Merge origin/main — integrate monitoring/Phase3 updates with Qwen3-8B no-think deployment
Resolved add/add conflicts in: - defaults/main.yml: kept our version (5 original models + Qwen3-8B x2 + rows 5-6) - tasks/swapmode.yml: kept our version (7-model GATE 2 assert) - templates/llama-server-router-preset.ini.j2: kept our version (+Qwen3-8B sections) - templates/llama-swap-config.yaml.j2: kept our version (+chat_template_file support) Remote changes incorporated from origin/main (14 commits): - Ansible Phase 3 integration (llama-swap.service.j2, tasks/monitoring.yml) - Prometheus monitoring: PrometheusRule, Grafana dashboard, scrape config - VRAM exporter script, llama-swap-phase3 cutover results - Day2 playbooks: nomic_embed, cpu_offload_aux, per_model_ctx, qwen38_ctx128k - Router: CPU-offload Coder-14B + Llama-3.1-8B - host_vars/astro-orbiter/vars.yml updates
This commit is contained in:
Hermes Agent service account
@@ -0,0 +1,148 @@
|
||||
# War Machine Phase 3 Cutover Results: 2026-08-18
|
||||
|
||||
## Execution Summary
|
||||
|
||||
**Date:** 2026-08-18
|
||||
**Component:** llama-swap Phase 3 Go-Live
|
||||
**Agent:** War Machine (Hermes Profile) / Wong (Infrastructure)
|
||||
**Status:** ✅ LIVE
|
||||
|
||||
---
|
||||
|
||||
## VRAM Baseline (Steady-State)
|
||||
|
||||
### Measured on astro-orbiter (RTX 3090 24 GB) at 18:45 UTC
|
||||
|
||||
```
|
||||
GPU Memory Profile (nvidia-smi)
|
||||
=================================
|
||||
Total VRAM: 24576 MiB
|
||||
Model loads (current):
|
||||
- Qwen3.8-27B-Q4_K_M: ~17,100 MiB (main model)
|
||||
- KV-cache @ 65K ctx: ~6,000 MiB (dynamic, per request)
|
||||
- llama-server overhead: ~460 MiB (llama.cpp runtime)
|
||||
|
||||
Steady-state used: ~18,560 MiB
|
||||
Free headroom: ~6,000 MiB (reserved for KV-cache peaks)
|
||||
```
|
||||
|
||||
**Key insight:** Qwen3.8-27B-Q4_K_M quantization (Q4_K_M) leaves ~6 GB for KV-cache, which comfortably holds 2-3 concurrent requests at max context (65K tokens each).
|
||||
|
||||
### Memory Pressure Profile
|
||||
|
||||
| Scenario | VRAM Used | Headroom | Status |
|
||||
|----------|-----------|----------|--------|
|
||||
| Idle (no requests) | 17,100 MiB | ~7.5 GB | ✅ Green |
|
||||
| 1 max-ctx request (65K) | ~23,100 MiB | ~1.5 GB | ⚠️ Yellow |
|
||||
| 2 concurrent mid-ctx (32K ea) | ~22,500 MiB | ~2 GB | ⚠️ Yellow |
|
||||
| 3+ concurrent or >65K demand | >24,000 MiB | 0 | 🔴 Red (OOM risk) |
|
||||
|
||||
**Alert thresholds set accordingly:**
|
||||
- **Critical:** > 24,000 MiB (90%+ of 24 GB)
|
||||
- **Warning:** > 23,000 MiB (94%+) — investigate request patterns
|
||||
|
||||
---
|
||||
|
||||
## KV-Cache Utilization
|
||||
|
||||
### Qwen3.8-27B @ 65,536 token context (Q4_K_M)
|
||||
|
||||
- **Allocated KV-cache per request:** ~6000 MiB ÷ (concurrent_requests) = ~2000 MiB per request (3 slots)
|
||||
- **Critical spill threshold:** 92% occupancy (triggers alert; requests may drop from queue)
|
||||
- **Observed during Phase 2 validation:** Never exceeded 45% under normal load; no spill observed
|
||||
|
||||
### Multi-Model Scenario (router mode, not active Phase 3)
|
||||
|
||||
If router mode were re-enabled with Coder (14B) + Llama (8B) models (CPU-offloaded), each would allocate a small KV slot (~500 MiB each at 16K/8K contexts). Qwen3.8's 6 GB slot dominates; co-resident models are negligible.
|
||||
|
||||
---
|
||||
|
||||
## Latency Profile
|
||||
|
||||
### Prediction Latency (tokens/second)
|
||||
|
||||
Measured under synthetic load (30 concurrent requests, each 100 tokens):
|
||||
|
||||
| Model | Ctx Size | Batch | Latency | Tokens/sec | Notes |
|
||||
|-------|----------|-------|---------|------------|-------|
|
||||
| Qwen3.8-27B | 65K | 4096 ubatch | 18 ms/tok | ~56 | Q4_K_M, GPU-resident |
|
||||
|
||||
**Observed degradation:** No throttling under sustained load in Phase 2 testing. Latency remained stable within ±2 ms variance, suggesting no thermal or memory-pressure effects.
|
||||
|
||||
---
|
||||
|
||||
## Request Queue Behavior
|
||||
|
||||
### Normal Load
|
||||
|
||||
- **Baseline queue depth:** 0-1 requests (immediate processing)
|
||||
- **Observed max during Phase 2:** 8 requests (occurred briefly when Hermes profile test script fired 10 parallel requests)
|
||||
- **Clear time (from max queue to idle):** ~90 seconds
|
||||
|
||||
### Alert Trigger
|
||||
|
||||
Queue depth > 5 sustained for >30s indicates model cannot keep up; investigate incoming request rate or queue timeout misconfiguration.
|
||||
|
||||
---
|
||||
|
||||
## Error Rate
|
||||
|
||||
**Observed in Phase 1-2 shadow testing:** 0 errors (100% success rate on valid requests).
|
||||
|
||||
- No HTTP 5xx responses
|
||||
- No request timeouts
|
||||
- No OOM-kills (even at 94% VRAM usage)
|
||||
- No kernel panics
|
||||
|
||||
**Phase 3 production (first 2 hours):** Monitoring TBD (dashboard not yet deployed).
|
||||
|
||||
---
|
||||
|
||||
## Comparison to Phase 2 Validation Gate Results
|
||||
|
||||
| Gate | Requirement | Phase 2 Result | Status |
|
||||
|------|-------------|----------------|--------|
|
||||
| Gate 1: Context | n_ctx >= 64000 | n_ctx_train = 1,010,000 (Qwen3.8-27B-Instruct-1M) | ✅ Pass |
|
||||
| Gate 2: Tool-calling | tool_calls on valid, none on invalid | 10/10 valid, 0/10 invalid (zero hallucinations) | ✅ Pass |
|
||||
| Gate 3: Throughput | >= 50 tokens/sec sustained | 56 tokens/sec @ 65K ctx, 4096 batch | ✅ Pass |
|
||||
| Gate 4: Stability | No OOM, no errors @ 94% VRAM | 2h continuous load, 0 errors | ✅ Pass |
|
||||
|
||||
All gates cleared; **Phase 3 production go-live approved.**
|
||||
|
||||
---
|
||||
|
||||
## Monitoring Gaps (Phase 3 Action Items)
|
||||
|
||||
The following monitoring components are **not yet deployed** as of cutover:
|
||||
|
||||
1. **VRAM textfile exporter** — this task (Wong)
|
||||
2. **Prometheus scrape config** — this task (Wong)
|
||||
3. **Grafana dashboard (6 panels)** — this task (Wong)
|
||||
4. **Alert rules (PrometheusRule CR)** — this task (Wong)
|
||||
|
||||
All are specified in the Ciro Luciotta monitoring pattern (`references/monitoring-llm-homelab-ciro-luciotta-2026.md`).
|
||||
|
||||
**ETA deployment:** 2026-08-18 (today, within 4 hours of cutover).
|
||||
|
||||
---
|
||||
|
||||
## Post-Launch Notes
|
||||
|
||||
- **Model was pre-downloaded** to `/opt/models/Qwen3.8-27B-Q4_K_M.gguf` (17.1 GB) on 2026-08-17 via manual `wget`.
|
||||
- **Configuration:** `/etc/llama-swap/config.yaml`, hand-authored in Phase 1, now templated in Ansible (see `templates/llama-swap-config.yaml.j2`).
|
||||
- **Service:** `systemctl status llama-swap` confirms it is running and has processed ~500+ requests in the first 30 minutes post-cutover.
|
||||
- **Next phase:** Once monitoring dashboard is live, track VRAM spikes under production Hermes workload (real tool-calling traffic, not synthetic).
|
||||
|
||||
---
|
||||
|
||||
## Sign-off
|
||||
|
||||
**Infrastructure readiness:** ✅ Confirmed by Wong
|
||||
**Hermes validation (tool-calling):** ✅ Confirmed by War Machine
|
||||
**Production cutover:** ✅ LIVE 2026-08-18 18:45 UTC
|
||||
|
||||
---
|
||||
|
||||
**Author:** War Machine (execution), Wong (documentation)
|
||||
**Reviewed by:** Ryan (approval)
|
||||
**Prepared for:** Hermes monitoring Phase 3 integration
|
||||
@@ -0,0 +1,278 @@
|
||||
# GPU/LLM Monitoring Pattern: Ciro Luciotta 2026
|
||||
|
||||
## Overview
|
||||
|
||||
This document describes the standardized monitoring stack for llama-swap and llama-server deployments on the homelab. It defines:
|
||||
|
||||
1. **VRAM textfile exporter** — nvidia-smi-based metrics written to node_exporter's textfile collector
|
||||
2. **llama-swap native /metrics endpoint** — built-in OpenMetrics output from llama.cpp
|
||||
3. **Prometheus scrape jobs** — configuration to ingest both sources
|
||||
4. **Grafana dashboard panels** — visualization of VRAM, KV-cache, latency, queue depth, errors, and context usage
|
||||
5. **Alert rules** — PrometheusRule CRs for VRAM saturation, KV-cache spill, and throughput degradation
|
||||
|
||||
## VRAM Textfile Exporter
|
||||
|
||||
### Purpose
|
||||
|
||||
The VRAM exporter runs as a 15-second cron job on the GPU host, using `nvidia-smi` to query instantaneous VRAM usage and writes a Prometheus-formatted `nvidia.prom` file to node_exporter's textfile collector (`/var/lib/node_exporter/textfile_collector/`).
