Delivery behavior and operational limits

Delivery happens on a worker thread rather than the application's calling thread. This keeps logging calls from waiting on Elasticsearch or OpenSearch, but it is not a durable queue:

  • Logs can be lost when the target cluster is unavailable and the bounded send buffer fills.
  • Pending logs can remain unsent when retries are exhausted or shutdown reaches shutdownTimeout.
  • HTTP 413 Payload Too Large responses cause the bulk request to be split recursively at event boundaries. Only an event that is still rejected after isolation is dropped.
  • A Bulk API response can have HTTP 200 and still contain item-level errors. Successful events are removed, events rejected with 429 or 5xx remain buffered for retry, and permanently rejected 4xx events are dropped with a warning.
  • Other HTTP 4xx responses clear the current send buffer. Other failures are retried according to maxRetries.
  • The appender uses the configured Bulk API URL directly. It does not discover cluster nodes or manage index templates and mappings.

For workloads that require durable, guaranteed delivery, use an external log shipper or queue between the application and Elasticsearch or OpenSearch.