Skip to content

Latest commit

 

History

History
51 lines (47 loc) · 2.8 KB

TODO.md

File metadata and controls

51 lines (47 loc) · 2.8 KB

TODO

  • instead of Node ids, work with ip addresses and ports

  • fix send on closed channel bug with update calls with too short a context

  • implement validation of the sample by probing

  • move transports to sub-package

  • refactor core

    • reduce nr of public methods necessary for tests
    • formalize alive state
    • fix the myriad of race conditions on shared memory variables
    • test for concurrent access

  • fix bug that causes cores sample not to be stable over time

  • fix big that cause some deactivated cores to linger in the sample of others

  • decide, use and test an actual network transport

  • implement and test the agent

  • create and test a command that runs the agent as a process

  • (fix) bug that prevent the network to grow from 1 to 2

  • test if nodes can succesfully join by just pushing there ID

  • (fix) leave of node causes peers to keep probing on real agent proc

  • make validation and update timeouts configurable

  • (fix) remove core.IsActive lock contention

  • (fix) remove readView lock contention

  • (fix) pull easily runs into deadline exceeded

  • (fix) prevent pulling of nodes that were recently invalidated

  • refactor probing to not probe double samples

  • (fix) make sure its not possible for others to push others node's "self" info

  • finish basic agent implementation

  • add a general interface to the agent to dissemate messages

  • add a simple way to dissemate custom message in brahmsd

  • use the crypto hash for node hashing also for sampling instead of farm hash

  • probe only a part of the sampled nodes at a time

  • (fix) make sure probes (validation) returns early if all respond in time

  • implement a limited push with a small proof of work to prevent large scale push pollution (as per paper)

  • encrypt messages using https or asymetric encryption (as per paper)

  • randomly clear samples and reset with new seed data to better shield against adverary contamination

  • add a cellular consensus mechanism on sets

  • adjust l1 and l2 as the network grobs using an esimate as described here

  • store the node's sample on disk

  • measure if lock contention on sampler is too high

  • only full shutdown gossip agent if no messages arrive anymore

  • for probing only a subset, use a randomized order approach (Like SWIM) instead of random picking

When to refresh the view

Refreshing the view seems to be an open design decision.

  • Always refresh unless push is too big
  • Only refresh if there are pushed or pullsed nodes
  • Only refresh if there are both push and pulls

Always refresh makes sure it is the most up-to-date but causes the view to easily fluctuate. but thats were the history sample is for?