spv: After losing all peers synced is never true.

[JoeGruffins]

After losing all peers, even if you get them back, synced from the rpc command syncstatus never becomes true.

Can be seen on mainnet:

1. Start up an spv wallet on mainnet.
2. After syncstatus returns true for synced turn off your internet.
3. Wait for getpeerinfo to return an empty list.
4. Start your internet again.
5. Check syncstatus after getting peers back and waiting a bit.

Also, if you have logs on trace when turning off peers, you will see the spinning described in #2433

[davecgh]

I'm sure @jrick will know better since he's far more familiar with this code and works on it, but I imagine it has to wait for a new block to know it's synced again.

Did you wait long enough to know if that's the case or not?

[JoeGruffins]

Yes. I have waited a few blocks. The change in block number is reflected in getinfo but synced stays false.

[jrick]

Yeah, I can see how this is occurring. The SPV syncer never does complete teardown (where the main package could recreate the syncer all over again in a loop) when all peers are lost.

This is indicative of a larger issue in this design: we almost certainly want to treat reconnection after all peers drop the same as initial sync, but we only perform initial sync once per syncer instance. I think tearing down the entire syncer (with Run() returning an error) is the best way to work around this problem.

[jrick]

Please test #2465 to see if it fixes this issue.

I believe this should also fix the cpu spinning as well.