CrashLoop OOM and Resource Pressure¶

1. Summary¶

Symptom¶

Replicas repeatedly restart with CrashLoopBackOff, OOMKilled, or ContainerTerminated events. Application may briefly serve traffic before crashing, or fail to start at all. Latency spikes and timeouts often precede or accompany crash cycles.

Why this scenario is confusing¶

Resource pressure manifests differently depending on the bottleneck (memory vs CPU vs startup time). An application bug can look like a resource limit issue, and vice versa. OOM kills may occur without obvious memory warnings if the spike is sudden.

Troubleshooting decision flow¶

graph TD
    A[Symptom: Frequent restarts] --> B{OOMKilled signal in logs?}
    B -->|Yes| H1[H1: Memory limit exceeded]
    B -->|No| C{Exit code 137?}
    C -->|Yes| H1
    C -->|No| D{Probe failures during startup?}
    D -->|Yes| H2[H2: CPU throttling delays startup]
    D -->|No| E{Crash after serving traffic?}
    E -->|Yes| H3[H3: Memory leak or burst allocation]
    E -->|No| H4[H4: Application startup crash]

2. Common Misreadings¶

• "Application bug only" — Resource pressure can trigger crashes in correct code.
• "Just raise memory limit" — Unbounded memory growth will still fail; root cause matters.
• "CPU throttling won't crash the app" — Probe timeouts due to CPU starvation cause restarts.
• "OOMKilled always means memory leak" — Can also be legitimate peak usage exceeding limits.
• "Exit code 1 means app bug" — Could be uncaught exception from resource exhaustion.

3. Competing Hypotheses¶

4. What to Check First¶

Metrics¶

• Memory working set vs memory limit over time
• CPU usage vs CPU limit (throttling indicator)
• Restart count trend
• Request latency percentiles (P95, P99) before crashes

Logs¶

let AppName = "ca-myapp";
ContainerAppSystemLogs_CL
| where ContainerAppName_s == AppName
| where TimeGenerated > ago(2h)
| where Reason_s has_any ("OOMKilled", "CrashLoopBackOff", "ContainerTerminated", "BackOff", "Killed")
   or Log_s has_any ("OOM", "killed", "terminated", "exit code", "signal 9", "signal 137")
| project TimeGenerated, RevisionName_s, ReplicaName_s, Reason_s, Log_s
| order by TimeGenerated desc

Platform Signals¶

# Check resource allocation
az containerapp show --name "$APP_NAME" --resource-group "$RG" \
  --query "properties.template.containers[0].resources" --output json

# Check probe configuration
az containerapp show --name "$APP_NAME" --resource-group "$RG" \
  --query "properties.template.containers[0].probes" --output json

# Check replica status and restart count
az containerapp replica list --name "$APP_NAME" --resource-group "$RG" --output table

5. Evidence to Collect¶

Required Evidence¶

Useful Context¶

• Application memory footprint at idle and under load
• Startup time requirements
• Recent code or dependency changes
• Traffic pattern during incidents

6. Validation and Disproof by Hypothesis¶

H1: Memory limit too low¶

Signals that support:

• Explicit OOMKilled in system logs
• Exit code 137 (128 + SIGKILL signal 9)
• Container terminated abruptly without graceful shutdown logs
• Memory usage approaching limit before termination

Signals that weaken:

• Memory usage consistently well below limit
• Graceful shutdown messages in logs
• Exit code 1 with application exception

What to verify:

// Find OOM signals
let AppName = "ca-myapp";
ContainerAppSystemLogs_CL
| where ContainerAppName_s == AppName
| where TimeGenerated > ago(6h)
| where Log_s has_any ("OOM", "137", "killed", "memory")
| project TimeGenerated, RevisionName_s, Log_s
| order by TimeGenerated desc

# Check current memory limits
az containerapp show --name "$APP_NAME" --resource-group "$RG" \
  --query "properties.template.containers[0].resources.memory" --output tsv

# Typical fix: increase memory
az containerapp update --name "$APP_NAME" --resource-group "$RG" \
  --memory "1.0Gi" --cpu "0.5"

H2: CPU throttling delays startup¶

Signals that support:

