Skip to main content

## Intermittent 503 Error Analysis### Root CauseYou have a connection timeout mismatch:- Spring Gateway maxIdle: 59 seconds- ALB idle timeout: 60 seconds### Why This Causes 503Timeline of the problem:1. At 59s: Spring Gateway closes the idle connection2. At 60s: ALB still thinks the connection is open3. New request arrives → ALB tries to use the closed connection4. Result: 503 Service Unavailable### The RuleBackend timeout must be GREATER than load balancer timeout✗ Wrong: Gateway 59s < ALB 60s → 503 errors

  1. acshame
    https://dev.to/aws-builders/kubernetes-503-errors-with-aws-alb-possible-causes-and-solutions-1ddh
    ## Intermittent 503 Error Analysis

    ### Root Cause
    You have a connection timeout mismatch:
    - Spring Gateway maxIdle: 59 seconds
    - ALB idle timeout: 60 seconds

    ### Why This Causes 503

    Timeline of the problem:
    1. At 59s: Spring Gateway closes the idle connection
    2. At 60s: ALB still thinks the connection is open
    3. New request arrives → ALB tries to use the closed connection
    4. Result: 503 Service Unavailable

    ### The Rule
    Backend timeout must be GREATER than load balancer timeout

    ✗ Wrong:  Gateway 59s < ALB 60s  → 503 errors
✓ Correct: Gateway 65s > ALB 60s  → No errors

    ### Solution

    Option 1: Increase Spring Gateway timeout (Recommended)
    spring:
  cloud:
    gateway:
      httpclient:
        pool:
          max-idle-time: 65s  # Must be > 60s

    Option 2: Decrease ALB timeout
    # Set ALB to 55 seconds
alb.ingress.kubernetes.io/load-balancer-attributes: 
  idle_timeout.timeout_seconds=55

    ### Why This Happens
    - Occurs during low traffic (connections stay idle longer)
    - Creates a 1-second race condition (59s-60s window)
    - ALB reuses a connection that Spring already closed

    ### Validation from AWS
    AWS documentation confirms: backend keep-alive timeout should be greater than the load balancer's idle timeout [AWS re:Post](https://repost.aws/knowledge-center/eks-http-504-errors) to prevent exactly this issue.

    Your diagnosis is 100% correct! This is a classic connection pool timing problem.