-
Type:
New Feature
-
Resolution: Unresolved
-
Priority:
Major - P3
-
None
-
Affects Version/s: None
-
Component/s: None
-
StorEng - Refinement Pipeline
Summary:
Write a Python module that any workgen workload can utilise to encourage cache pressure along with the workload.
Detailed description:
Most workloads tested are somewhat static in what they do. An initial populate phase is followed by a consistent workload that could vary in types and number of operations done. There can also be variability in the key distribution for the operations.
These workloads generally lead to a stable cache constitution or predictable eviction states. I will like to test the system when it is pushed out of the stable state, specifically concerning cache constitution and eviction. Frequently customers (or MongoDB operations) end up running operations that generate cache pressure and destabilize the overall workload. Having the means to replicate some of these situations and then quantify how well we behave will be a starting point to make things better.
Note:
This work is somewhat experimental in nature and will be more clearly defined as we go.