les: unresponsive server up until sync is done

[farazdagi]

System information

Geth version: 1.5.8-stable
OS & Version: OSX
Commit hash : f58fb322

Expected behavior

• RPC server is expected to be available during header syncing

Actual behaviour

• Everything blocks up until sync is done

Steps to reproduce the behaviour

• While this issue was first found when we tried to run eth.* commands using web3, this issue can be reproduced in console as well
• Basically, geth --light --testnet console, then once sync starts, you cannot call any commands (they block), you can't event dump eth object by typing eth ENTER:
  

More Notes

• the issue is introduced by this commit: c57c54c (I've tested code prior to that commit - no blocking exists there)
• the very same or really similar issue was already accepted and fix was provided by les: fix private net issues, enable adding peers manually again #3607. From description of that PR:

removes the mechanism which delayed starting the LES server side protocol until ETH was fully synced. This mechanism solved a problem that does not exist anymore, clients now check server's reported head before sending requests and choose a server that can actually serve it. At the same time it caused other problems, like not starting LES server at all when the node was mining alone on a private net. Also, not starting the protocol stalled connections to other LES servers through ETH, even though they did not want to do anything with each other through LES.

• probably some aspects of the problem are solved with that PR, but the problem reported in this issue (that's blocking calls to eth - using RPC, via console or web3) is still there

Severity

• This issue is really severe. Blocking/delaying light client server up until full sync (especially on mobile devices where we use LES, and where sync is a bit slower) is huge concern, as it prevents users from using our app up until sync is done

[0xc1c4da]

Related PR #3660

[karalabe]

Should be fixed on 1.6.0.