Practical Guide to Nginx API Gateway Performance Tuning

Scenario

An e-commerce platform uses Nginx as an API gateway. Users report slow API responses; p99 latency jumped from 200ms to 2s.

Symptoms

• Increased 502/504 errors from upstream
• Nginx error logs: "upstream timed out"
• Active connections near worker_connections limit

Diagnosis

1. Check Nginx error log: tail -f /var/log/nginx/error.log
2. Log upstream response time: add $upstream_response_time to access log format
3. Monitor connection pool: curl -s http://localhost/nginx_status | grep Active
4. Analyze buffer settings: default proxy_buffering can cause head-of-line blocking

Commands

Adjust worker_connections (in /etc/nginx/nginx.conf):

events {
    worker_connections 4096;
    multi_accept on;
}

Enable and tune buffers (in /etc/nginx/conf.d/api_gateway.conf):

location /api/ {
    proxy_pass http://upstream;
    proxy_buffering on;
    proxy_buffer_size 8k;
    proxy_buffers 8 32k;
    proxy_busy_buffers_size 64k;
    proxy_read_timeout 60s;
    proxy_send_timeout 60s;
    # Enable upstream keepalive
    proxy_http_version 1.1;
    proxy_set_header Connection "";
    keepalive_requests 1000;
    keepalive_timeout 120s;
}

Verify syntax: nginx -t; reload: systemctl reload nginx

Risk Controls

• Test on a canary instance first
• Monitor CPU, memory, and connection counts
• Ensure upstream service can handle increased connection reuse

Rollback

Keep backup of original config as .bak, replace and reload:

cp /etc/nginx/conf.d/api_gateway.conf.bak /etc/nginx/conf.d/api_gateway.conf
nginx -s reload

Verification

• Monitor p99 latency reduction via Prometheus/Grafana
• Check error log for fewer timeouts
• Load test: ab -n 10000 -c 100 http://gateway/api/test

When to Submit an OpsGlobal Ticket

If latency persists after tuning, or root cause is upstream application slowness, database queries, or DNS issues, submit a ticket for OpsGlobal SRE team deep analysis.