With gatling 2.0.3 and debug logging enabled [logback.groovy:logger(“io.gatling.http”, DEBUG) ], reported response times for our simulations are close enough to wireshark times to be acceptable for trend analysis (e.g. 15-20ms mean across 12 service calls per user).
If, however, we disable debug logging, the reported response times go up significantly, from 3x to 10x actual, making the response time reports highly suspect.
I have been running various permutations of our simulations (going back in time, etc) to try and isolate the issue. The one common factor is leaving logging enabled ensures valid close to valid response times.
I’ve examined the source code and can find no clear indication as to why the response times would become so inflated when debug logging is disabled.
Our load generators are heavy-duty 32 core, 32GB, server-grade systems in the data center and c3.large in AWS. Load tests have ranged from 1K CCU to 800K CCU.
Any advice or thoughts about how we might normalize the reported response times when logging is enabled or disabled would be greatly appreciated.