Web socket syncing

[tdroxler]

Resolves: #353

After thousands of tests and adjustment I think we can start the discussion, it still needs some unit tests, but I'm not sure yet how easy it will be to test the websockets.

The PR needs https://github.com/alephium/dev-alephium/pull/827 from the node, because we need the events to match the data we receive from rest calls.

The strategy is:

• Sync with BlockFlowSyncService as usual
• Once we are up-to-date: latest block is less than 30s old we move to
  websocket syncing
• If websocket messages are late: > 30s old, we close the websocket and
  move back to http syncing. This can happen in case of network issue,
  this is safer as we could suddenly receive lots of message through the
  websocket and the DB can't follow (happen once in my tests)
• If the websocket close for anyreason we also move back to http
  syncing.

The main idea is to always rely and fall back on our BlockFlowSyncService which is well tested in production.

I tried various edge cases, like cutting my network etc.

[simerplaha]

Neat! We have WebSockets! Is the way to test this is to do a full sync and then manually terminate the server to test how it behaves?

Tricky one to test. I'm letting it sync overnight and kinda manually causing abrupt server shutdowns to see how the connection & sync behaves, with debug statements, if I'm not careful enough it corrupts my RocksDB instance and I have to start over.

Surely there is an easier way to test. Any suggestion?

it still needs some unit tests, but I'm not sure yet how easy it will be to test the websockets.

Yep, tricky one indeed.

[tdroxler]

Tricky one to test. I'm letting it sync overnight and kinda manually causing abrupt server shutdowns to see how the connection & sync behaves, with debug statements, if I'm not careful enough it corrupts my RocksDB instance and I have to start over.

If you shutdown the node, then the explorer-backend should also stop, but it sounds super weird that the node's rockDB get corrupted, the explorer-backend shouldn't have any influence on the node.

how are you causing the abrupt shutdown? I could try to reproduce

[simerplaha]

Oh it's not a bug. It's me intentionally causing abrupt/force shutdown to see how the connection behaves with println debug statement. Node and explorer behave as expected. Not sure how else to test it. Was trying to convey that testing is tricky, as you've pointed out.