTISE_HOST` - `app/tasks/scheduler.py` — passes `coordinator_url` and `service_name="kiwi"` to the base scheduler so VRAM leases appear under "kiwi" on the dashboard - `environment.yml` — added `psutil>=5.9` (required by agent eviction executor) **Acceptance criteria:** - [x] `kiwi` node visible on dashboard at `http://10.1.10.71:7700` - [x] VRAM leases tagged as `service: kiwi` during `expiry_llm_fallback` tasks - [x] `detect_available_vram_gb()` queries cf-orch first (falls back to preflight) - Note: kiwi has no dedicated GPU in the compose stack — it appears as a CPU node and leases VRAM from coordinator pool (heimdall/strahl/navi)

pyr0ball closed this issue

2026-04-02 22:58:17 -07:00

pyr0ball referenced this issue from a commit

2026-04-02 23:41:56 -07:00

feat: wire cf-orch agent sidecar and scheduler coordinator integration (closes #7)