note.txt

I decided that this is too complex. Tracker will be pure DHT based.
* mind long-term and part-of announcements
* localized DHT branch
* optionaly store and verify signature of mutable link
* mind ephemeral Node IDs


Groups, Swarms, Download, Fetch, Sources and DHT.

Fetch is the Encapsulation of all the sub-classes. One FetchJob is started per
ObjectID. Sources can be added to a fetch as:
  - directly address
  - group ID/reference

Fetch class is used to select best peer to download from. It uses data from
other classes, but remembers the failed sources to not retry them.
  
First a DHT lookup is performed, then the closest peers (by RTT or Hop) are
queried for better peers in few iteretions.

This lookup is performed by a PeerGroup class. Only one instance per GroupID
should be created, cached and reused for multiple objects. PeerGroup is
observable and can notify multiple observers for new peers. The peers are
always kept sorted by distance (Hop, RTT).

Peer addresses can be added to Fetch, so it must have a list for peer refs
outside of groups.

Fetch:
  otclient
  otconnection
  outputstream
  observers
  groups
  direct peers
  failed peers

SourceSelect(grop[],direct[],blacklist[])

Group instance is initialized for specific group ID. Group contains peer
addresses and DHT-IDs. Both can be added manualy, but the group id is present
by default. When no usable peer is present, a DHT lookup is performed to next
DHT-ID. Query type is GroupMembersQuery or dhtGetNode, depending on flag.
Othervise a GetCloserMembers query is sent to closest usable peer to get more,
possibly better peers.

Results of Group lookup are used after few iterations to ensure good peer
selection. Subsequent downloads (or retries) from the same group can get better
peers (the lookup is running in background).

Peer queries:
  * DHT based
    * trackerGetPeers
  * non-dht
    * direct
    * indirect

Have list of tried nodes for each DHT bucket.  When rcv new node for that bucket
and not on tried list, send beat. Bucket has few (2) slots for hop-near nodes,
where closer nodes are prefered. Normal slots prefer older nodes.