Model testing

[marvin-j97]

!!!

See the model-testing branch for basic setup

!!!

Run both normal and a key-value separated tree in model-based test, where:

• random keys are written
• random keys/ranges/prefixes are read (forwards and reverse ranges/prefixes!)
• here and there a compaction is run; synchronously, so it's deterministic (Tree::flush_active_memtable); followed by a compaction as well (Tree::compact)
• the state of the tree is compared to an in-memory BTreeMap after every operation
• if the trees do not match -> fail

Considerations

• IO errors should just stop execution (unwrap)
• because at some point the trees and op log could become too large, we may need to nuke all data some times and start with a clean slate
• all operations need to be logged, so we can see where something is going wrong (should probably be stored to disk)
• keys need to be overwritten and deleted sometimes, so updates/tombstone shadowing are tested as well; and blob GC can also be invoked some times
• test snapshots? may be harder - for now, just use an incrementing sequence number generator for both inserts and reads
• Checking the entire tree after every operation is expensive - if we assume, let's say, an insert not to disrupt other, irrelevant data, we can only check for that key (and maybe some ranges that include it), and only do a full tree check here and there
• especially the blob tree should sometimes receive large values for performing key-value separation with

[harshithsaiv]

Hi @marvin-j97 , I'm interested in taking on this

[marvin-j97]

Hi @marvin-j97 , I'm interested in taking on this

I pushed a branch (model-testing) yesterday with a basic scaffold in the model folder; though the blob tree cannot be tested yet until #157 is merged into 3.0.0.