Backend/API Diagnostics

Purpose: Troubleshoot backend services, APIs, and application-level performance issues.

Common Backend Issues

1. Slow API Response

Symptoms:

• API response time >1 second
• Users report slow loading
• Timeout errors

Diagnosis:

Check Application Logs

# Check for slow requests
grep "duration" /var/log/application.log | awk '{if ($5 > 1000) print}'

# Check error rate
grep "ERROR" /var/log/application.log | wc -l

# Check recent errors
tail -f /var/log/application.log | grep "ERROR"

Red flags:

• Repeated errors for same endpoint
• Increasing response times
• Timeout errors

---

Check Application Metrics

# CPU usage
top -bn1 | grep "node\|java\|python"

# Memory usage
ps aux | grep "node\|java\|python" | awk '{print $4, $11}'

# Thread count
ps -eLf | grep "node\|java\|python" | wc -l

# Open file descriptors
lsof -p <PID> | wc -l

Red flags:

• CPU >80%
• Memory increasing over time
• Thread count increasing (thread leak)
• File descriptors increasing (connection leak)

---

Check Database Query Time

# If slow, likely database issue
# See database-diagnostics.md

# Check if query time matches API response time
# API response time = Query time + Application processing

---

Check External API Calls

# Check if calling external APIs
grep "http.request" /var/log/application.log

# Check external API response time
# Use APM tools or custom instrumentation

Red flags:

• External API taking >500ms
• External API rate limiting (429 errors)
• External API errors (5xx errors)

Mitigation:

• Cache external API responses
• Add timeout (don't wait >5s)
• Circuit breaker pattern
• Fallback data

---

2. 5xx Errors (500, 502, 503, 504)

Symptoms:

• Users getting error messages
• Monitoring alerts for error rate
• Some/all requests failing

Diagnosis by Error Code:

500 Internal Server Error

Cause: Application code error

Diagnosis:

# Check application logs for exceptions
grep "Exception\|Error" /var/log/application.log | tail -20

# Check stack traces
tail -100 /var/log/application.log

Common causes:

• NullPointerException / TypeError
• Unhandled promise rejection
• Database connection error
• Missing environment variable

Mitigation:

• Fix bug in code
• Add error handling
• Add input validation
• Add monitoring for this error

---

502 Bad Gateway

Cause: Reverse proxy can't reach backend

Diagnosis:

# Check if application is running
ps aux | grep "node\|java\|python"

# Check application port
netstat -tlnp | grep <PORT>

# Check reverse proxy logs (nginx, apache)
tail -f /var/log/nginx/error.log

Common causes:

• Application crashed
• Application not listening on expected port
• Firewall blocking connection
• Reverse proxy misconfigured

Mitigation:

• Restart application
• Check application logs for crash reason
• Verify port configuration
• Check reverse proxy config

---

503 Service Unavailable

Cause: Application overloaded or unhealthy

Diagnosis:

# Check application health
curl http://localhost:<PORT>/health

# Check connection pool
# Database connections, HTTP connections

# Check queue depth
# Message queues, task queues

Common causes:

• Too many concurrent requests
• Database connection pool exhausted
• Dependency service down
• Health check failing

Mitigation:

• Scale horizontally (add more instances)
• Increase connection pool size
• Rate limiting
• Circuit breaker for dependencies

---

504 Gateway Timeout

Cause: Application took too long to respond

Diagnosis:

# Check what's slow
# Database query? External API? Long computation?

# Check application logs for slow operations
grep "slow\|timeout" /var/log/application.log

Common causes:

• Slow database query
• Slow external API call
• Long-running computation
• Deadlock

Mitigation:

• Optimize slow operation
• Add timeout to prevent indefinite wait
• Async processing (return 202 Accepted)
• Increase timeout (last resort)

---

3. Memory Leak (Backend)

Symptoms:

• Memory usage increasing over time
• Application crashes with OutOfMemoryError
• Performance degrades over time

Diagnosis:

Monitor Memory Over Time

# Linux
watch -n 5 'ps aux | grep <PROCESS> | awk "{print \$4, \$5, \$6}"'

# Get heap dump (Java)
jmap -dump:format=b,file=heap.bin <PID>

# Get heap snapshot (Node.js)
node --inspect index.js
# Chrome DevTools → Memory → Take heap snapshot

Red flags:

• Memory increasing linearly
• Memory not released after GC
• Large arrays/objects in heap dump

---

Common Causes

// 1. Event listeners not removed
emitter.on('event', handler); // Never removed

// 2. Timers not cleared
setInterval(() => { /* ... */ }, 1000); // Never cleared

// 3. Global variables growing
global.cache = {}; // Grows forever

// 4. Closures holding references
function createHandler() {
  const largeData = new Array(1000000);
  return () => {
    // Closure keeps largeData in memory
  };
}

