Investigate options to support a better recovery of overloaded backends

[a-thaler]

Description

When using SAP Cloud Logging with the dev plan as log backend, a high log throughput can bring the instance into an overload situation easy. If the refusal rate is getting too high, the buffer used in fluent-bit will fill up as expected, and the items in the buffer will be retried repeatedly using a exponential backoff period. That retries of the buffered items will bring the system under load even more into trouble and a recovery is getting unlikely.
Especially, if the input log load is reduced, the retries of the buffered items will keep the backend so much under pressure that the recover might not succeed.

Goal: Investigate options to support a recovery of a overloaded backend

Ideas:

• Support some kind of "dev" mode where the buffer is switched off or perform just one retry
• Support a configurable "time-in-queue" setting to have the amount of retries configurable
• Increase the minimal time between retries

First experiments showed that a increased minimal retry period has no big effect. Instead reducing the amount of retries to a low value (3) relaxed the situation very fast

[github-actions]

This issue has been automatically marked as stale due to the lack of recent activity. It will soon be closed if no further activity occurs.
Thank you for your contributions.