p2p sync dead peer handoff

[pmesnier]

Relates to #291. When a new peer node is added to the network, it must be quickly brought up to the head of the chain with the rest of the nodes. As described in #344, this is done by requesting smallish chunks of blocks from many peers, and caching these until they can be assembled into a local clone of the chain.

But if one of those peers dies while delivering a chunk? The solution is to simply transfer the chunk to another peer and request that peer finish supplying the missing blocks. There is a function used to generate new chunk requests, all that is needed here is to push any partially filled chunks onto a queue and hand these out to the peers as they are able to take more requests.

[pmesnier]

I am still working this issue, but concurrently working to fix up synchronization issues that have cropped up today. To the solution of #408, I've defined a "sync_manager" class that simplifies the task of distributing sections of historical blocks, and is able to transfer responsibility for a section from one peer to another.

[pmesnier]

daily status (a bit delayed) : Transfer of sync request from a dead peer to another works fine. Adding a timer detecting slow peers. working through the implications of that scenario.

[pmesnier]

Timer work done, Syncing now tolerates both dead peers and peers that are just slow. Need to add debug-driver code so that these conditions may be reliably tested. Also, the sync only takes the client up to the current last irreversible block. Still need to add the support for sending notices and fetching remaining blocks and pending transactions. Not quite ready for a PR just yet.

[pmesnier]

taking off the in progress label for now.

[pmesnier]

Addressed by PR #500. Waiting for review. It's a big one.

[pmesnier]

Closing, PR #500 resolves this issue.