Project Planning July 25, 2018

Henry Till

Michael Birch (Unlicensed)

• Suspect that it is in the Java layer - worth performing testing in LMDB to ensure that all exceptions are propagated
• Observed that the log being provided to CASPER not always complete.  Have not yet isolated the log.
• It stands to reason if anything in LMDB can fail without throwing an exception or error.  The C layer has it's kind of error handling, and the Java layer will attempt to handle this by passing up Java exceptions. 
• Steps:
  • Comb through their issue tracker → Medha Parlikar (Unlicensed) - see if there are any issues  Assign to one of the RSpace team members.
  • Use Michael's test case, reduce the test case further. 
  • We should consider creating more tests for other areas in LMDB.
  • 2 pronged approach - 1 person goes and ferrets out the map size error and where it is happening.
  • 1 person goes and hunts for other sources of non-determinism.
    • Run Jeremy's perf test with some very high 10GB map size limit.
  • Nothing in the RSpace design is coupled to LMDB - FYI.  It would take approx a sprint to swap it out for another KVS. 
  • What is the status of the in-memory RSpace implementation?
    • It has been tested in the context of RSpace.  There are no tests for Casper /Rholang.  There are not higher level integration tests for it.  
    • Dom has been keeping it up to date with the current one.  It is available to us as well.
    • Definitely a good intermediate step before using another KVS.
    • No in memory history implementation.
    • Create a ticket to run with the in memory implementation & run Jeremy's script.
      • Swap out the normal store for RSpace in memory
      • If the test proves good
        • Create in memory trie store.  Lower priority.
  • Next steps
    • Run with Jeremy's test with an high map size, 10GB. Attempt to create anything that would non-determinism.
    • Task RSpace team to switch to the in memory store
    • Look at LMDB issue log
    • Audit projects that consume rspace to ensure that they are correctly handling exceptions coming from RSpace and below.

• These things in flight or are not started and are at risk due to lack of resources
  • Non-Determinism bugs are taking time to work through
  • Slashing work not started
  • Challenge/response protocol work not started.
  • Needs help from Rholang team if possible.
    • Rholang team will help starting next sprint.  Need to get some more convenience methods done.

• Cost accounting implementation appears to be on track
• Still some Rholang features for testnet that are not yet done.
  • • Convenience methods on collections - Daniyar is on it - List concatenation is the most important one.
    • String concatenation would also be helpful (Birch). 
    • Map and fold should also work on native lists (Butt)
      • Map operations would happen at a public name   
      • If there are features Consensus needs related to map and fold, this decision should be made soon
  • Registry work - without it - someone else can listen on your contract.  Open for a Denial of Service.
  • We could assume that the test net will not DOS each other.
  • Can we update the test net around end of September. 
    • Clients will need to update their code as a result of the update.  We will need to keep them prepared.
    • DECISION -Defer Registry until end of September.

• Clean up circular references in Communications between Kademlia & RChain protocol & transport
• Possible impacts to communications which will impede consensus testing.
• Work off of feature branch & test? 
• Signing messages with validator private key
  • Discussion done with Kyle earlier
  • Pawel to write up findings
• Concern: how many connections should node consistently keep running?
  • Raised with Nash. Waiting for response.
  • Pawel and Sebastian to write up to make a decision for how to understand this risk related to testnet and mainnet

• Read only nodes cannot make blocks.  Not running a CASPER instance with a validator identiy.
• They receive blocks but cannot produce blocks.  They cannot make their own blocks.
• Gossiping: Mike and Medha have discussed, and work is not in scope at this time