|
||||
|
||||
node_exporter automatically discovers `.prom` files in this directory and exposes them at `GET /metrics`, so new metrics appear immediately without restarting node_exporter.
|
||||
|
||||
### Script (`nvidia-smi-vram-exporter.sh`)
|
||||
|
||||
Location: `roles/llm-inference-multimodel/scripts/nvidia-smi-vram-exporter.sh`
|
||||
|
||||
```bash
|
||||
#!/bin/bash
|
||||
# Description: NVIDIA VRAM textfile exporter for Prometheus
|
||||
# Writes llamacpp_vram_used_mib to node_exporter's textfile collector.
|
||||
# Cron: */1 * * * * (every 1 minute, the script runs every 15s internally)
|
||||
# Output: /var/lib/node_exporter/textfile_collector/nvidia.prom
|
||||
|
||||
TEXTFILE_DIR="/var/lib/node_exporter/textfile_collector"
|
||||
OUTPUT_FILE="${TEXTFILE_DIR}/nvidia.prom"
|
||||
TMPFILE="${OUTPUT_FILE}.tmp"
|
||||
|
||||
# Query nvidia-smi for GPU 0 (RTX 3090)
|
||||
GPU_INDEX=0
|
||||
VRAM_MIB=$(nvidia-smi --query-gpu=memory.used --format=csv,noheader,nounits --id=$GPU_INDEX)
|
||||
|
||||
# Handle nvidia-smi failure
|
||||
if [ -z "$VRAM_MIB" ] || ! [[ "$VRAM_MIB" =~ ^[0-9]+$ ]]; then
|
||||
VRAM_MIB=0
|
||||
fi
|
||||
|
||||
# Write metric to temp file (atomic swap)
|
||||
cat > "$TMPFILE" << EOF
|
||||
# HELP llamacpp_vram_used_mib GPU VRAM used in MiB (nvidia-smi)
|
||||
# TYPE llamacpp_vram_used_mib gauge
|
||||
llamacpp_vram_used_mib $VRAM_MIB
|
||||
EOF
|
||||
|
||||
# Atomic swap to avoid partial reads
|
||||
mv "$TMPFILE" "$OUTPUT_FILE"
|
||||
```
|
||||
|
||||
**Invocation:** Every minute via cron. The script itself is idempotent and cheap to run.
|
||||
|
||||
### Metric Produced
|
||||
|
||||
```
|
||||
llamacpp_vram_used_mib{instance="10.1.71.130:9100",job="node"} 18560
|
||||
```
|
||||
|
||||
- **Metric name:** `llamacpp_vram_used_mib`
|
||||
- **Type:** Gauge
|
||||
- **Unit:** MiB
|
||||
- **Update frequency:** ~1 minute (node_exporter scrape interval)
|
||||
- **Cardinality:** 1 per GPU host (no labels beyond Prometheus scrape labels)
|
||||
|
||||
### Installation
|
||||
|
||||
Deployed by `roles/llm-inference-multimodel/tasks/monitoring.yml` (Phase X — TBD).
|
||||
|
||||
1. Copy script to `/opt/llama-server-monitoring/nvidia-smi-vram-exporter.sh` (owned by `jarvis:jarvis`, mode 0755)
|
||||
2. Create crontab entry: `* * * * * /opt/llama-server-monitoring/nvidia-smi-vram-exporter.sh`
|
||||
3. Verify: `stat /var/lib/node_exporter/textfile_collector/nvidia.prom` (file should update every minute)
|
||||
|
||||
---
|
||||
|
||||
## llama-swap Native Metrics (`/metrics` endpoint)
|
||||
|
||||
### Purpose
|
||||
|
||||
llama.cpp (and llama-swap's embedded instance) exposes Prometheus metrics natively at port 8001 (or the configured `llm_swapmode_port`), under the `/metrics` path.
|
||||
|
||||
This endpoint requires **no additional exporter process** — it's built into llama-swap binary.
|
||||
|
||||
### Metrics Exposed
|
||||
|
||||
**Per-model metrics** (labelled with `model="<model-id>"`):
|
||||
|
||||
- `llamacpp_tokens_predicted_total` — cumulative tokens generated (counter)
|
||||
- `llamacpp_tokens_evaluated_total` — cumulative tokens processed (counter)
|
||||
- `llamacpp_kv_cache_usage_ratio` — KV-cache occupancy as fraction [0.0, 1.0] (gauge)
|
||||
- `llamacpp_time_predict_ms` — per-token prediction latency in milliseconds (histogram)
|
||||
- `llamacpp_queue_size` — current request queue depth (gauge)
|
||||
|
||||
**Global metrics:**
|
||||
|
||||
- `llamacpp_vram_max_mib` — total VRAM available (gauge, set once at startup)
|
||||
- No global VRAM "used" metric (use the textfile exporter for that)
|
||||
|
||||
### Example Scrape
|
||||
|
||||
```
|
||||
GET http://10.1.71.130:8001/metrics HTTP/1.1
|
||||
|
||||
HTTP/1.1 200 OK
|
||||
Content-Type: application/openmetrics-text; version=1.0.0; charset=utf-8
|
||||
|
||||
# HELP llamacpp_tokens_predicted_total Total tokens predicted by llama.cpp
|
||||
# TYPE llamacpp_tokens_predicted_total counter
|
||||
llamacpp_tokens_predicted_total{model="Qwen3.8-27B-Q4_K_M"} 42512
|
||||
llamacpp_tokens_predicted_total{model="Meta-Llama-3.1-8B-Instruct-Q4_K_M"} 18956
|
||||
...
|
||||
```
|
||||
|
||||
### Prometheus Scrape Job
|
||||
|
||||
Defined in `cluster/applications/monitoring/values.yaml`:
|
||||
|
||||
```yaml
|
||||
additionalScrapeConfigs:
|
||||
- job_name: llama-swap
|
||||
static_configs:
|
||||
- targets: ["10.1.71.130:8001"]
|
||||
scrape_interval: 30s
|
||||
scrape_timeout: 10s
|
||||
honor_labels: true
|
||||
metrics_path: /metrics
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Grafana Dashboard Panels
|
||||
|
||||
### Panel 1: VRAM over time (stacked area)
|
||||
|
||||
- **Title:** GPU VRAM Usage
|
||||
- **Metric:** `llamacpp_vram_used_mib{job="node"}`
|
||||
- **Graph type:** Stacked area chart
|
||||
- **Time range:** Last 24 hours (configurable)
|
||||
- **Y-axis:** MiB, max ~24576 (RTX 3090 physical limit)
|
||||
- **Alert line:** 24000 MiB (90% threshold for warning)
|
||||
|
||||
Displays the textfile-exporter VRAM as a single time series. Spike analysis shows when models load/unload or garbage-collection occurs.
|
||||
|
||||
### Panel 2: KV-cache utilization per model (gauge + time series)
|
||||
|
||||
- **Title:** KV-Cache Utilization by Model
|
||||
- **Metrics:**
|
||||
- Gauge (multi-stat): `llamacpp_kv_cache_usage_ratio{model="..."}`
|
||||
- Time series: same metric over time
|
||||
- **Thresholds:**
|
||||
- 0.0 - 0.8: Green ("Healthy")
|
||||
- 0.8 - 0.92: Yellow ("Caution")
|
||||
- 0.92 - 1.0: Red ("Critical")
|
||||
- **Alert line:** 0.92 (spill threshold)
|
||||
|
||||
Each model gets its own gauge and time series below. Tracks which models are approaching context-window limits.
|
||||
|
||||
### Panel 3: Latency by model (histogram)
|
||||
|
||||
- **Title:** Prediction Latency by Model
|
||||
- **Metric:** `rate(llamacpp_time_predict_ms_sum[5m]) / rate(llamacpp_time_predict_ms_count[5m])` (moving avg)
|
||||
- **Graph type:** Line chart, one series per model
|
||||
- **Y-axis:** Milliseconds per token (lower is faster)
|
||||
- **Legend:** Show model names
|
||||
|
||||
Tracks per-token generation speed. Degradation indicates queueing or memory pressure.
|
||||
|
||||
### Panel 4: Queue depth (line)
|
||||
|
||||
- **Title:** Request Queue Depth
|
||||
- **Metric:** `llamacpp_queue_size{model="..."}`
|
||||
- **Graph type:** Line chart, stacked (one per model) or overlaid
|
||||
- **Y-axis:** Number of pending requests
|
||||
- **Alert line:** 5+ requests (threshold for investigation)
|
||||
|
||||
High queue depth indicates the model cannot keep up with incoming load.
|
||||
|
||||
### Panel 5: Error rate (counter)
|
||||
|
||||
- **Title:** Request Errors
|
||||
- **Metric:** Rate of HTTP 5xx / network errors (inferred from llama-swap logs or a custom counter, TBD)
|
||||
- **Graph type:** Line chart
|
||||
- **Y-axis:** Errors per minute
|
||||
|
||||
Currently no native llama-swap error counter; may require a custom sidecar or log-shipper to emit this. Mark as "TBD" for now; use for post-incident analysis.
|
||||
|
||||
### Panel 6: Context-used distribution (histogram)
|
||||
|
||||
- **Title:** Context Window Usage Distribution
|
||||
- **Metric:** Histogram of `context_window_tokens` per request (if llama-swap exposes this; fallback: model's n_ctx_train)
|
||||
- **Graph type:** Histogram / distribution chart
|
||||
- **X-axis:** Token count bins
|
||||
- **Y-axis:** Frequency (request count)
|
||||
|
||||
Shows whether workload is sparse (small contexts) or dense (full context windows). Helps capacity planning.
|
||||
|
||||
---
|
||||
|
||||
## Alert Rules
|
||||
|
||||
Defined in `roles/llm-inference-multimodel/templates/llama-swap-alerts.yml.j2` and applied via ArgoCD as a PrometheusRule CR.
|
||||
|
||||
### Alert 1: VRAM saturation (Critical)
|
||||
|
||||
```yaml
|
||||
alert: LlamaSwapVramSaturation
|
||||
expr: llamacpp_vram_used_mib > 24000
|
||||
for: 1m
|
||||
severity: critical
|
||||
description: GPU VRAM usage exceeds 24000 MiB on {{ $labels.instance }}
|
||||
```
|
||||
|
||||
**Threshold:** > 24000 MiB (90% of 24 GB RTX 3090)
|
||||
**Duration:** Sustained for 1 minute
|
||||
**Action:** Page oncall. Model(s) will begin OOM-killing processes within minutes if this is not resolved.
|
||||
|
||||
### Alert 2: KV-cache spill (Warning)
|
||||
|
||||
```yaml
|
||||
alert: LlamaSwapKvCacheSpill
|
||||
expr: llamacpp_kv_cache_usage_ratio{model="..."} > 0.92
|
||||
for: 2m
|
||||
severity: warning
|
||||
description: KV-cache utilization {{ $value }} on model {{ $labels.model }}
|
||||
```
|
||||
|
||||
**Threshold:** > 0.92 (92% of allocated KV-cache)
|
||||
**Duration:** Sustained for 2 minutes
|
||||
**Action:** Investigate incoming request context-window distribution. Consider reducing `n_ctx` for non-critical models or routing long-context requests to a different model.
|
||||
|
||||
### Alert 3: Throughput degradation (Warning)
|
||||
|
||||
```yaml
|
||||
alert: LlamaSwapThroughputDegradation
|
||||
expr: rate(llamacpp_tokens_predicted_total[5m]) < (baseline_tokens_per_minute * 0.8)
|
||||
for: 5m
|
||||
severity: warning
|
||||
description: Prediction throughput on {{ $labels.model }} is {{ $value }}% of baseline
|
||||
```
|
||||
|
||||
**Threshold:** < 80% of baseline tokens/minute
|
||||
**Duration:** Sustained for 5 minutes
|
||||
**Action:** Check queue depth, VRAM usage, and model temperatures. May indicate thermal throttling or resource contention.
|
||||
|
||||
**Baseline:** Set per-model during validation Phase 2. Example: Qwen3.8-27B at 65K context should sustain ~200 tokens/min under continuous load.
|
||||
|
||||
---
|
||||
|
||||
## Dashboarding Best Practices
|
||||
|
||||
1. **Time ranges:** Default to "Last 24 hours"; allow user selection from 1h to 7d.
|
||||
2. **Refresh rate:** 30 seconds (matches Prometheus scrape interval).
|
||||
3. **Alerting integration:** Grafana "Alert state" panel shows active alerts and provides one-click drill-down.
|
||||
4. **Annotations:** Mark model deployments, upgrades, or maintenance windows with vertical lines.
|
||||
5. **Multi-instance support:** If homelab expands to multiple GPU hosts, use `instance` label in all queries to keep dashboards reusable.
|
||||
|
||||
---
|
||||
|
||||
## Validation Checklist (Deployment)
|
||||
|
||||
- [ ] VRAM exporter script installed, executable, and cron job active
|
||||
- [ ] VRAM metric appears in node_exporter's `/metrics` within 2 minutes
|
||||
- [ ] Prometheus scrape of `10.1.71.130:8001/metrics` returns HTTP 200
|
||||
- [ ] All 6 dashboard panels render without errors
|
||||
- [ ] Alert rules parse without syntax errors in Prometheus
|
||||
- [ ] Alert rules return the correct cardinality (e.g., one alert per model for KV-cache thresholds)
|
||||
|
||||
---
|
||||
|
||||
## References
|
||||
|
||||
- Ciro Luciotta, "Real-time Observability for Edge LLM Inference", 2026 (internal)
|
||||
- llama.cpp metrics documentation: https://github.com/ggerganov/llama.cpp/blob/master/examples/main/README.md#metrics
|
||||
- Prometheus AlertManager routing: https://prometheus.io/docs/prometheus/latest/configuration/alerting_rules/
|
||||
@@ -0,0 +1,57 @@
|
||||
#!/bin/bash
|
||||
# ==============================================================================
|
||||
# FILE: roles/llm-inference-multimodel/scripts/nvidia-smi-vram-exporter.sh
|
||||
# DESCRIPTION: NVIDIA VRAM textfile exporter for Prometheus
|
||||
# Queries nvidia-smi for GPU VRAM usage and writes Prometheus-
|
||||
# formatted metrics to node_exporter's textfile collector
|
||||
# (/var/lib/node_exporter/textfile_collector/).
|
||||
#
|
||||
# Designed for 1-minute cron execution (idempotent; atomic writes).
|
||||
# Outputs: llamacpp_vram_used_mib (gauge, MiB)
|
||||
#
|
||||
# CRON ENTRY: * * * * * /opt/llama-server-monitoring/nvidia-smi-vram-exporter.sh
|
||||
# OUTPUT FILE: /var/lib/node_exporter/textfile_collector/nvidia.prom
|
||||
#
|
||||
# AUTHOR: Wong (Infrastructure Automation Specialist)
|
||||
# DATE: 2026-08-18
|
||||
# ==============================================================================
|
||||
|
||||
set -euo pipefail
|
||||
|
||||
# Configuration
|
||||
TEXTFILE_DIR="/var/lib/node_exporter/textfile_collector"
|
||||
OUTPUT_FILE="${TEXTFILE_DIR}/nvidia.prom"
|
||||
TMPFILE="${OUTPUT_FILE}.tmp.$$"
|
||||
GPU_INDEX="${1:-0}" # Allow override via first positional arg; default GPU 0
|
||||
|
||||
# Ensure textfile collector directory exists
|
||||
if [ ! -d "$TEXTFILE_DIR" ]; then
|
||||
echo "ERROR: $TEXTFILE_DIR does not exist. Create it with:" >&2
|
||||
echo " mkdir -p $TEXTFILE_DIR" >&2
|
||||
echo " chown prometheus:prometheus $TEXTFILE_DIR" >&2
|
||||
exit 1
|
||||
fi
|
||||
|
||||
# Query nvidia-smi for instantaneous GPU VRAM usage
|
||||
# Format: plain number (MiB), or empty if nvidia-smi fails
|
||||
VRAM_MIB=$(nvidia-smi --query-gpu=memory.used \
|
||||
--format=csv,noheader,nounits \
|
||||
--id="$GPU_INDEX" 2>/dev/null || echo "")
|
||||
|
||||
# Validate output is a number; default to 0 if nvidia-smi fails
|
||||
if [ -z "$VRAM_MIB" ] || ! [[ "$VRAM_MIB" =~ ^[0-9]+$ ]]; then
|
||||
VRAM_MIB=0
|
||||
fi
|
||||
|
||||
# Write metric to temp file (atomic swap to avoid partial reads)
|
||||
cat > "$TMPFILE" << EOF
|
||||
# HELP llamacpp_vram_used_mib GPU VRAM used in MiB (nvidia-smi)
|
||||
# TYPE llamacpp_vram_used_mib gauge
|
||||
llamacpp_vram_used_mib $VRAM_MIB
|
||||
EOF
|
||||
|
||||
# Atomic swap: move temp file to final location
|
||||
# This ensures node_exporter never reads a partial file
|
||||
mv "$TMPFILE" "$OUTPUT_FILE"
|
||||
|
||||
exit 0
|
||||
@@ -60,3 +60,37 @@
|
||||
- include_tasks: preset.yml
|
||||
when: llm_router_preset_enabled | default(false)
|
||||
tags: [always]
|
||||
|
||||
# Phase S — llama-swap mode hot-swap proxy (port 8001)
|
||||
# Gates on llm_swapmode_enabled (default false — complete no-op until enabled).
|
||||
# Replaces router mode entirely: single Go binary + YAML config, no INI presets.
|
||||
# Additive deployment (non-invasive); production router (port 8002) stays running during Phase 1 shadow.
|
||||
#
|
||||
# When llm_swapmode_enabled: true, this phase:
|
||||
# swapmode_binary — download + install llama-swap binary
|
||||
# swapmode_config — render config.yaml.j2 template
|
||||
# swapmode_systemd — deploy llama-swap.service unit
|
||||
# swapmode_firewall — open port 8001 scoped to Hermes subnet
|
||||
# swapmode_verify — start service, run validation gates
|
||||
#
|
||||
# Added 2026-08-18 (t_c1e44190): llama-swap Phase 3 Ansible integration — Wong.
|
||||
- include_tasks: swapmode.yml
|
||||
when: llm_swapmode_enabled | default(false)
|
||||
tags: [always]
|
||||
|
||||
# Phase M — GPU/LLM Monitoring (VRAM exporter + Prometheus + Grafana)
|
||||
# Gates on llm_monitoring_enabled (default true — but can be disabled per-host).
|
||||
# Deploys:
|
||||
# - VRAM textfile exporter script (runs every minute via cron)
|
||||
# - Prometheus scrape config template (for GitOps deployment)
|
||||
# - Grafana dashboard JSON template (for GitOps deployment)
|
||||
# - PrometheusRule alert rules template (for GitOps deployment)
|
||||
#
|
||||
# No cluster-facing changes here; templates are staged for manual review
|
||||
# and committed via Git. ArgoCD syncs them automatically afterward.
|
||||
#
|
||||
# Reference: roles/llm-inference-multimodel/references/monitoring-llm-homelab-ciro-luciotta-2026.md
|
||||
# Added 2026-08-18 (t_57a9f82f): GPU/LLM monitoring Phase 3 — Wong.
|
||||
- include_tasks: monitoring.yml
|
||||
when: llm_monitoring_enabled | default(true)
|
||||
tags: [always]
|
||||
|
||||
174
ansible/roles/llm-inference-multimodel/tasks/monitoring.yml
Normal file
174
ansible/roles/llm-inference-multimodel/tasks/monitoring.yml
Normal file
@@ -0,0 +1,174 @@
|
||||
---
|
||||
# ==============================================================================
|
||||
# FILE: roles/llm-inference-multimodel/tasks/monitoring.yml
|
||||
# DESCRIPTION: Phase X — GPU/LLM monitoring deployment for llama-swap.
|
||||
# Deploys:
|
||||
# 1. VRAM textfile exporter script + cron job
|
||||
# 2. Prometheus scrape config template (for GitOps deployment)
|
||||
# 3. Grafana dashboard JSON template (for GitOps deployment)
|
||||
# 4. PrometheusRule CR template (for GitOps deployment)
|
||||
#
|
||||
# REFERENCED BY: tasks/main.yml (call with `- include_tasks: monitoring.yml`)
|
||||
# GATED BY: llm_monitoring_enabled (default: true)
|
||||
#
|
||||
# AUTHOR: Wong (Infrastructure Automation Specialist)
|
||||
# DATE: 2026-08-18
|
||||
# ==============================================================================
|
||||
|
||||
- name: GPU/LLM Monitoring | Conditional gate
|
||||
debug:
|
||||
msg: "GPU/LLM monitoring deployment gated: llm_monitoring_enabled={{ llm_monitoring_enabled }}"
|
||||
when: not llm_monitoring_enabled
|
||||
|
||||
- name: GPU/LLM Monitoring | Create monitoring script directory
|
||||
ansible.builtin.file:
|
||||
path: /opt/llama-server-monitoring
|
||||
state: directory
|
||||
owner: "{{ llm_service_user }}"
|
||||
group: "{{ llm_service_user }}"
|
||||
mode: "0755"
|
||||
when: llm_monitoring_enabled
|
||||
|
||||
- name: GPU/LLM Monitoring | Deploy VRAM exporter script
|
||||
ansible.builtin.copy:
|
||||
src: nvidia-smi-vram-exporter.sh
|
||||
dest: "{{ llm_vram_exporter_script }}"
|
||||
owner: root
|
||||
group: root
|
||||
mode: "0755"
|
||||
when: llm_monitoring_enabled
|
||||
notify: restart vram exporter cron
|
||||
|
||||
- name: GPU/LLM Monitoring | Create cron job for VRAM exporter
|
||||
ansible.builtin.cron:
|
||||
name: "llama-swap GPU VRAM exporter"
|
||||
minute: "{{ llm_vram_exporter_cron_minute }}"
|
||||
hour: "*"
|
||||
day: "*"
|
||||
month: "*"
|
||||
weekday: "*"
|
||||
job: "{{ llm_vram_exporter_script }}"
|
||||
state: present
|
||||
when: llm_monitoring_enabled
|
||||
|
||||
- name: GPU/LLM Monitoring | Verify VRAM exporter textfile directory exists
|
||||
ansible.builtin.file:
|
||||
path: "{{ llm_vram_textfile_dir }}"
|
||||
state: directory
|
||||
owner: "{{ llm_service_user }}"
|
||||
group: "{{ llm_service_user }}"
|
||||
mode: "0755"
|
||||
when: llm_monitoring_enabled
|
||||
|
||||
- name: GPU/LLM Monitoring | Force initial VRAM exporter run
|
||||
ansible.builtin.shell:
|
||||
cmd: "{{ llm_vram_exporter_script }}"
|
||||
register: vram_exporter_run
|
||||
changed_when: false
|
||||
when: llm_monitoring_enabled
|
||||
|
||||
- name: GPU/LLM Monitoring | Verify VRAM exporter output
|
||||
ansible.builtin.stat:
|
||||
path: "{{ llm_vram_textfile_dir }}/nvidia.prom"
|
||||
register: vram_exporter_output
|
||||
retries: 5
|
||||
delay: 2
|
||||
until: vram_exporter_output.stat.exists
|
||||
when: llm_monitoring_enabled
|
||||
|
||||
- name: GPU/LLM Monitoring | Display VRAM exporter output
|
||||
ansible.builtin.debug:
|
||||
msg: "VRAM exporter metric created: {{ vram_exporter_output.stat.path }}"
|
||||
when:
|
||||
- llm_monitoring_enabled
|
||||
- vram_exporter_output.stat.exists
|
||||
|
||||
# -----------------------------------------------------------------------
|
||||
# Prometheus & Grafana templates (for GitOps deployment via ArgoCD)
|
||||
# -----------------------------------------------------------------------
|
||||
|
||||
- name: GPU/LLM Monitoring | Template Prometheus scrape config
|
||||
ansible.builtin.template:
|
||||
src: llama-swap-prometheus-scrape.yml.j2
|
||||
dest: /tmp/llama-swap-prometheus-scrape.yml
|
||||
owner: root
|
||||
group: root
|
||||
mode: "0644"
|
||||
when: llm_monitoring_enabled
|
||||
register: prometheus_scrape_config
|
||||
|
||||
- name: GPU/LLM Monitoring | Template Grafana dashboard JSON
|
||||
ansible.builtin.template:
|
||||
src: llama-swap-grafana-dashboard.json.j2
|
||||
dest: /tmp/llama-swap-grafana-dashboard.json
|
||||
owner: root
|
||||
group: root
|
||||
mode: "0644"
|
||||
when: llm_monitoring_enabled
|
||||
register: grafana_dashboard_config
|
||||
|
||||
- name: GPU/LLM Monitoring | Template PrometheusRule alert rules
|
||||
ansible.builtin.template:
|
||||
src: llama-swap-alerts.yml.j2
|
||||
dest: /tmp/llama-swap-alerts.yml
|
||||
owner: root
|
||||
group: root
|
||||
mode: "0644"
|
||||
when: llm_monitoring_enabled
|
||||
register: prometheus_alerts_config
|
||||
|
||||
- name: GPU/LLM Monitoring | Validate Prometheus alert rules (YAML syntax)
|
||||
ansible.builtin.debug:
|
||||
msg: "Alert rules template ready at {{ prometheus_alerts_config.dest }}"
|
||||
when:
|
||||
- llm_monitoring_enabled
|
||||
- prometheus_alerts_config is changed
|
||||
|
||||
- name: GPU/LLM Monitoring | Validate Grafana dashboard JSON (JSON syntax)
|
||||
ansible.builtin.debug:
|
||||
msg: "Grafana dashboard template ready at {{ grafana_dashboard_config.dest }}"
|
||||
when:
|
||||
- llm_monitoring_enabled
|
||||
- grafana_dashboard_config is changed
|
||||
|
||||
- name: GPU/LLM Monitoring | Summary
|
||||
ansible.builtin.debug:
|
||||
msg: |
|
||||
GPU/LLM Monitoring Deployment Summary
|
||||
======================================
|
||||
Status: {{ 'ENABLED' if llm_monitoring_enabled else 'DISABLED' }}
|
||||
|
||||
Deployed Components:
|
||||
1. VRAM exporter: {{ llm_vram_exporter_script }}
|
||||
- Cron: Every minute (*/1 * * * *)
|
||||
- Output: {{ llm_vram_textfile_dir }}/nvidia.prom
|
||||
- Status: ✓ Running
|
||||
|
||||
2. Prometheus scrape config: /tmp/llama-swap-prometheus-scrape.yml
|
||||
- Target: {{ llm_bind_address }}:{{ llm_swapmode_port }}/metrics
|
||||
- Interval: {{ llm_prometheus_scrape_interval }}
|
||||
- Status: ✓ Templated (ready for GitOps deployment)
|
||||
|
||||
3. Grafana dashboard: /tmp/llama-swap-grafana-dashboard.json
|
||||
- Title: {{ llm_grafana_dashboard_title }}
|
||||
- UID: {{ llm_grafana_dashboard_uid }}
|
||||
- Panels: 6 (VRAM, KV-cache, Latency, Queue, Throughput, Percentiles)
|
||||
- Status: ✓ Templated (ready for GitOps deployment)
|
||||
|
||||
4. PrometheusRule alerts: /tmp/llama-swap-alerts.yml
|
||||
- Critical: VRAM > {{ llm_vram_critical_mib }} MiB
|
||||
- Warning: KV-cache > {{ llm_kv_cache_spill_ratio | round(2) }}
|
||||
- Warning: Throughput < {{ llm_throughput_baseline_tokens_per_min }} tokens/min
|
||||
- Status: ✓ Templated (ready for GitOps deployment)
|
||||
|
||||
Next Steps:
|
||||
1. Copy dashboard JSON to cluster/applications/monitoring/dashboards.yaml
|
||||
2. Copy alert rules to cluster/applications/monitoring/rules/ (K8s manifest)
|
||||
3. Add Prometheus scrape config to cluster/applications/monitoring/values.yaml
|
||||
4. Commit to Git and push (ArgoCD syncs automatically)
|
||||
5. Verify metrics appear in Prometheus UI within 2 minutes
|
||||
|
||||
Documentation:
|
||||
- Pattern spec: references/monitoring-llm-homelab-ciro-luciotta-2026.md
|
||||
- Phase 3 results: references/llama-swap-phase3-cutover-results-2026-08-18.md
|
||||
when: llm_monitoring_enabled
|
||||
@@ -65,7 +65,7 @@
|
||||
ansible.builtin.debug:
|
||||
msg:
|
||||
- "Qwen (:{{ llm_qwen_port }}) serving: {{ llm_qwen_models.json.data | map(attribute='id') | list }}"
|
||||
- "Verified n_ctx (must be >= 64000, not just requested): {{ llm_qwen_models.json.data | map(attribute='meta') | map(attribute='n_ctx') | list }}"
|
||||
- "Verified n_ctx (must be >= 64000, not just requested): {{ llm_qwen_models.json.data | map(attribute='meta', default={}) | map(attribute='n_ctx', default=0) | list }}"
|
||||
when:
|
||||
- llm_qwen_service_enabled | default(false)
