Auto-sync: 20260116-155016

MONITORING.md

@@ -98,19 +98,21 @@ Monitors Claude Code token usage across all machines to track subscription consu
 ```
 Claude Code (MacBook/Mac Mini)
       │
-      ▼ (OTLP HTTP push)
+      ▼ (OTLP HTTP push every 60s)
       │
 OTEL Collector (docker-host:4318)
       │
-      ▼ (Remote Write)
+      ▼ (Prometheus exporter on :8889)
       │
-Prometheus (docker-host:9090)
+Prometheus (docker-host:9090) ─── scrapes ───► otel-collector:8889
       │
       ├──► Grafana Dashboard
       │
       └──► Alertmanager (burn rate alerts)
 ```
 
+**Note**: Uses Prometheus exporter instead of Remote Write because Claude Code sends Delta temporality metrics, which Remote Write doesn't support.
+
 **Monitored Devices**:
 All Claude Code sessions on any device automatically push metrics via OTLP.
 
@@ -132,23 +134,22 @@ All Claude Code sessions on any device automatically push metrics via OTLP.
 | No Metrics | Scrape fails for 5min | Info |
 
 **Configuration Files**:
-- Claude settings: `~/.claude/settings.json` (on each Mac - synced via Syncthing)
+- Shell config: `~/.zshrc` (on each Mac - synced via Syncthing)
 - OTEL Collector: `/opt/monitoring/otel-collector/config.yaml` (docker-host)
 - Alert rules: `/opt/monitoring/prometheus/rules/claude-code.yml` (docker-host)
 
-**Claude Code Settings** (in `~/.claude/settings.json`):
-```json
-{
-  "env": {
-    "CLAUDE_CODE_ENABLE_TELEMETRY": "1",
-    "OTEL_METRICS_EXPORTER": "otlp",
-    "OTEL_EXPORTER_OTLP_ENDPOINT": "http://10.10.10.206:4318",
-    "OTEL_EXPORTER_OTLP_PROTOCOL": "http/protobuf",
-    "OTEL_METRIC_EXPORT_INTERVAL": "60000"
-  }
-}
+**Shell Environment Setup** (in `~/.zshrc`):
+```bash
+# Claude Code OpenTelemetry Metrics (push to OTEL Collector)
+export CLAUDE_CODE_ENABLE_TELEMETRY=1
+export OTEL_METRICS_EXPORTER=otlp
+export OTEL_EXPORTER_OTLP_ENDPOINT="http://10.10.10.206:4318"
+export OTEL_EXPORTER_OTLP_PROTOCOL="http/protobuf"
+export OTEL_METRIC_EXPORT_INTERVAL=60000
 ```
 
+**Note**: These can be set either in shell environment (`~/.zshrc`) or in `~/.claude/settings.json` under the `env` block. Both methods work.
+
 **OTEL Collector Config** (`/opt/monitoring/otel-collector/config.yaml`):
 ```yaml
 receivers:
@@ -164,36 +165,51 @@ processors:
     timeout: 10s
 
 exporters:
-  prometheusremotewrite:
-    endpoint: "http://prometheus:9090/api/v1/write"
+  prometheus:
+    endpoint: 0.0.0.0:8889
+    resource_to_telemetry_conversion:
+      enabled: true
 
 service:
   pipelines:
     metrics:
       receivers: [otlp]
       processors: [batch]
-      exporters: [prometheusremotewrite]
+      exporters: [prometheus]
+```
+
+**Prometheus Scrape Config** (add to `/opt/monitoring/prometheus/prometheus.yml`):
+```yaml
+  - job_name: "claude-code"
+    static_configs:
+      - targets: ["otel-collector:8889"]
+        labels:
+          group: "claude-code"
 ```
 
 **Useful PromQL Queries**:
 ```promql
-# Total tokens this session
-sum(claude_code_token_usage_total)
+# Total tokens by model
+sum(claude_code_token_usage_tokens_total) by (model)
 
 # Burn rate (tokens/hour)
-sum(rate(claude_code_token_usage_total[1h])) * 3600
+sum(rate(claude_code_token_usage_tokens_total[1h])) * 3600
 
-# Usage by device
-sum(claude_code_token_usage_total) by (device)
+# Total cost by model
+sum(claude_code_cost_usage_USD_total) by (model)
 
-# Projected weekly usage
-sum(increase(claude_code_token_usage_total[24h])) * 7
+# Usage by type (input, output, cacheRead, cacheCreation)
+sum(claude_code_token_usage_tokens_total) by (type)
+
+# Projected weekly usage (rough estimate)
+sum(increase(claude_code_token_usage_tokens_total[24h])) * 7
 ```
 
 **Important Notes**:
-- Claude Code must be restarted after changing telemetry settings
+- After changing `~/.zshrc`, start a new terminal/shell session before running Claude Code
 - Metrics only flow while Claude Code is running
 - Weekly subscription resets Monday 1am (America/New_York)
+- Verify env vars are set: `env | grep OTEL`
 
 **Added**: 2026-01-16