Zebra should try harder to discover and distribute good peer addresses

[teor2345]

Is your feature request related to a problem? Please describe.

Sometimes Zebra runs out of peer addresses, and stops working.

Describe the solution you'd like

Zebra should advertise its own inbound listener address:

• add its own address to GetAddr responses in the inbound service (Collect and send more accurate peer addresses #2123)
  • security: set the time to now, then sanitise it
  • security: add our address to the list, then shuffle and truncate
• send its own addresses to other nodes using unsolicited Addr messages
  • during the handshake with each new peer
    • https://developer.bitcoin.org/reference/p2p_networking.html#addr
    • https://github.com/ZcashFoundation/zebra/blob/main/zebra-network/src/peer/handshake.rs#L374
  • to all peers every 24 hours
    • I've seen this in a bitcoin wiki, but I can't find the reference right now

Zebra should regularly try to get new addresses:

• regularly make requests to the DNS seeders
  • set the request interval to longer than the typical DNS seeder TTL
  • security: DNS attacks can be used to pollute the node's peer set, so we shouldn't do this too often
• fanout peer crawling to multiple peers, to increase the number and diversity of addresses received (Fix a deadlock between the crawler and dialer, and other hangs #1950)
• from canonical addresses in Version messages (Collect and send more accurate peer addresses #2123)

Zebra currently uses gossiped outbound peer addresses, and detected inbound peer addresses. We could also add the address_from in received Version messages to our address book, but that could result in multiple connections to a single peer. To avoid these duplicate connections, we try new gossiped addresses before new canonical addresses (#2123).

[mpguerra]

Adding this to current sprint (9) as it seems we are working on it

[teor2345]

Adding this to current sprint (9) as it seems we are working on it

I implemented some of these features in #2123, to increase integration test coverage for #2120.

But the rest of these features aren't urgent, so I don't want to do any more work on this ticket during the current sprint.

[teor2345]

The current performance is good enough, we can open specific tickets if we need to.