You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Use Case:
We are benchmarking multiple systems to compare throughput. One specific metric that we are unable to test is whether pointing reads at a replica pool would improve the throughput for TPC-C load tests. We would still want all writes to proceed through the primary connection.
Considerations:
Replicas in a relational database like Postgresql have the potential for lag between the primary and the replica. Provided you have a system that can minimize this like Amazon Aurora or EDB BigAnimal this can be negligible and allow for the majority of applications to run separate read and write connections and have a low error/bad-data percentage. For a business that needs to add additional replicas to the read pool under high read load would be more likely to be in a position to allow a certain percentage of errors or inconsistent results in order to handle the increased system load.
Feature Request:
Add the ability to pass a separate read connection for reads in the configuration for TPC-C workloads
All read traffic (within reason) is sent through the separate read connection
If at all possible it would be ideal to have some measurement around incorrect result but this is likely a stretch
The text was updated successfully, but these errors were encountered:
Use Case:
We are benchmarking multiple systems to compare throughput. One specific metric that we are unable to test is whether pointing reads at a replica pool would improve the throughput for TPC-C load tests. We would still want all writes to proceed through the primary connection.
Considerations:
Replicas in a relational database like Postgresql have the potential for lag between the primary and the replica. Provided you have a system that can minimize this like Amazon Aurora or EDB BigAnimal this can be negligible and allow for the majority of applications to run separate read and write connections and have a low error/bad-data percentage. For a business that needs to add additional replicas to the read pool under high read load would be more likely to be in a position to allow a certain percentage of errors or inconsistent results in order to handle the increased system load.
Feature Request:
The text was updated successfully, but these errors were encountered: