Debugging Challenges

Introduction

Debugging MongoDB problems is a learnable craft: the symptoms (slow queries, lag, OOMs, connection storms) map to a small set of causes. These challenges train the pattern-matching muscle so production triage takes minutes, not hours.

Understanding the topic

Symptom → likely cause map:

• Spike in CPU on primary → unindexed query, runaway aggregation, or noisy neighbour.
• Replication lag growing → slow secondary, long-running op on primary blocking oplog apply.
• Connection count near limit → app pool misconfigured or connection leak.
• OOM on a node → working set > RAM, or huge in-memory $sort/$group.
• "Cursor not found" errors → cursor timeout (default 10 min) exceeded.
• WriteConflict storm → many concurrent updates to the same doc.

Informative example

Find and kill a runaway query in seconds:

db.currentOp({
  active: true,
  "command.aggregate": { $exists: true },
  microsecs_running: { $gt: 2 * 1000000 }
});

// Identify opid, then:
db.killOp(<opid>);

Best practices

• Keep a personal "symptoms → causes" cheat sheet; update it after every incident.
• Reproduce in staging with realistic data before "fixing" in prod.
• Always grab evidence (Atlas Profiler export, currentOp snapshot) before mitigating.