|
||||
- llm_qwen_models is defined
|
||||
|
||||
@@ -18,9 +18,11 @@ ExecStart={{ llm_binary_path }} \
|
||||
--models-max {{ llm_router_models_max }} \
|
||||
--host {{ llm_router_bind_address }} \
|
||||
--port {{ llm_router_port }} \
|
||||
{% if not (llm_router_preset_enabled | default(false)) %}
|
||||
--n-gpu-layers {{ llm_router_gpu_layers }} \
|
||||
--ctx-size {{ llm_router_ctx_size }} \
|
||||
--flash-attn {{ llm_router_flash_attn }} \
|
||||
{% endif %}
|
||||
--cache-type-k {{ llm_router_cache_type_k }} \
|
||||
--cache-type-v {{ llm_router_cache_type_v }} \
|
||||
--batch-size {{ llm_router_batch_size }} \
|
||||
@@ -28,7 +30,7 @@ ExecStart={{ llm_binary_path }} \
|
||||
--parallel {{ llm_router_parallel }} \
|
||||
--metrics
|
||||
|
||||
# ROUTER MODE NOTES (2026-08-12, t_0cca74a2 / updated t_9adf0889):
|
||||
# ROUTER MODE NOTES (2026-08-12, t_0cca74a2 / updated t_9adf0889 / updated t_72646029):
|
||||
# - NO -m/--model flag: this is what enables llama-server router/supervisor mode.
|
||||
# Without -m, llama-server discovers all .gguf files in --models-dir, or uses
|
||||
# the per-model definitions in a --models-preset INI file.
|
||||
@@ -38,12 +40,16 @@ ExecStart={{ llm_binary_path }} \
|
||||
# The preset INI is at {{ llm_router_preset_path | default('/opt/llama-server-router-preset.ini') }}.
|
||||
# Both the section name and the alias field in the INI work as model IDs.
|
||||
# GH #22364 (extra "default" entry in /v1/models) is expected in preset mode — cosmetic.
|
||||
# - --n-gpu-layers is INTENTIONALLY OMITTED from preset mode (t_72646029, 2026-08-17):
|
||||
# In --models-preset mode every model section in the INI sets n-gpu-layers explicitly.
|
||||
# A global CLI --n-gpu-layers has HIGHEST precedence in llama.cpp (CLI > model-section > global-INI)
|
||||
# and would override per-model INI values (e.g. n-gpu-layers=0 for CPU offload).
|
||||
# When preset mode is disabled (--models-dir), --n-gpu-layers is emitted normally.
|
||||
# - --models-max {{ llm_router_models_max }} is driven by llm_router_models_max
|
||||
# (default 1 in defaults/main.yml; overridden to 4 in host_vars/astro-orbiter
|
||||
# as of t_33acbb2e after VRAM budget review — see host_vars for OOM risk note).
|
||||
# Default llama-server cap is 4 simultaneous — OOM on 24GB if all 3 current
|
||||
# GGUFs load at once. LRU eviction mitigates in practice but review before adding
|
||||
# models. See host_vars/astro-orbiter/vars.yml for full VRAM breakdown.
|
||||
# as of t_72646029 after CPU-offload enabling — CPU models count against models-max
|
||||
# and hold ~1.4-1.7GB CUDA-context VRAM each (llama.cpp 6ea215d allocates it even at
|
||||
# n-gpu-layers=0); steady-state ~24,004 MiB, below the 24,576 MiB physical limit).
|
||||
# - Clients select a model via "model": "<section-name-or-alias>" in their
|
||||
# chat completion request. Hermes sends model: "<id>" on every request already.
|
||||
# - Cold model load on first request: ~30-60s for Qwen3.6-35B. First response
|
||||
|
||||
@@ -0,0 +1,132 @@
|
||||
# ==============================================================================
|
||||
# FILE: roles/llm-inference-multimodel/templates/llama-swap-alerts.yml.j2
|
||||
# DESCRIPTION: PrometheusRule CustomResource for llama-swap alert rules.
|
||||
# Defines CRITICAL, WARNING, and INFO alerts per the Ciro Luciotta
|
||||
# monitoring pattern (references/monitoring-llm-homelab-ciro-luciotta-2026.md).
|
||||
#
|
||||
# Deployed by ArgoCD as a K8s resource in the monitoring namespace.
|
||||
# Prometheus loads these rules automatically on sync.
|
||||
#
|
||||
# SCOPE: Alerts fire when:
|
||||
# - VRAM exceeds physical limit (24GB) — pending OOM-kill
|
||||
# - KV-cache spills to CPU (>92% utilization) — requests may drop
|
||||
# - Throughput degrades below baseline — model may be throttled
|
||||
#
|
||||
# AUTHOR: Wong (Infrastructure Automation Specialist)
|
||||
# DATE: 2026-08-18
|
||||
# ==============================================================================
|
||||
|
||||
apiVersion: monitoring.coreos.com/v1
|
||||
kind: PrometheusRule
|
||||
metadata:
|
||||
name: llama-swap-alerts
|
||||
namespace: monitoring
|
||||
labels:
|
||||
prometheus: kube-prometheus
|
||||
spec:
|
||||
groups:
|
||||
- name: llama-swap.rules
|
||||
interval: 30s
|
||||
rules:
|
||||
|
||||
# ====================================================================
|
||||
# CRITICAL: GPU VRAM saturation (OOM risk)
|
||||
# ====================================================================
|
||||
- alert: LlamaSwapVramSaturation
|
||||
expr: (llamaswap_gpu_memory_used_bytes{job=\"llama-swap\"} / 1048576) > {{ llm_swapmode_vram_max_mib | int }}
|
||||
for: 1m
|
||||
labels:
|
||||
severity: critical
|
||||
component: llm-inference
|
||||
annotations:
|
||||
summary: "GPU VRAM saturation on {{ $labels.instance }}"
|
||||
description: |
|
||||
GPU VRAM usage is {{ $value | humanize }}MiB (critical threshold: {{ llm_swapmode_vram_max_mib }}MiB).
|
||||
|
||||
The system is at risk of out-of-memory (OOM) kernel-kill events.
|
||||
Immediate action required:
|
||||
1. Check Prometheus dashboard for request queue depth and active models
|
||||
2. Identify which model(s) are consuming VRAM
|
||||
3. If queue depth is high, consider rate-limiting or routing requests
|
||||
4. If a single request caused the spike, investigate context-window size
|
||||
|
||||
Instance: {{ $labels.instance }}
|
||||
Time: {{ $value | humanizeDuration }}
|
||||
|
||||
# ====================================================================
|
||||
# WARNING: KV-cache spill risk (context cache pressure)
|
||||
# ====================================================================
|
||||
- alert: LlamaSwapVramPressure
|
||||
expr: llamaswap_gpu_memory_util_percent{job="llama-swap"} > 92
|
||||
for: 2m
|
||||
labels:
|
||||
severity: warning
|
||||
component: llm-inference
|
||||
annotations:
|
||||
summary: "GPU memory utilization high (possible VRAM pressure)"
|
||||
description: |
|
||||
KV-cache utilization on {{ $labels.model }} is {{ $value | humanizePercentage }}
|
||||
(warning threshold: 92%).
|
||||
|
||||
The model's context cache is nearly full. Requests with large context windows
|
||||
may not fit and could be dropped from the queue. Consider:
|
||||
1. Reviewing incoming request context-window distribution
|
||||
2. Reducing n_ctx for non-critical models (if router mode is active)
|
||||
3. Routing long-context requests to a different model with more capacity
|
||||
4. Investigating whether concurrent requests are competing for KV space
|
||||
|
||||
Model: {{ $labels.model }}
|
||||
Instance: {{ $labels.instance }}
|
||||
|
||||
# ====================================================================
|
||||
# WARNING: Throughput degradation (possible throttling)
|
||||
# ====================================================================
|
||||
- alert: LlamaSwapInferenceStall
|
||||
expr: |
|
||||
(llamaswap_gpu_util_percent{job="llama-swap"} == 0) and (llamaswap_gpu_memory_util_percent{job="llama-swap"} > 50)
|
||||
for: 5m
|
||||
labels:
|
||||
severity: warning
|
||||
component: llm-inference
|
||||
annotations:
|
||||
summary: "GPU compute stall detected (memory loaded but no utilization)"
|
||||
description: |
|
||||
The RTX 3090 has >50% memory utilization but 0% compute utilization
|
||||
for more than 5 minutes. This may indicate:
|
||||
|
||||
This may indicate:
|
||||
1. Thermal throttling (GPU temperature limiting frequency)
|
||||
2. Memory pressure (even if VRAM not full, latency can increase)
|
||||
3. CPU contention (if models are CPU-offloaded)
|
||||
4. Incoming request rate exceeds model capacity (check queue depth)
|
||||
|
||||
Recommended actions:
|
||||
- Check nvidia-smi output for GPU temperature and throttle flags
|
||||
- Compare queue depth to baseline (alert if >5 sustained)
|
||||
- Check CPU usage and interrupt frequency (vmstat 1 1)
|
||||
- Review log tail for errors or warnings from llama-swap
|
||||
|
||||
Model: {{ $labels.model }}
|
||||
Instance: {{ $labels.instance }}
|
||||
|
||||
# ====================================================================
|
||||
# INFO: Scrape failures (monitoring health)
|
||||
# ====================================================================
|
||||
- alert: LlamaSwapScrapeFailed
|
||||
expr: up{job="llama-swap"} == 0
|
||||
for: 2m
|
||||
labels:
|
||||
severity: warning
|
||||
component: monitoring
|
||||
annotations:
|
||||
summary: "llama-swap Prometheus scrape failed"
|
||||
description: |
|
||||
Prometheus cannot scrape llama-swap's /metrics endpoint at
|
||||
http://{{ $labels.instance }}/metrics (HTTP {{ $value }} or timeout).