• Probe failures during startup phase
• Startup takes longer than probe initialDelaySeconds
• CPU limit is very low (e.g., 0.25 cores)
• Logs show slow initialization steps

Signals that weaken:

• Startup completes within probe window
• CPU usage well below limit
• Crashes occur after stable running period

What to verify:

# Check probe configuration
az containerapp show --name "$APP_NAME" --resource-group "$RG" \
  --query "properties.template.containers[0].probes" --output json

# Check CPU allocation
az containerapp show --name "$APP_NAME" --resource-group "$RG" \
  --query "properties.template.containers[0].resources.cpu" --output tsv

// Find probe failures
let AppName = "ca-myapp";
ContainerAppSystemLogs_CL
| where ContainerAppName_s == AppName
| where TimeGenerated > ago(2h)
| where Reason_s == "ProbeFailed" or Log_s has "probe"
| project TimeGenerated, RevisionName_s, Log_s
| order by TimeGenerated desc

H3: Memory leak or burst allocation¶

Signals that support:

• Memory grows steadily over time (saw-tooth pattern)
• Crashes occur after serving traffic for some period
• Restart temporarily fixes the issue
• Specific endpoints or operations correlate with crashes

Signals that weaken:

• Memory stable throughout operation
• Crashes during startup before any traffic

What to verify:

// Correlate crashes with traffic
let AppName = "ca-myapp";
let Crashes = ContainerAppSystemLogs_CL
| where ContainerAppName_s == AppName
| where TimeGenerated > ago(6h)
| where Reason_s has_any ("OOMKilled", "ContainerTerminated")
| project CrashTime=TimeGenerated;
// Cross-reference with request patterns in Application Insights if available

# Check for memory-related app settings
az containerapp show --name "$APP_NAME" --resource-group "$RG" \
  --query "properties.template.containers[0].env[?contains(name, 'MEMORY') || contains(name, 'HEAP')]" \
  --output table

H4: Application startup crash¶

Signals that support:

• Exit code 1 with exception stack trace
• Crashes immediately at startup
• Missing configuration or environment variables
• Dependency connection failures in logs

Signals that weaken:

• Successful startup, crashes only under load
• OOMKilled or exit code 137

What to verify:

# Get console logs for stack traces
az containerapp logs show --name "$APP_NAME" --resource-group "$RG" \
  --type console --tail 200

// Find startup errors
let AppName = "ca-myapp";
ContainerAppConsoleLogs_CL
| where ContainerAppName_s == AppName
| where TimeGenerated > ago(2h)
| where Log_s has_any ("error", "exception", "failed", "traceback", "Error:", "Exception:")
| project TimeGenerated, Log_s
| order by TimeGenerated desc
| take 50

7. Likely Root Cause Patterns¶

8. Immediate Mitigations¶

1. If OOM: Increase memory limit

   az containerapp update --name "$APP_NAME" --resource-group "$RG" \
     --memory "2.0Gi" --cpu "1.0"
   

1. If probe timeout: Relax probe settings

   az containerapp update --name "$APP_NAME" --resource-group "$RG" \
     --yaml probe-config.yaml  # With increased initialDelaySeconds
   

1. If startup crash: Roll back to known good revision

   az containerapp ingress traffic set --name "$APP_NAME" --resource-group "$RG" \
     --revision-weight "<previous-revision>=100"
   

1. If memory leak: Restart replicas while investigating

   az containerapp revision restart --name "$APP_NAME" --resource-group "$RG" \
     --revision "<current-revision>"
   

9. Prevention¶

• Baseline resource profiles in staging before production
• Set resource requests based on observed P95 usage + headroom
• Implement memory monitoring and alerts in Application Insights
• Add startup health endpoints that fail fast on missing dependencies
• Use liveness probes with appropriate failure thresholds
• Apply performance regression checks in CI pipeline

See Also¶

• Container Start Failure
• Probe Failure and Slow Start
• HTTP Scaling Not Triggering
• Replica Crash Signals KQL
• Restart Timing Correlation KQL
• Repeated Startup Attempts KQL

Sources¶

• Set up health probes for a container app
• Manage resource allocation in Azure Container Apps
• Troubleshoot Azure Container Apps