Four most-used shard key patterns: single_hashed, ranged_compound, single_ranged, id_hashed

Test plan

How long does analyzeShardKey take to run by itself

without samples

T1. Run ycsb_load for {5M, 50M, 1000M} documents, followed by analyzeShardKey for {field0: 1, field1: "hashed"}. This will tell us whether run-time of the command is linear w.r.t. number of documents.

T2. Repeat test T1 for {field0: 1}

T2a. Repeat test T1 for two analyzeShardKey commands running concurrently, one command for each key.

with samples

T3. For each of the clusters containing {50M} documents, run {30, 60} minutes of ycsb_read50update50 with sampling rate of 50/sec, followed by analyzeShardKey for {field0: 1, field1: "hashed"}. This will tell us how much longer the command takes to gather read/write distribution metrics.

T4. Repeat test T3 for {field0: 1}

T4a. Repeat test T3 for two analyzeShardKey commands running concurrently, one command for each key.

How does analyzeShardKey and query sampling impact concurrent workload performance

T5. Run ycsb_load for {1000M} documents, followed by a long run of ycsb_read50update50 with sampling rate of 50/sec. While ycsb_read50update50 is running, run analyzeShardKey every X minutes. (X depends on how long analyzeShardKey takes to run.)

Full test descriptions are here: https://docs.google.com/document/d/1FjAvT-XCASxseYEFos4CX57ZFI_vSrb7UgQtTVzLvgU/edit?usp=sharing