// 5. Connection leaks
const conn = await db.connect();
// Never closed → connection pool exhausted

Mitigation:

// 1. Remove event listeners
const handler = () => { /* ... */ };
emitter.on('event', handler);
// Later:
emitter.off('event', handler);

// 2. Clear timers
const intervalId = setInterval(() => { /* ... */ }, 1000);
// Later:
clearInterval(intervalId);

// 3. Use LRU cache
const LRU = require('lru-cache');
const cache = new LRU({ max: 1000 });

// 4. Be careful with closures
function createHandler() {
  return () => {
    const largeData = loadData(); // Load when needed
  };
}

// 5. Always close connections
const conn = await db.connect();
try {
  await conn.query(/* ... */);
} finally {
  await conn.close();
}

---

4. High CPU Usage

Symptoms:

• CPU at 100%
• Slow response times
• Server becomes unresponsive

Diagnosis:

Identify CPU-heavy Process

# Top CPU processes
top -bn1 | head -20

# CPU per thread (Java)
top -H -p <PID>

# Profile application (Node.js)
node --prof index.js
node --prof-process isolate-*.log

Common causes:

• Infinite loop
• Heavy computation (parsing, encryption)
• Regular expression catastrophic backtracking
• Large JSON parsing

Mitigation:

// 1. Break up heavy computation
async function processLargeArray(items) {
  for (let i = 0; i < items.length; i++) {
    await processItem(items[i]);

    // Yield to event loop
    if (i % 100 === 0) {
      await new Promise(resolve => setImmediate(resolve));
    }
  }
}

// 2. Use worker threads (Node.js)
const { Worker } = require('worker_threads');
const worker = new Worker('./heavy-computation.js');

// 3. Cache results
const cache = new Map();
function expensiveOperation(input) {
  if (cache.has(input)) return cache.get(input);
  const result = /* heavy computation */;
  cache.set(input, result);
  return result;
}

// 4. Fix regex
// Bad: /(.+)*/ (catastrophic backtracking)
// Good: /(.+?)/ (non-greedy)

---

5. Connection Pool Exhausted

Symptoms:

• "Connection pool exhausted" errors
• "Too many connections" errors
• Requests timing out

Diagnosis:

Check Connection Pool

# Database connections
# PostgreSQL:
SELECT count(*) FROM pg_stat_activity;

# MySQL:
SHOW PROCESSLIST;

# Application connection pool
# Check application metrics/logs

Red flags:

• Connections = max pool size
• Idle connections in transaction
• Long-running queries holding connections

Common causes:

• Connections not released (missing .close())
• Connection leak in error path
• Pool size too small
• Long-running queries

Mitigation:

// 1. Always close connections
async function queryDatabase() {
  const conn = await pool.connect();
  try {
    const result = await conn.query('SELECT * FROM users');
    return result;
  } finally {
    conn.release(); // CRITICAL
  }
}

// 2. Use connection pool wrapper
const pool = new Pool({
  max: 20, // max connections
  idleTimeoutMillis: 30000,
  connectionTimeoutMillis: 2000,
});

// 3. Monitor pool metrics
pool.on('error', (err) => {
  console.error('Pool error:', err);
});

// 4. Increase pool size (if needed)
// But investigate leaks first!

---

Backend Performance Metrics

Response Time:

• p50: <100ms
• p95: <500ms
• p99: <1s

Throughput:

• Requests per second (RPS)
• Requests per minute (RPM)

Error Rate:

• Target: <0.1%
• 4xx errors: Client errors (validation)
• 5xx errors: Server errors (bugs, downtime)

Resource Usage:

• CPU: <70% average
• Memory: <80% of limit
• Connections: <80% of pool size

Availability:

• Target: 99.9% (8.76 hours downtime/year)
• 99.99%: 52.6 minutes downtime/year
• 99.999%: 5.26 minutes downtime/year

---

Backend Diagnostic Checklist

When diagnosing slow backend:

• Check application logs for errors
• Check CPU usage (target: <70%)
• Check memory usage (target: <80%)
• Check database query time (see database-diagnostics.md)
• Check external API calls (timeout, errors)
• Check connection pool (target: <80% used)
• Check error rate (target: <0.1%)
• Check response time percentiles (p95, p99)
• Check for thread leaks (increasing thread count)
• Check for memory leaks (increasing memory over time)

Tools:

• Application logs
• APM tools (New Relic, DataDog, AppDynamics)
• top, htop, ps, lsof
• curl with timing
• Profilers (node --prof, jstack, py-spy)

---

Related Documentation

• SKILL.md - Main SRE agent
• database-diagnostics.md - Database troubleshooting
• infrastructure.md - Server/network troubleshooting
• monitoring.md - Observability tools