Common Log Errors — Fix 502s, OOMKilled Pods, and Deadlocks Fast

Why Log Errors Matter (And Cost You Real Money)

Let's talk numbers. According to Erwood Group's 2025 research, the average cost of IT downtime is now $14,056 per minute for mid-sized companies. For large enterprises, that number jumps to $23,750 per minute. Do the math: a 30-minute outage caused by an undiagnosed 502 Bad Gateway error costs your company over $400,000.

The old Gartner estimate from 2014 pegged downtime at $5,600/minute. In 2024, inflation and increased digital dependency have more than doubled that cost. If you're in retail e-commerce during Black Friday? You're looking at $1-2 million per hourin lost revenue, according to BigPanda's 2024 outage analysis.

But here's the thing: Most outages aren't caused by infrastructure failures. They're caused by slow incident detection. The DORA metrics (DevOps Research and Assessment) show that elite engineering teams have a Mean Time to Recovery (MTTR) under 1 hour. Low-performing teams? 1 to 6 months. The difference isn't luck — it's process.

How to Use This Error Library

This isn't just a list of error codes. It's a battle-tested troubleshooting system built from thousands of production incidents. Here's how to use it effectively:

Error Categories and Quick Fixes

These are the errors that keep you up at night. We've organized them by platform and included SQL queries, debugging workflows, and real-world solutions used by Google SRE teams, DevOps engineers, and platform architects.

SELECT
  time_local,
  request,
  status,
  upstream_response_time,
  upstream_addr
FROM logs
WHERE status IN (502, 504)
ORDER BY time_local DESC
LIMIT 100;

SELECT
  timestamp,
  pod_name,
  namespace,
  message
FROM logs
WHERE message ILIKE '%OOMKilled%'
   OR message ILIKE '%CrashLoopBackOff%'
ORDER BY timestamp DESC
LIMIT 50;

SELECT
  timestamp,
  message,
  duration_ms
FROM logs
WHERE message ILIKE '%deadlock%'
   OR duration_ms > 5000
ORDER BY timestamp DESC
LIMIT 100;

SELECT
  timestamp,
  request_id,
  error_code,
  key
FROM logs
WHERE error_code IN ('AccessDenied', 'ThrottlingException')
GROUP BY error_code, key
ORDER BY COUNT(*) DESC
LIMIT 20;

The 5-Step Debugging Workflow

When an error appears, don't panic. Follow this systematic process used by Google SRE teams:

1. 1

   Isolate the Symptom

   What's the error code? Which service? What timestamp? Use WHERE clauses to filter logs.

2. 2

   Measure Frequency

   Is this a one-off spike or sustained failure? Use GROUP BY time_bucket to plot error rates.

3. 3

   Find Correlations

   What changed recently? Deployments? Config updates? Use JOIN with deployment logs.

4. 4

   Trace the Request Path

   Follow request IDs across services. Use WHERE request_id = '...' to build trace timeline.

5. 5

   Test the Fix

   Apply the solution. Monitor error rates for 15 minutes. If errors persist, rollback and repeat from step 1.

When to Escalate

Not every error needs immediate escalation. But some do. Here's when to page the on-call engineer:

• →Error persists >1 hour despite troubleshooting attempts
• →Cascading failures across multiple services or regions
• →Data loss risk (e.g., database corruption, replication failure)
• →Security breach indicators (unauthorized access, injection attempts)

According to DORA's 2024 metrics, elite teams have clear escalation policies that reduce MTTR by 60%. Know when to ask for help.

Preventative Strategies

The best way to handle errors is to catch them before they become incidents. Here's how: