MongoDB Write Performance: Checkpoint and Cache Eviction

Hi All,

We are running a write performance test using stand-alone instances. We are seeing write latencies going up quite a bit. My initial theory is about database cache eviction and checkpoint.

My setup:

  • mongod pods running on k8s
  • wiredTiger cache size is set to 11GB
  • Using YCSB workload A (50% Queries / 50% Updates)
  • Always ramp up wiredTiger cache, bringing it to 80% utilization before every run

In the mongod.log file, we can see checkpoint messages showing the size of each checkpoint and how long it took to finish it. Parsing these messages from the log file isn’t a complex task, but I was wondering if there are any performance counters we could look at to identify as a signature of this situation?

Any thoughts?

All the best,

– Rodrigo

1 Like