Algolia Observability

Algolia Observability

Overview

Algolia provides built-in analytics in the dashboard, but production systems need application-level observability: latency histograms, error rate counters, distributed traces, and alerts. This skill instruments the algoliasearch v5 client with Prometheus, OpenTelemetry, and structured logging.

Prerequisites

• Access to the application's metrics, tracing, and logging platform.
• A stable service name and index labels that distinguish production from test traffic.
• Permission to inspect Algolia request IDs and API responses without logging API keys or user PII.

Key Metrics to Track

Instructions

Examples

The client wrapper, metrics endpoint, tracing, and alert examples below create a correlated view of search latency, errors, and index activity. Preserve request IDs and aggregate labels; never emit search credentials in telemetry.

Step 1: Instrumented Algolia Client Wrapper

// src/algolia/instrumented-client.ts
import { algoliasearch, ApiError } from 'algoliasearch';
import { Counter, Histogram, Gauge, Registry } from 'prom-client';

const registry = new Registry();

const searchLatency = new Histogram({
  name: 'algolia_search_duration_seconds',
  help: 'Algolia search request duration in seconds',
  labelNames: ['index', 'status'],
  buckets: [0.01, 0.025, 0.05, 0.1, 0.25, 0.5, 1, 2.5],
  registers: [registry],
});

const searchTotal = new Counter({
  name: 'algolia_search_requests_total',
  help: 'Total Algolia search requests',
  labelNames: ['index', 'status'],
  registers: [registry],
});

const searchErrors = new Counter({
  name: 'algolia_errors_total',
  help: 'Total Algolia errors by type',
  labelNames: ['index', 'error_type', 'status_code'],
  registers: [registry],
});

const indexRecords = new Gauge({
  name: 'algolia_index_records',
  help: 'Number of records in Algolia index',
  labelNames: ['index'],
  registers: [registry],
});

const client = algoliasearch(process.env.ALGOLIA_APP_ID!, process.env.ALGOLIA_ADMIN_KEY!);

export async function instrumentedSearch<T = any>(
  indexName: string,
  searchParams: Record<string, any>
) {
  const timer = searchLatency.startTimer({ index: indexName });

  try {
    const result = await client.searchSingleIndex<T>({ indexName, searchParams });
    timer({ status: 'success' });
    searchTotal.inc({ index: indexName, status: 'success' });
    return result;
  } catch (error) {
    timer({ status: 'error' });
    searchTotal.inc({ index: indexName, status: 'error' });

    if (error instanceof ApiError) {
      searchErrors.inc({
        index: indexName,
        error_type: error.status === 429 ? 'rate_limit' : 'api_error',
        status_code: String(error.status),
      });
    } else {
      searchErrors.inc({
        index: indexName,
        error_type: 'network',
        status_code: '0',
      });
    }
    throw error;
  }
}

// Periodic index stats collection (run every 5 minutes)
export async function collectIndexMetrics() {
  const { items } = await client.listIndices();
  for (const idx of items) {
    indexRecords.set({ index: idx.name }, idx.entries || 0);
  }
}

export { registry };

Step 2: Prometheus Metrics Endpoint

// src/api/metrics.ts (Express example)
import express from 'express';
import { registry, collectIndexMetrics } from '../algolia/instrumented-client';

const app = express();

app.get('/metrics', async (_req, res) => {
  res.set('Content-Type', registry.contentType);
  res.send(await registry.metrics());
});

// Collect index stats every 5 minutes
setInterval(collectIndexMetrics, 5 * 60 * 1000);

Step 3: OpenTelemetry Distributed Tracing

// src/algolia/tracing.ts
import { trace, SpanStatusCode, type Span } from '@opentelemetry/api';

const tracer = trace.getTracer('algolia-service', '1.0.0');

export async function tracedSearch<T>(
  indexName: string,
  query: string,
  searchParams: Record<string, any> = {}
): Promise<T> {
  return tracer.startActiveSpan(`algolia.search ${indexName}`, async (span: Span) => {
    span.setAttribute('algolia.index', indexName);
    span.setAttribute('algolia.query', query);
    span.setAttribute('algolia.hitsPerPage', searchParams.hitsPerPage || 20);

    try {
      const result = await client.searchSingleIndex<T>({
        indexName,
        searchParams: { query, ...searchParams },
      });

      span.setAttribute('algolia.nbHits', result.nbHits);
      span.setAttribute('algolia.processingTimeMS', result.processingTimeMS);
      span.setStatus({ code: SpanStatusCode.OK });
      return result as T;
    } catch (error: any) {
      span.setStatus({ code: SpanStatusCode.ERROR, message: error.message });
      span.recordException(error);
      throw error;
    } finally {
      span.end();
    }
  });
}

Step 4: Structured Logging

// src/algolia/logger.ts
import pino from 'pino';

const logger = pino({ name: 'algolia', level: process.env.LOG_LEVEL || 'info' });

export function logSearch(params: {
  index: string;
  query: string;
  nbHits: number;
  processingTimeMS: number;
  page: number;
  userId?: string;
}) {
  logger.info({
    event: 'algolia.search',
    index: params.index,
    query: params.query,
    hits: params.nbHits,
    latency_ms: params.processingTimeMS,
    page: params.page,
    user: params.userId,
  });
}

export function logSearchError(params: {
  index: string;
  query: string;
  error: string;
  statusCode?: number;
}) {
  logger.error({
    event: 'algolia.search.error',
    index: params.index,
    query: params.query,
    error: params.error,
    status_code: params.statusCode,
  });
}

Step 5: Alert Rules (Prometheus AlertManager)

# alerts/algolia.yml
groups:
  - name: algolia
    rules:
      - alert: AlgoliaHighErrorRate
        expr: |
          rate(algolia_errors_total[5m]) /
          rate(algolia_search_requests_total[5m]) > 0.05
        for: 5m
        labels: { severity: warning }
        annotations:
          summary: "Algolia error rate > 5% for 5 minutes"

      - alert: AlgoliaHighLatency
        expr: |
          histogram_quantile(0.95,
            rate(algolia_search_duration_seconds_bucket[5m])
          ) > 0.5
        for: 5m
        labels: { severity: warning }
        annotations:
          summary: "Algolia P95 search latency > 500ms"

      - alert: AlgoliaRateLimited
        expr: rate(algolia_errors_total{error_type="rate_limit"}[5m]) > 0
        for: 2m
        labels: { severity: critical }
        annotations:
          summary: "Algolia returning 429 rate limit errors"

      - alert: AlgoliaIndexStale
        expr: algolia_index_records == 0
        for: 10m
        labels: { severity: warning }
        annotations:
          summary: "Algolia index has 0 records — possible sync failure"

Grafana Dashboard Queries

# Search rate: rate(algolia_search_requests_total[5m])
# Error rate: rate(algolia_errors_total[5m]) / rate(algolia_search_requests_total[5m])
# P50 latency: histogram_quantile(0.5, rate(algolia_search_duration_seconds_bucket[5m]))
# P95 latency: histogram_quantile(0.95, rate(algolia_search_duration_seconds_bucket[5m]))
# Records per index: algolia_index_records

Output

The service exposes actionable search health signals: latency and error rates, request traces, structured logs, dashboards, and alert conditions. Operators can distinguish an application regression from an Algolia incident using the correlated evidence.

Error Handling

Resources

• Prometheus Client
• OpenTelemetry JS
• Algolia Dashboard Analytics
• pino Logger

Next Steps

For incident response, see algolia-incident-runbook.