Make sure that causal consistency performance impact expectations match the reality and quantify the coefficients.
reads secondary, auth is off: proportional to lag
reads secondary, auth is on: proportional to lag + constant overhead
reads primary, auth is off: no effect
reads primary, auth is on: constant overhead
also measure the performance when reads are from secondary and WaitForSecondaryBeforeNoopWriteMS is set to less then and greater then secondary lag. The latter case must demonstrate the step function as soon as noop writes will be generated.
- depends on
-
SERVER-30779 Investigate performance degradation in Causal Consistency enabled benchmark
- Closed