Slow start mode for Ring-hash load balancer

[leosarra]

Title: Implement "Slow Start mode" for "RING_HASH" Load Balancer in Istio/Envoy

Description:

I would like to propose an improvement to the load balancer behavior to support a 'slow start mode' for the ring-hash load balancer configuration.

Currently, the "least request/RR" load balancing strategy supports a "slow start up" feature, which is beneficial for warming up Java microservices before they begin serving requests at full capacity. This mechanism is especially useful during scale-out operations, as it prevents new instances from being overwhelmed by traffic immediately upon startup.

We find the ring-hash load balancing strategy especially useful to direct similar queries to the same instance, ensuring efficient cache utilization. However, the absence of a "slow start mode" result in new instances receiving a disproportionate amount of traffic, leading to request rejections and triggering Istio's outlier detection mechanism, which removes these instances from the pool.

In the original design document, here, it was mentioned that introducing periodic rebalancing was a prerequisite for potentially introducing a slow start mode in the hashing-based LB.

Is there any plan at the moment to bring a slow start mode for the ring hash load balancing strategy?

Relevant Links:

• Envoy Slow Start Config Documentation
• Original slow start mode PR

[tsaarni]

@nezdolik I was wondering if you might know if anyone has worked towards this or if you have any other pointers?

[github-actions]

This issue has been automatically marked as stale because it has not had activity in the last 30 days. It will be closed in the next 7 days unless it is tagged "help wanted" or "no stalebot" or other activity occurs. Thank you for your contributions.

[nezdolik]

As far as i know, no one has worked on supporting slow start for Ring Hash. Having low capacity at the moment due to other initiatives, but if anyone is interested to give it a try I will help out.