In depth review of other algorithms

Design document

Protocol

• Feature Flag (!19978 (merged))
• Change the sampler to select bakers using the SWRR algorithm
  • Implement SWRR (!20084 (merged))
  • Implement baker selection with shift for rounds (!20084 (merged))
  • Plug in the new sampler (!20180)
    • delegate_sampler for baking_rights_owner (with the feature flag)
    • delegate_cycles: compute the rights for future cycles
• Update Credits for deactivated delegates at cycle end (!20180)
• Benchmarks + optimizations
  • Optimize IO accesses in SWRR algorithm (!20084 (merged))
  • Cache access to selected baker list for a given cycle
  • Optimize round 0 attribution for faster access (even for higher rounds)

Baker

• Rounds computation might need to be changed/optimized (through a change to the Validators RPC)

Client

Nothing to change

RPC

• New RPCs
  • Return the current SWRR state (the remaining credits) (!20180)
    • This would be useful if external tools try to reproduce the algorithm locally. This is the only state info they would need, in addition to the stake distribution
  • Return the selected bakers for a given cycle (can specify the cycle, up to current + consensus_rights_delay) (!20180)
• Old RPC Changes
  • Validators
    • Info remains consistent. However, the rounds computation might need to be optimized

Dissemination (tool providers, bakers…)

• Design document
• HLP
• Milestone
• Communication on the new RPC
• Communication on the other RPC changes
• Communication on the selection algorithm, upsides and downside (use Design doc as ref)
• Update Documentation and Changelog

Tests

• RPC calls: tezt/tests/RPC_test.ml
• Sampler
  • Correct length
  • No empty value (Signature.Public_key_hash.zero)
  • Craft a scenario (choose the values, number of bakers, etc...), and expect the exact result (computed by hand) (determinism test)
  • Check Variance between -1 and 1 (for round 0) (number of blocks per cycle = mainnet value)
• Credit
  • Try inject value >> Int64.max in storage, run the sampler, see the result
• get_baker
  • Compare Selected_bakers in storage with Validators RPC result
• Deactivated delegates
  • They have 0 credits (check multiple cycles)
  • They don't appear in the sampler result in the storage
• Run the whole test suite with the flag on
• Manual/Sandbox tests (checked with !20180)
  • Run a baker in sandboxed mode, bake some cycles
• Large scale tests
  • Testnet/Teztale
    • Benchmark performance

Future Cleanup

• Remove alias sampler
• Remove Seed
• Remove Feature Flag
• Remove unused storages