Events with encoding errors produce misbehaving retries with some outputs

[jsoriano]

When events cannot be encoded by the output because they contain invalid values as NaN they should be dropped as the Elasticsearch output does.
But at least with logstash output the observed behaviour is that the whole batch including the wrong event is retried and/or dropped, leading to repeated or missing events, and to repeating messages in logs as the wrong event is also retried.

[marwojt]

Hi,
is there any progress in bug fixing or workaround for metricbeat and logstash output ?

[michalpristas]

@jsoriano was this fixed by any generic approach or just for prometheus in #12084

[jsoriano]

#12084 prevents the prometheus module to generate events with NaN values.

The issue reported here is that when some event cannot be encoded, the output sometimes misbehaves, affecting the whole batch and not only the invalid event.

[elasticmachine]

Pinging @elastic/integrations-services (Team:Services)

[elasticmachine]

Pinging @elastic/elastic-agent-data-plane (Team:Elastic-Agent-Data-Plane)