Enable block archival sync

[antiochp]

Archival Peers (as of 20210315) -

• 78.47.168.92:3414
• 46.101.247.184:3414
• 3.81.230.109:3414
• 23.124.76.209:3414

# configure the above as peers_preferred to help archive node connectivity
peers_preferred = ["78.47.168.92:3414", "46.101.247.184:3414", "3.81.230.109:3414", "23.124.76.209:3414"]

---

Two relatively minor changes that allow a node to run in "block archival mode" and sync fully from other archival peers.
This works in conjunction with various related PRs recently merged.

1. Simply return false for check_txhashset_needed() if running with archive_mode enabled
2. No longer skip compaction during sync (compact periodically during full archival sync).

---

The local chain_data dir should be cleared out prior to a full archival sync - the node will sync from scratch.

A node will only be able to sync historical blocks from other archival nodes.
The best way of configuring the local node to peer with archival nodes is via the following config.

#
# with [server] section -
#

archive_mode = true


#
# within [server.p2p_config] section -
# 

# These are known archival nodes. Sync works best if these are both "preferred" and "allow".
peers_preferred = ["85.10.201.143:3414", "49.12.108.173:3414", "176.9.86.219:3414", "116.203.154.223:3414", "213.239.217.14:3414", "46.101.247.184:3414"]
peers_allow = ["85.10.201.143:3414", "49.12.108.173:3414", "176.9.86.219:3414", "116.203.154.223:3414", "213.239.217.14:3414", "46.101.247.184:3414"]

---

Once the node has completed an archival sync we can update the config to allow other nodes to connect.

# No explicit peers_allow list, so all peers implicitly allowed.
# peers_allow = ["85.10.201.143:3414", "49.12.108.173:3414", "176.9.86.219:3414", "116.203.154.223:3414", "213.239.217.14:3414", "46.101.247.184:3414"]

---

Full archival sync takes several hours to complete, probably close to 24 hours.
On completion the local node will have full block history from height 1 onwards.

Disk usage is approx 9.5GB for a full archival node.

---

It is anticipated that peers_preferred will be a required config for archival nodes, to ensure we good good connectivity.

The additional peers_allow config is a temporary solution while we test this out prior to having nodes correctly advertising archival capabilities.

---

TODO

• uncomment the BLOCK_HIST capability filtering (we only sync from other archival nodes)
• collect a list of known good ip addresses for existing archival nodes so we can publish a list of preferred peers for bootstrapping
• release this as part of 5.1.0

[antiochp]

If we have consensus that this is something we want to roll out to mainnet then I propose we consider a 5.1.0 release in the next couple of weeks. This would include this change along with other changes currently on master.

The "bootstrapping" change (discussed above) will ensure network traffic and peering is unaffected for non-archive nodes (archival sync restricted to known archive peers).

[antiochp]

I understand @quentinlesceller is experiencing really slow archive sync.
This appears limited to some local nodes as we are not seeing this across every archive node. Possibly slow disk or some other local environmental factor.
Solving that is outside the scope of this PR - we can (and will) investigate and optimize in a subsequent PR.

[bladedoyle]

I completed a sync on bare metal i9 cpu + ssd in just over 38 hours.
ending disk use: 7.9 GB
Everything looks good to me.

[antiochp]

For reference a $5/month Digital Ocean instance sync'd in about 20 hours.
So maybe bandwidth is a bigger factor than disk speed.

[bladedoyle]

My test sync was done with 1 GB fiber...
None of the peers data xfer rate ever got above about 12K/sec, including a local full node peer.

ftr: the sync started with 3 peers, gained peers over the first few hours, and ended with (currently has) 7 peers.

one more note: my test sync did not allow inbound connections, all outbound

[antiochp]

My test sync was done with 1 GB fiber...
None of the peers data xfer rate ever got above about 12K/sec, including a local full node peer.

ftr: the sync started with 3 peers, gained peers over the first few hours, and ended with (currently has) 7 peers.

one more note: my test sync did not allow inbound connections, all outbound

Is this with allow and preferred lists configured against known archival nodes?

[antiochp]

Going to merge this to master.
Anybody wanting to run this code with archive_mode=true enabled is advised to specify an explicit set of known archive nodes in their peers_preferred config.