Discuss & plan moving to async-io

[elcritch]

@tbekas and myself were discussing the issue of blocking disk io in the client. This impacts both file and sqlite access. We'd put moving to a multithreaded io until after the testnet release.

Before working on this we need to have a couple of client team meetings specifically to discuss what and how to do this. Ideally we can share perspectives and get agreements on approaches and how to break it up and plan things.

Suggested points to discuss:

• multi-thread datastore vs multithread file-ops
• multi-threading sqlite
• implications for datastore update operation
• data models, specifically
  • using seqs vs SharedPtr types
  • implications refc vs orc, copying overhead

[tbekas]

Thanks for the summary, I think that solving this should be a priority. Even if the tradeoff is data-copying with refc.

[elcritch]

Blocked by https://github.com/codex-storage/nim-codex/issues/775

Essentially moving to non-blocking disk IO leads to lots of issues with QueryIterator and AsyncIterator's.