Skip to content

How to find the cause of CPU spikes

Finding the cause of CPU spikes

Question

When we get AWS CloudWatch alerts about high CPU usage on a server (shard/cluster), what is the best way to find the root cause in FusionReactor?

  1. Check if the issue is still ongoing or has resolved back to normal state.

    • If it was an alert from an earlier time (e.g., overnight), note the time and go to this timeframe within the Cloud UI (use the time selector in the top bar).

  2. Check the Metrics screen:

    • Look at the amount of transactions running at that time.
    • Review their completion time and CPU usage to identify when the spike started.

  3. View previous time frames:

    • When looking at the Web Requests, Transactions, or JDBC screens, go to ‘Saved in Cloud’ to view historical data.

    !Screenshot

  4. Investigate requests during the spike:

    • Identify requests with increased completion time and investigate the cause.
    • Open longer/erroring requests to check traces/sub-transactions and determine where extra time was spent.

  5. Check transaction types:

    • Determine if many larger scheduled tasks were executing simultaneously.

  6. Analyze the Applications screen:

    • Ensure you are looking at the correct timeframe.
    • Check performance graphs for anomalies or unexpected increases in throughput.

  7. Review Logging for the timeframe in question:

    • Look for unusual behavior.
    • Check the IP sending requests to detect if a bot is making multiple requests.

  8. Explore Traces for affected Services:

    • Run a query for the specific timeframe.
    • Focus on services with the longest durations to identify the root cause.