
The AWSLogs container log driver's non-blocking mode prevents application blocking during logging server outages by buffering logs in memory, with testing recommending a 25MB max-buffer-size to avoid log loss for up to 5MB/s throughput rates.


<div><p>This article explains how to prevent log loss in Amazon ECS using the AWSLogs container log driver's non-blocking mode, which prioritizes application availability over guaranteed log delivery.</p><ul><li>Blocking mode (default) halts application execution when logs cannot be sent; non-blocking mode buffers logs in memory</li><li>Recommended setting: max-buffer-size of 25MB for up to 5MB/s log output rate</li><li>Testing showed no log loss for ≤2MB/s with 4MB buffer; ≤5MB/s with 25MB buffer</li><li>Above 6MB/s sustained logging, log loss becomes difficult to prevent</li><li>Cross-region log uploads are unreliable; same-region uploads strongly recommended</li><li>No built-in metrics or alerts for detecting log loss with AWSLogs driver</li><li>Calculate per-container log throughput using CloudWatch IncomingBytes metric</li></ul><p>The article recommends 25MB max-buffer-size for most use cases and emphasizes same-region log uploads to minimize log loss risk while maintaining application availability.</p></div>


Related articles