Consumer appears to spam connections to fluentd in-forward plugin causing system instability

[pavgup]

I'm having a hard time pinning down exactly what is happening to a machine using this consumer in concert with fluentd's in-forward plugin, but with low flush times and a slow flush executable, it appears fluentd's in-forward plugin begins to be spammed with connection attempts to flush the kafka-fluentd-consumer buffers despite fluentd reporting its buffers are full. Is it possible to ask this consumer to initiate connections to the in-forward plugin at a slower clip and more successful clip? This might leave behind fewer lingering connections and perhaps improve stability.

[pavgup]

This issue an be closed. It was later identified we were experiencing hardware problems.