Temporal Affiliation
* [x] I am a Temporal employee
Page URL
https://docs.temporal.io/develop/worker-performance#choosing-slot-supplier-types
Content Issue
This section offers great suggestions for choosing a slot supplier, though I wonder if it would serve users better if it explicitly suggested fixed-sized tuners over resource-based tuners.
Currently the section reads as though it may suggest using the resource-based tuner by default, (i.e. "_You want acceptable performance with minimum effort_"). This is a bit of a dangerous suggestion, given that there are foot-guns with using the resource-based tuner, it is tricky to use effectively and really only suitable for specific workloads.
I think it's suggested use cases should be limited to just:
* Fluctuating workloads with low per-Task consumption
* Protection from out-of-memory & over-subscription in the face of unpredictable per-task consumption
in the "_The following use cases are well suited to resource-based auto-tuning slot suppliers:_" section.
In it's place, we should provide recommended default values for fixed-sized tuners, for each language.
Temporal Affiliation
* [x] I am a Temporal employee
Page URL
https://docs.temporal.io/develop/worker-performance#choosing-slot-supplier-types
Content Issue
This section offers great suggestions for choosing a slot supplier, though I wonder if it would serve users better if it explicitly suggested fixed-sized tuners over resource-based tuners.
Currently the section reads as though it may suggest using the resource-based tuner by default, (i.e. "_You want acceptable performance with minimum effort_"). This is a bit of a dangerous suggestion, given that there are foot-guns with using the resource-based tuner, it is tricky to use effectively and really only suitable for specific workloads.
I think it's suggested use cases should be limited to just:
* Fluctuating workloads with low per-Task consumption
* Protection from out-of-memory & over-subscription in the face of unpredictable per-task consumption
in the "_The following use cases are well suited to resource-based auto-tuning slot suppliers:_" section.
In it's place, we should provide recommended default values for fixed-sized tuners, for each language.