|
||||
|
||||
The monitoring pipeline is degraded. Check:
|
||||
1. llama-swap service status: systemctl status llama-swap
|
||||
2. Network reachability: curl http://{{ $labels.instance }}/metrics
|
||||
3. Prometheus scrape logs in Prometheus UI (Alerts -> llama-swap)
|
||||
|
||||
Instance: {{ $labels.instance }}
|
||||
@@ -0,0 +1,534 @@
|
||||
{
|
||||
"annotations": {
|
||||
"list": [
|
||||
{
|
||||
"builtIn": 1,
|
||||
"datasource": "-- Grafana --",
|
||||
"enable": true,
|
||||
"hide": true,
|
||||
"iconColor": "rgba(0, 211, 255, 1)",
|
||||
"name": "Annotations & Alerts",
|
||||
"type": "dashboard"
|
||||
}
|
||||
]
|
||||
},
|
||||
"editable": true,
|
||||
"gnetId": null,
|
||||
"graphTooltip": 0,
|
||||
"id": null,
|
||||
"links": [],
|
||||
"panels": [
|
||||
{
|
||||
"datasource": "Prometheus",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"color": {
|
||||
"mode": "palette-classic"
|
||||
},
|
||||
"custom": {
|
||||
"axisLabel": "MiB",
|
||||
"axisPlacement": "auto",
|
||||
"barAlignment": 0,
|
||||
"drawStyle": "line",
|
||||
"fillOpacity": 10,
|
||||
"gradientMode": "none",
|
||||
"hideFrom": {
|
||||
"tooltip": false,
|
||||
"viz": false,
|
||||
"legend": false
|
||||
},
|
||||
"lineInterpolation": "linear",
|
||||
"lineWidth": 1,
|
||||
"pointSize": 5,
|
||||
"scaleDistribution": {
|
||||
"type": "linear"
|
||||
},
|
||||
"showPoints": "auto",
|
||||
"spanNulls": false,
|
||||
"stacking": {
|
||||
"group": "A",
|
||||
"mode": "none"
|
||||
},
|
||||
"thresholdsStyle": {
|
||||
"mode": "off"
|
||||
}
|
||||
},
|
||||
"mappings": [],
|
||||
"max": 24576,
|
||||
"min": 0,
|
||||
"thresholds": {
|
||||
"mode": "absolute",
|
||||
"steps": [
|
||||
{
|
||||
"color": "green",
|
||||
"value": null
|
||||
},
|
||||
{
|
||||
"color": "yellow",
|
||||
"value": 23000
|
||||
},
|
||||
{
|
||||
"color": "red",
|
||||
"value": 24000
|
||||
}
|
||||
]
|
||||
},
|
||||
"unit": "short"
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"gridPos": {
|
||||
"h": 8,
|
||||
"w": 12,
|
||||
"x": 0,
|
||||
"y": 0
|
||||
},
|
||||
"id": 1,
|
||||
"options": {
|
||||
"legend": {
|
||||
"calcs": [
|
||||
"last",
|
||||
"max"
|
||||
],
|
||||
"displayMode": "table",
|
||||
"placement": "right"
|
||||
},
|
||||
"tooltip": {
|
||||
"mode": "single"
|
||||
}
|
||||
},
|
||||
"pluginVersion": "8.0.0",
|
||||
"targets": [
|
||||
{
|
||||
"expr": "llamaswap_gpu_memory_used_bytes{job=\"llama-swap\"} / 1048576",
|
||||
"interval": "",
|
||||
"legendFormat": "VRAM Used",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
"title": "GPU VRAM Usage (MiB)",
|
||||
"type": "timeseries"
|
||||
},
|
||||
{
|
||||
"datasource": "Prometheus",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"color": {
|
||||
"mode": "thresholds"
|
||||
},
|
||||
"mappings": [],
|
||||
"max": 1,
|
||||
"min": 0,
|
||||
"thresholds": {
|
||||
"mode": "absolute",
|
||||
"steps": [
|
||||
{
|
||||
"color": "green",
|
||||
"value": null
|
||||
},
|
||||
{
|
||||
"color": "yellow",
|
||||
"value": 0.8
|
||||
},
|
||||
{
|
||||
"color": "orange",
|
||||
"value": 0.92
|
||||
},
|
||||
{
|
||||
"color": "red",
|
||||
"value": 0.95
|
||||
}
|
||||
]
|
||||
},
|
||||
"unit": "percentunit"
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"gridPos": {
|
||||
"h": 8,
|
||||
"w": 12,
|
||||
"x": 12,
|
||||
"y": 0
|
||||
},
|
||||
"id": 2,
|
||||
"options": {
|
||||
"orientation": "auto",
|
||||
"reduceOptions": {
|
||||
"values": false,
|
||||
"fields": "",
|
||||
"calcs": [
|
||||
"lastNotNull"
|
||||
]
|
||||
},
|
||||
"showThresholdLabels": false,
|
||||
"showThresholdMarkers": true
|
||||
},
|
||||
"pluginVersion": "8.0.0",
|
||||
"targets": [
|
||||
{
|
||||
"expr": "llamaswap_gpu_memory_util_percent{job=\"llama-swap\"}",
|
||||
"interval": "",
|
||||
"legendFormat": "{{ model }}",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
"title": "GPU Memory Utilization %",
|
||||
"type": "gauge"
|
||||
},
|
||||
{
|
||||
"datasource": "Prometheus",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"color": {
|
||||
"mode": "palette-classic"
|
||||
},
|
||||
"custom": {
|
||||
"axisLabel": "%",
|
||||
"axisPlacement": "auto",
|
||||
"barAlignment": 0,
|
||||
"drawStyle": "line",
|
||||
"fillOpacity": 0,
|
||||
"gradientMode": "none",
|
||||
"hideFrom": {
|
||||
"tooltip": false,
|
||||
"viz": false,
|
||||
"legend": false
|
||||
},
|
||||
"lineInterpolation": "linear",
|
||||
"lineWidth": 1,
|
||||
"pointSize": 5,
|
||||
"scaleDistribution": {
|
||||
"type": "linear"
|
||||
},
|
||||
"showPoints": "never",
|
||||
"spanNulls": true,
|
||||
"stacking": {
|
||||
"group": "A",
|
||||
"mode": "none"
|
||||
},
|
||||
"thresholdsStyle": {
|
||||
"mode": "off"
|
||||
}
|
||||
},
|
||||
"mappings": [],
|
||||
"thresholds": {
|
||||
"mode": "absolute",
|
||||
"steps": [
|
||||
{
|
||||
"color": "green",
|
||||
"value": null
|
||||
}
|
||||
]
|
||||
},
|
||||
"unit": "percent"
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"gridPos": {
|
||||
"h": 8,
|
||||
"w": 12,
|
||||
"x": 0,
|
||||
"y": 8
|
||||
},
|
||||
"id": 3,
|
||||
"options": {
|
||||
"legend": {
|
||||
"calcs": [
|
||||
"mean",
|
||||
"max"
|
||||
],
|
||||
"displayMode": "table",
|
||||
"placement": "right"
|
||||
},
|
||||
"tooltip": {
|
||||
"mode": "single"
|
||||
}
|
||||
},
|
||||
"pluginVersion": "8.0.0",
|
||||
"targets": [
|
||||
{
|
||||
"expr": "llamaswap_gpu_util_percent{job=\"llama-swap\"}",
|
||||
"interval": "",
|
||||
"legendFormat": "{{ model }}",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
"title": "GPU Utilization %",
|
||||
"type": "timeseries"
|
||||
},
|
||||
{
|
||||
"datasource": "Prometheus",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"color": {
|
||||
"mode": "palette-classic"
|
||||
},
|
||||
"custom": {
|
||||
"axisLabel": "%",
|
||||
"axisPlacement": "auto",
|
||||
"barAlignment": 0,
|
||||
"drawStyle": "line",
|
||||
"fillOpacity": 0,
|
||||
"gradientMode": "none",
|
||||
"hideFrom": {
|
||||
"tooltip": false,
|
||||
"viz": false,
|
||||
"legend": false
|
||||
},
|
||||
"lineInterpolation": "linear",
|
||||
"lineWidth": 1,
|
||||
"pointSize": 5,
|
||||
"scaleDistribution": {
|
||||
"type": "linear"
|
||||
},
|
||||
"showPoints": "never",
|
||||
"spanNulls": true,
|
||||
"stacking": {
|
||||
"group": "A",
|
||||
"mode": "none"
|
||||
},
|
||||
"thresholdsStyle": {
|
||||
"mode": "off"
|
||||
}
|
||||
},
|
||||
"mappings": [],
|
||||
"thresholds": {
|
||||
"mode": "absolute",
|
||||
"steps": [
|
||||
{
|
||||
"color": "green",
|
||||
"value": null
|
||||
},
|
||||
{
|
||||
"color": "yellow",
|
||||
"value": 3
|
||||
},
|
||||
{
|
||||
"color": "red",
|
||||
"value": 5
|
||||
}
|
||||
]
|
||||
},
|
||||
"unit": "percent"
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"gridPos": {
|
||||
"h": 8,
|
||||
"w": 12,
|
||||
"x": 12,
|
||||
"y": 8
|
||||
},
|
||||
"id": 4,
|
||||
"options": {
|
||||
"legend": {
|
||||
"calcs": [
|
||||
"mean",
|
||||
"max"
|
||||
],
|
||||
"displayMode": "table",
|
||||
"placement": "right"
|
||||
},
|
||||
"tooltip": {
|
||||
"mode": "single"
|
||||
}
|
||||
},
|
||||
"pluginVersion": "8.0.0",
|
||||
"targets": [
|
||||
{
|
||||
"expr": "avg(llamaswap_cpu_util_percent{job=\"llama-swap\"})",
|
||||
"interval": "",
|
||||
"legendFormat": "{{ model }}",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
"title": "CPU Utilization %",
|
||||
"type": "timeseries"
|
||||
},
|
||||
{
|
||||
"datasource": "Prometheus",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"color": {
|
||||
"mode": "palette-classic"
|
||||
},
|
||||
"custom": {
|
||||
"axisLabel": "W",
|
||||
"axisPlacement": "auto",
|
||||
"barAlignment": 0,
|
||||
"drawStyle": "line",
|
||||
"fillOpacity": 0,
|
||||
"gradientMode": "none",
|
||||
"hideFrom": {
|
||||
"tooltip": false,
|
||||
"viz": false,
|
||||
"legend": false
|
||||
},
|
||||
"lineInterpolation": "linear",
|
||||
"lineWidth": 1,
|
||||
"pointSize": 5,
|
||||
"scaleDistribution": {
|
||||
"type": "linear"
|
||||
},
|
||||
"showPoints": "never",
|
||||
"spanNulls": true,
|
||||
"stacking": {
|
||||
"group": "A",
|
||||
"mode": "none"
|
||||
},
|
||||
"thresholdsStyle": {
|
||||
"mode": "off"
|
||||
}
|
||||
},
|
||||
"mappings": [],
|
||||
"thresholds": {
|
||||
"mode": "absolute",
|
||||
"steps": [
|
||||
{
|
||||
"color": "green",
|
||||
"value": null
|
||||
}
|
||||
]
|
||||
},
|
||||
"unit": "watt"
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"gridPos": {
|
||||
"h": 8,
|
||||
"w": 12,
|
||||
"x": 0,
|
||||
"y": 16
|
||||
},
|
||||
"id": 5,
|
||||
"options": {
|
||||
"legend": {
|
||||
"calcs": [
|
||||
"mean"
|
||||
],
|
||||
"displayMode": "table",
|
||||
"placement": "right"
|
||||
},
|
||||
"tooltip": {
|
||||
"mode": "single"
|
||||
}
|
||||
},
|
||||
"pluginVersion": "8.0.0",
|
||||
"targets": [
|
||||
{
|
||||
"expr": "llamaswap_gpu_power_draw_watts{job=\"llama-swap\"}",
|
||||
"interval": "",
|
||||
"legendFormat": "{{ model }} (tokens/min)",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
"title": "GPU Power Draw (W)",
|
||||
"type": "timeseries"
|
||||
},
|
||||
{
|
||||
"datasource": "Prometheus",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"color": {
|
||||
"mode": "palette-classic"
|
||||
},
|
||||
"custom": {
|
||||
"axisLabel": "load",
|
||||
"axisPlacement": "auto",
|
||||
"barAlignment": 0,
|
||||
"drawStyle": "bars",
|
||||
"fillOpacity": 100,
|
||||
"gradientMode": "none",
|
||||
"hideFrom": {
|
||||
"tooltip": false,
|
||||
"viz": false,
|
||||
"legend": false
|
||||
},
|
||||
"lineInterpolation": "linear",
|
||||
"lineWidth": 1,
|
||||
"pointSize": 5,
|
||||
"scaleDistribution": {
|
||||
"type": "linear"
|
||||
},
|
||||
"showPoints": "never",
|
||||
"spanNulls": true,
|
||||
"stacking": {
|
||||
"group": "A",
|
||||
"mode": "normal"
|
||||
},
|
||||
"thresholdsStyle": {
|
||||
"mode": "off"
|
||||
}
|
||||
},
|
||||
"mappings": [],
|
||||
"thresholds": {
|
||||
"mode": "absolute",
|
||||
"steps": [
|
||||
{
|
||||
"color": "green",
|
||||
"value": null
|
||||
}
|
||||
]
|
||||
},
|
||||
"unit": "short"
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"gridPos": {
|
||||
"h": 8,
|
||||
"w": 12,
|
||||
"x": 12,
|
||||
"y": 16
|
||||
},
|
||||
"id": 6,
|
||||
"options": {
|
||||
"legend": {
|
||||
"calcs": [],
|
||||
"displayMode": "list",
|
||||
"placement": "bottom"
|
||||
},
|
||||
"tooltip": {
|
||||
"mode": "single"
|
||||
}
|
||||
},
|
||||
"pluginVersion": "8.0.0",
|
||||
"targets": [
|
||||
{
|
||||
"expr": "llamaswap_load_average{interval=\"5m\"}",
|
||||
"interval": "",
|
||||
"legendFormat": "p95 latency",
|
||||
"refId": "A"
|
||||
},
|
||||
{
|
||||
"expr": "llamaswap_load_average{interval=\"5m\"}",
|
||||
"interval": "",
|
||||
"legendFormat": "p99 latency",
|
||||
"refId": "B"
|
||||
}
|
||||
],
|
||||
"title": "System Load Average (5m)",
|
||||
"type": "timeseries"
|
||||
}
|
||||
],
|
||||
"refresh": "30s",
|
||||
"schemaVersion": 27,
|
||||
"style": "dark",
|
||||
"tags": [
|
||||
"llm",
|
||||
"llama-swap",
|
||||
"gpu-monitoring",
|
||||
"ciro-luciotta"
|
||||
],
|
||||
"templating": {
|
||||
"list": []
|
||||
},
|
||||
"time": {
|
||||
"from": "now-24h",
|
||||
"to": "now"
|
||||
},
|
||||
"timepicker": {},
|
||||
"timezone": "",
|
||||
"title": {{ llm_grafana_dashboard_title }},
|
||||
"uid": {{ llm_grafana_dashboard_uid }},
|
||||
"version": 1
|
||||
}
|
||||
@@ -0,0 +1,39 @@
|
||||
# ==============================================================================
|
||||
# FILE: roles/llm-inference-multimodel/templates/llama-swap-prometheus-scrape.yml.j2
|
||||
# DESCRIPTION: Prometheus scrape job configuration for llama-swap's native
|
||||
# /metrics endpoint (OpenMetrics format).
|
||||
#
|
||||
# This template is rendered and deployed to the Prometheus
|
||||
# config via GitOps (cluster/applications/monitoring/values.yaml).
|
||||
# Does NOT include this file inline here; it is referenced and
|
||||
# rendered by Ansible roles/llm-inference-multimodel/tasks/*.yml.
|
||||
#
|
||||
# TARGET HOST: astro-orbiter ({{ llm_bind_address }}:{{ llm_swapmode_port }})
|
||||
# METRICS: llamacpp_tokens_predicted_total, llamacpp_kv_cache_usage_ratio,
|
||||
# llamacpp_time_predict_ms, llamacpp_queue_size, etc. (per llama.cpp)
|
||||
#
|
||||
# AUTHOR: Wong (Infrastructure Automation Specialist)
|
||||
# DATE: 2026-08-18
|
||||
# ==============================================================================
|
||||
|
||||
---
|
||||
- job_name: llama-swap
|
||||
static_configs:
|
||||
- targets: ["{{ llm_bind_address }}:{{ llm_swapmode_port }}"]
|
||||
labels:
|
||||
component: llm-inference
|
||||
service: llama-swap
|
||||
environment: homelab
|
||||
scrape_interval: 30s
|
||||
scrape_timeout: 10s
|
||||
honor_labels: true
|
||||
metrics_path: /metrics
|
||||
|
||||
# Relabeling: extract model name from metric labels for dashboard grouping
|
||||
# llama-swap exposes llamaswap_* metrics (GPU VRAM, utilization, power, CPU,
|
||||
# network, load average). Per-model inference metrics are not available at the
|
||||
# proxy level. Filter to keep only llamaswap_* metrics to reduce cardinality.
|
||||
metric_relabel_configs:
|
||||
- source_labels: [__name__]
|
||||
regex: 'llamaswap_.*'
|
||||
action: keep
|
||||
@@ -0,0 +1,53 @@
|
||||
{#
|
||||
FILE: roles/llm-inference-multimodel/templates/llama-swap.service.j2
|
||||
DESCRIPTION: llama-swap systemd unit template.
|
||||
Single Go binary, no subprocess management — just a /usr/local/bin/llama-swap
|
||||
process reading /etc/llama-swap/config.yaml.
|
||||
|
||||
Design:
|
||||
- Type=simple (no forking)
|
||||
- User={{ llm_swapmode_service_user }} (jarvis)
|
||||
- Restart=on-failure, RestartSec=10
|
||||
- Logs to journald (StandardOutput/StandardError=journal)
|
||||
- After nvidia-persistenced.service (NVIDIA driver dependency)
|
||||
|
||||
Config location: /etc/llama-swap/config.yaml (rendered by swapmode_config phase)
|
||||
Listen address: 127.0.0.1 inside the container (exposed by --listen flag)
|
||||
#}
|
||||
[Unit]
|
||||
Description=llama-swap — hot-swap model proxy (port {{ llm_swapmode_port }})
|
||||
Documentation=https://github.com/mostlygeek/llama-swap
|
||||
After=network.target nvidia-persistenced.service
|
||||
Wants=nvidia-persistenced.service
|
||||
|
||||
[Service]
|
||||
Type=simple
|
||||
User={{ llm_swapmode_service_user }}
|
||||
Group={{ llm_swapmode_service_user }}
|
||||
Environment="HOME=/home/{{ llm_swapmode_service_user }}"
|
||||
|
||||
ExecStart=/usr/local/bin/llama-swap \
|
||||
--config {{ llm_swapmode_config_file }} \
|
||||
--listen {{ llm_swapmode_bind_address }}:{{ llm_swapmode_port }}
|
||||
|
||||
# LLAMA-SWAP NOTES (2026-08-18, t_c1e44190):
|
||||
# - Single Go binary, zero runtime dependencies (llama.cpp statically linked).
|
||||
# - Upstream servers (llama-server instances) are spawned on-demand per config.yaml model definitions.
|
||||
# - --listen can override config.yaml's listen key; this flag takes precedence.
|
||||
# Double-check consistency between ExecStart and config.yaml.
|
||||
# - CUDA_VISIBLE_DEVICES can be set via Environment= if GPU isolation is needed.
|
||||
# Default: inherit from parent (systemd likely has it unset, picks all GPUs).
|
||||
# - No jinja flag needed: llama.cpp model templates are embedded in each model's GGUF.
|
||||
|
||||
Restart=on-failure
|
||||
RestartSec=10
|
||||
TimeoutStartSec=600
|
||||
StandardOutput=journal
|
||||
StandardError=journal
|
||||
SyslogIdentifier=llama-swap
|
||||
|
||||
# Resource limits (optional; adjust per VRAM budget)
|
||||
# MemoryMax=24G # Enforce hard limit; uncomment if runaway is a concern
|
||||
|
||||
[Install]
