Currently, we check the overall utilization to be below 75% to allow exceeding the soft concurrency limit. However, this check is global and doesn’t differentiate between standard and premium runners. Consequently, if the standard utilization is low, but a single customer with premium runners requests a large number of vCPUs, they could consume all available premium capacity. As a result, other customers with premium access might be blocked from getting premium runners. If both standard and premium utilization are above 75%, we don’t allow exceeding the soft concurrency limit. If premium utilization is high, but standard utilization is low, we allow exceeding the soft concurrency limit without a premium upgrade. To accommodate requested premium customers, I reduced the limit to 50% for free upgrades.
3.3 KiB
3.3 KiB