here’s the raw note: * Performance Test Dashboards

• Things we want to chart - things users actually do

• Adding a shard (1TB sharded collection)

• moveChunk

• Throughput (1TB sharded collection)
• Impact on cluster (refresh latency)

• Range Deleter (500GB)

• Throughput
• Impact on Cluster

• Removing a shard (1TB shard collection)

• moveChunk

• Throughput (1TB sharded collection)
• Impact on cluster (refresh latency)

• Range Deleter (500GB)

• Throughput
• Impact on Cluster

• shardCollection performance (1TB collection)
• movePrimary performance (1TB unsharded collection)

• Does the shape of data, indexes, matter?
• Chunk is logical (documents) instead of physical (pages)
• Any plan to provide low cost options in Atlas? NO
• How do you want to compare it? (version to version, build to build, feature flags?)
• REST API to add and remove?
• Steps to get there:

• Find 1TB data (Jim has it!)
• Create script to provision cluster
• Create script to run the test

• Run test
• Add shared
• Remove shared
• Repeat 3 times

• Create script to collect data

• Throughput/Execution time
• Latency graph (P50/P99) for test

• Create script to analyze data

•  

• Create script to clean up

•  

• Create report

•  

• Automate this