|
||||
WantedBy=multi-user.target
|
||||
248
ansible/roles/llm-inference-multimodel/verify-monitoring-deployment.sh
Executable file
248
ansible/roles/llm-inference-multimodel/verify-monitoring-deployment.sh
Executable file
@@ -0,0 +1,248 @@
|
||||
#!/bin/bash
|
||||
# ==============================================================================
|
||||
# VERIFICATION SCRIPT: GPU/LLM Monitoring Deployment (Task t_57a9f82f)
|
||||
# ==============================================================================
|
||||
# Run this script AFTER Ansible role deployment to verify all monitoring
|
||||
# components are installed and functional.
|
||||
#
|
||||
# Usage:
|
||||
# bash verify-monitoring-deployment.sh
|
||||
#
|
||||
# Expected output: All checks ✓ (green)
|
||||
# ==============================================================================
|
||||
|
||||
set -euo pipefail
|
||||
|
||||
ROLE_DIR="/home/hermes/git/homelab/ansible/roles/llm-inference-multimodel"
|
||||
VRAM_EXPORTER_SCRIPT="/opt/llama-server-monitoring/nvidia-smi-vram-exporter.sh"
|
||||
VRAM_EXPORTER_OUTPUT="/var/lib/node_exporter/textfile_collector/nvidia.prom"
|
||||
|
||||
CHECKS_PASSED=0
|
||||
CHECKS_FAILED=0
|
||||
|
||||
# Colors for output
|
||||
RED='\033[0;31m'
|
||||
GREEN='\033[0;32m'
|
||||
YELLOW='\033[1;33m'
|
||||
NC='\033[0m' # No Color
|
||||
|
||||
# Helper function for check results
|
||||
check_pass() {
|
||||
local desc="$1"
|
||||
echo -e "${GREEN}✓${NC} $desc"
|
||||
((CHECKS_PASSED++))
|
||||
}
|
||||
|
||||
check_fail() {
|
||||
local desc="$1"
|
||||
local reason="${2:-Unknown reason}"
|
||||
echo -e "${RED}✗${NC} $desc"
|
||||
echo " Reason: $reason"
|
||||
((CHECKS_FAILED++))
|
||||
}
|
||||
|
||||
echo "================================================================================"
|
||||
echo "GPU/LLM Monitoring Deployment Verification"
|
||||
echo "================================================================================"
|
||||
echo ""
|
||||
|
||||
# 1. Check role structure
|
||||
echo "1. Role Structure & Deliverables"
|
||||
echo "=================================="
|
||||
|
||||
if [ -f "$ROLE_DIR/references/monitoring-llm-homelab-ciro-luciotta-2026.md" ]; then
|
||||
check_pass "Reference docs: monitoring-llm-homelab-ciro-luciotta-2026.md exists"
|
||||
else
|
||||
check_fail "Reference docs: monitoring-llm-homelab-ciro-luciotta-2026.md NOT FOUND"
|
||||
fi
|
||||
|
||||
if [ -f "$ROLE_DIR/references/llama-swap-phase3-cutover-results-2026-08-18.md" ]; then
|
||||
check_pass "Phase 3 results: llama-swap-phase3-cutover-results-2026-08-18.md exists"
|
||||
else
|
||||
check_fail "Phase 3 results: llama-swap-phase3-cutover-results-2026-08-18.md NOT FOUND"
|
||||
fi
|
||||
|
||||
if [ -f "$ROLE_DIR/scripts/nvidia-smi-vram-exporter.sh" ]; then
|
||||
check_pass "VRAM exporter script: nvidia-smi-vram-exporter.sh exists"
|
||||
else
|
||||
check_fail "VRAM exporter script: nvidia-smi-vram-exporter.sh NOT FOUND"
|
||||
fi
|
||||
|
||||
if [ -x "$ROLE_DIR/scripts/nvidia-smi-vram-exporter.sh" ]; then
|
||||
check_pass "VRAM exporter script: executable"
|
||||
else
|
||||
check_fail "VRAM exporter script: not executable"
|
||||
fi
|
||||
|
||||
if [ -f "$ROLE_DIR/templates/llama-swap-prometheus-scrape.yml.j2" ]; then
|
||||
check_pass "Prometheus scrape config template exists"
|
||||
else
|
||||
check_fail "Prometheus scrape config template NOT FOUND"
|
||||
fi
|
||||
|
||||
if [ -f "$ROLE_DIR/templates/llama-swap-grafana-dashboard.json.j2" ]; then
|
||||
check_pass "Grafana dashboard template exists"
|
||||
else
|
||||
check_fail "Grafana dashboard template NOT FOUND"
|
||||
fi
|
||||
|
||||
if [ -f "$ROLE_DIR/templates/llama-swap-alerts.yml.j2" ]; then
|
||||
check_pass "Alert rules template exists"
|
||||
else
|
||||
check_fail "Alert rules template NOT FOUND"
|
||||
fi
|
||||
|
||||
if [ -f "$ROLE_DIR/tasks/monitoring.yml" ]; then
|
||||
check_pass "Monitoring tasks file exists"
|
||||
else
|
||||
check_fail "Monitoring tasks file NOT FOUND"
|
||||
fi
|
||||
|
||||
echo ""
|
||||
|
||||
# 2. Check runtime deployment (if on astro-orbiter)
|
||||
echo "2. Runtime Deployment Status (astro-orbiter)"
|
||||
echo "=============================================="
|
||||
|
||||
if [ -x "$VRAM_EXPORTER_SCRIPT" ]; then
|
||||
check_pass "VRAM exporter script deployed at $VRAM_EXPORTER_SCRIPT"
|
||||
|
||||
# Try to run it
|
||||
if output=$($VRAM_EXPORTER_SCRIPT 2>&1) && [ -f "$VRAM_EXPORTER_OUTPUT" ]; then
|
||||
check_pass "VRAM exporter runs successfully"
|
||||
|
||||
# Check metric format
|
||||
if grep -q "llamacpp_vram_used_mib" "$VRAM_EXPORTER_OUTPUT"; then
|
||||
check_pass "VRAM metric format is correct"
|
||||
|
||||
# Extract and display the value
|
||||
vram_value=$(grep "llamacpp_vram_used_mib " "$VRAM_EXPORTER_OUTPUT" | awk '{print $NF}')
|
||||
echo " Current VRAM usage: ${vram_value} MiB"
|
||||
else
|
||||
check_fail "VRAM metric format incorrect" "Expected 'llamacpp_vram_used_mib' in output"
|
||||
fi
|
||||
else
|
||||
check_fail "VRAM exporter failed to run" "$output"
|
||||
fi
|
||||
else
|
||||
echo -e "${YELLOW}⊘${NC} VRAM exporter not deployed yet (expected if running on non-astro-orbiter)"
|
||||
fi
|
||||
|
||||
if crontab -l 2>/dev/null | grep -q "nvidia-smi-vram-exporter"; then
|
||||
check_pass "VRAM exporter cron job is installed"
|
||||
else
|
||||
echo -e "${YELLOW}⊘${NC} VRAM exporter cron job not installed (expected if not on astro-orbiter)"
|
||||
fi
|
||||
|
||||
echo ""
|
||||
|
||||
# 3. Check Ansible variables
|
||||
echo "3. Ansible Configuration Variables"
|
||||
echo "===================================="
|
||||
|
||||
if grep -q "llm_monitoring_enabled" "$ROLE_DIR/defaults/main.yml"; then
|
||||
check_pass "llm_monitoring_enabled variable defined"
|
||||
else
|
||||
check_fail "llm_monitoring_enabled variable NOT FOUND"
|
||||
fi
|
||||
|
||||
if grep -q "llm_vram_critical_mib" "$ROLE_DIR/defaults/main.yml"; then
|
||||
check_pass "Alert threshold variables defined"
|
||||
else
|
||||
check_fail "Alert threshold variables NOT FOUND"
|
||||
fi
|
||||
|
||||
if grep -q "llm_grafana_dashboard_uid" "$ROLE_DIR/defaults/main.yml"; then
|
||||
check_pass "Grafana dashboard variables defined"
|
||||
else
|
||||
check_fail "Grafana dashboard variables NOT FOUND"
|
||||
fi
|
||||
|
||||
echo ""
|
||||
|
||||
# 4. Syntax validation
|
||||
echo "4. Template & Configuration Syntax"
|
||||
echo "===================================="
|
||||
|
||||
# Validate shell script
|
||||
if bash -n "$ROLE_DIR/scripts/nvidia-smi-vram-exporter.sh" 2>/dev/null; then
|
||||
check_pass "VRAM exporter script syntax (bash)"
|
||||
else
|
||||
check_fail "VRAM exporter script syntax error"
|
||||
fi
|
||||
|
||||
# Validate JSON dashboard (without Jinja2 rendering)
|
||||
if python3 -m json.tool "$ROLE_DIR/templates/llama-swap-grafana-dashboard.json.j2" > /dev/null 2>&1; then
|
||||
check_pass "Grafana dashboard template syntax (JSON)"
|
||||
else
|
||||
check_fail "Grafana dashboard template syntax error"
|
||||
fi
|
||||
|
||||
# Validate YAML structure (basic check)
|
||||
if grep -q "^- job_name:" "$ROLE_DIR/templates/llama-swap-prometheus-scrape.yml.j2"; then
|
||||
check_pass "Prometheus scrape template structure (YAML)"
|
||||
else
|
||||
check_fail "Prometheus scrape template structure error"
|
||||
fi
|
||||
|
||||
if grep -q "^kind: PrometheusRule" "$ROLE_DIR/templates/llama-swap-alerts.yml.j2"; then
|
||||
check_pass "Alert rules template structure (YAML)"
|
||||
else
|
||||
check_fail "Alert rules template structure error"
|
||||
fi
|
||||
|
||||
echo ""
|
||||
|
||||
# 5. Documentation completeness
|
||||
echo "5. Documentation Completeness"
|
||||
echo "=============================="
|
||||
|
||||
if grep -q "VRAM textfile exporter" "$ROLE_DIR/references/monitoring-llm-homelab-ciro-luciotta-2026.md"; then
|
||||
check_pass "Monitoring pattern docs include VRAM exporter section"
|
||||
else
|
||||
check_fail "Monitoring pattern docs incomplete: missing VRAM exporter section"
|
||||
fi
|
||||
|
||||
if grep -q "Grafana Dashboard Panels" "$ROLE_DIR/references/monitoring-llm-homelab-ciro-luciotta-2026.md"; then
|
||||
check_pass "Monitoring pattern docs include dashboard panels section"
|
||||
else
|
||||
check_fail "Monitoring pattern docs incomplete: missing dashboard panels section"
|
||||
fi
|
||||
|
||||
if grep -q "Alert Rules" "$ROLE_DIR/references/monitoring-llm-homelab-ciro-luciotta-2026.md"; then
|
||||
check_pass "Monitoring pattern docs include alert rules section"
|
||||
else
|
||||
check_fail "Monitoring pattern docs incomplete: missing alert rules section"
|
||||
fi
|
||||
|
||||
if grep -q "18560" "$ROLE_DIR/references/llama-swap-phase3-cutover-results-2026-08-18.md"; then
|
||||
check_pass "Phase 3 results include VRAM baseline figures"
|
||||
else
|
||||
check_fail "Phase 3 results incomplete: missing VRAM baseline"
|
||||
fi
|
||||
|
||||
echo ""
|
||||
|
||||
# 6. Summary
|
||||
echo "================================================================================"
|
||||
echo "Summary"
|
||||
echo "================================================================================"
|
||||
echo "Checks passed: ${GREEN}${CHECKS_PASSED}${NC}"
|
||||
echo "Checks failed: ${RED}${CHECKS_FAILED}${NC}"
|
||||
echo ""
|
||||
|
||||
if [ $CHECKS_FAILED -eq 0 ]; then
|
||||
echo -e "${GREEN}All checks passed! ✓${NC}"
|
||||
echo ""
|
||||
echo "Next steps:"
|
||||
echo " 1. Copy Grafana dashboard JSON to cluster/applications/monitoring/"
|
||||
echo " 2. Add Prometheus scrape config to cluster/applications/monitoring/values.yaml"
|
||||
echo " 3. Deploy PrometheusRule CR to cluster/applications/monitoring/"
|
||||
echo " 4. Commit to Git and push (ArgoCD syncs automatically)"
|
||||
echo " 5. Verify metrics in Prometheus UI: http://imagineering.local.mk-labs.cloud/prometheus"
|
||||
echo " 6. Verify dashboard in Grafana UI: http://imagineering.local.mk-labs.cloud/grafana"
|
||||
exit 0
|
||||
else
|
||||
echo -e "${RED}Some checks failed. See above for details.${NC}"
|
||||
exit 1
|
||||
fi
|
||||
Reference in New Issue
Block a user