@nodellmcache/otel

OpenTelemetry exporter for NodeLLMCache observability. Bridges a MetricsCollector to OpenTelemetry instruments so your cache metrics flow into whatever backend your OTel SDK exports to (OTLP, Prometheus, Datadog, ...).

Install

npm install @nodellmcache/otel @nodellmcache/observability @nodellmcache/core @opentelemetry/api

Quick start

import { attachOtel } from '@nodellmcache/otel'
import { observability } from '@nodellmcache/observability'
import { PromptCache } from '@nodellmcache/prompt-cache'
import { MemoryAdapter } from '@nodellmcache/memory'

// Your app configures an OTel MeterProvider + exporter as usual.

// Wire cache metrics in (defaults to the shared observability singleton):
const handle = attachOtel()

const cache = new PromptCache({ adapter: new MemoryAdapter(), metrics: observability })
// ...later: handle.detach()

Works whether or not a MeterProvider is configured — with none, the OTel API uses a no-op meter, so attaching is always safe.

Instruments

| Instrument | Type | Attributes | |------------|------|-----------| | nodellmcache.cache.requests | counter | result (hit/miss), cache_type | | nodellmcache.cache.latency | histogram (ms) | cache_type | | nodellmcache.cache.tokens_saved | counter | cache_type | | nodellmcache.cache.sets | counter | cache_type | | nodellmcache.cache.evictions | counter | cache_type |

Options

attachOtel(collector, {
  meter,            // a custom OTel Meter (defaults to metrics.getMeter(...))
  meterName,        // name when one isn't injected (default '@nodellmcache/otel')
  prefix,           // instrument name prefix (default 'nodellmcache')
})

attachOtel returns { detach() } to remove the listeners.

License

MIT