General review

[andresberrios]

State Track Plugin

• General concerns
  • Performance
  • JSON serializing on the plugin
  • Making transaction processing take longer and hit CPU time limit
    • Only for transactions sent directly to this node?
    • Does it affect validated transactions/blocks received from the network?
  • Forks
  • How to induce and test recovery
  • Consensus
  • Exceptions
  • Other potential disruptions
  • Socket communications
  • Sending messages right now blocks until they are received (PUSH socket)
  • This is useful in case receiver goes down, since nodeos will wait for it to come back up
    • Instead of maintaining a local in-memory queue of messages that can build up too much if it carries on
  • Lost messages
    • When receiver pulls a message from the socket but then crashes
    • Implement message acknowledgement signal?
    • Implement an intermediate receiver that only manages the queue?
    • Implement a very safe received-messages queue in receiver to recover from if it crashes?
• Chainbase
  • Differences in exception handling between emit functions
  • Chain Link
  • Chainbase Fork Link
  • This line seems to have been just removed without a purpose Link
• Plugin
  • General code review
• Receiver
  • Remove undoer
  • Add Readme

[mmcs85]

1 - Since JSON serialization adds overhead on tx execution if handled directly.

1.a) first solution is to delegate to the receiver side for parsing it later.

     Upside is that serialization can be delayed until is needed.
     Downside is that you need to fetch abi for each (can be cached) op to be able to parse it.

1.b) second solution is to create a temporary queue that serializes all op to JSON after tx is done executing.

Upside delay stream of OP until a tx is done then stream it to receivers, it avoids streaming tx with exceptions and the undo ops that result from it with a proper filter.
Downside is that database op stream is not directly on tx when executing.

2 - (General concerns lost messages)

Is possible to use a ZFS system to create differential backups and use it as a generic snapshot system to revert to a given point in time nodeos and(or) mongodb.

3 - (chainbase)

3.a) emit function on chainbase is using a std::function instead of a signal. so it handles different exceptions.

I have been thinking in trying to support the signal in this case. Cuz I realized that it may be useful for multiple plugins enabled at once trying to subscribe this events.

3.b) the line is not removed is added later _indices.erase( itr );

[andresberrios]

Current pending changes

• [x] Signals implementation
• [x] Bring back line that was removed (stack function)
• [x] Exception handling in emit function should be the same as in Chain library
• [x] Check for refactoring opportunities in plugin code
• [x] Add more info to Readme

[mmcs85]

fixed some of the issues