Crates.io | lightning-rapid-gossip-sync |
lib.rs | lightning-rapid-gossip-sync |
version | 0.0.125 |
source | src |
created_at | 2022-06-09 00:10:50.944189 |
updated_at | 2024-10-14 19:30:56.654489 |
description | Utility to process gossip routing data from Rapid Gossip Sync Server. |
homepage | |
repository | https://github.com/lightningdevkit/rust-lightning |
max_upload_size | |
id | 602310 |
size | 69,232 |
This crate exposes functionality for rapid gossip graph syncing, aimed primarily at mobile clients. Its server counterpart is the rapid-gossip-sync-server repository.
The (presumed) server sends a compressed gossip response containing gossip data. The gossip data is formatted compactly, omitting signatures and opportunistically incremental where previous channel updates are known.
Essentially, the serialization structure is as follows:
76, 68, 75
(the first three bytes are ASCII for LDK
)
u32
)default_cltv_expiry_delta
default_htlc_minimum_msat
default_fee_base_msat
default_fee_proportional_millionths
default_htlc_maximum_msat
(u64
, and if the default is no maximum, u64::MAX
)You will also notice that NodeAnnouncement
messages are omitted altogether as the node IDs are
implicitly extracted from the channel announcements and updates.
The data is then applied to the current network graph, artificially dated to the timestamp of the latest seen message less one week, be it an announcement or an update, from the server's perspective. The network graph should not be pruned until the graph sync completes.
In version 2 of the RGS protocol, node IDs may be followed by supplemental feature and socket address data. The presence of those additional fields is indicated by utilizing the unused bits of the 33-byte-pubkey parity byte as follows:
128 | 64 | 32 | 16 | 8 | 4 | 2 | 1 |
---|---|---|---|---|---|---|---|
Additional data | Reminder | Feature data | Feature data | Feature data | Address data | Always set | Odd y-coordinate |
Note that bit indices 3-5 all indicate feature data. Specifically, if none of the bits are set, that means there is no feature data that follows the pubkey. If a subset of them are set, the bit triplet is interpreted as an index (less one) of the default node features that were supplied prior. If all three bits are set, a custom feature combination is sent.
If there have been no changes to a node, bit index 6 can be set to function as a reminder absent any address or feature data.
Lastly, bit index 7 indicates the presence of additional data, which will allow forwards compatibility.
To achieve compactness and avoid data repetition, we're sending a significantly stripped down version of the channel announcement message, which contains only the following data:
channel_features
: u16
+ n
, where n
is the number of bytes indicated by the first u16
short_channel_id
: CompactSize
(incremental CompactSize
deltas starting from 0)node_id_1_index
: CompactSize
(index of node id within the previously sent sequence)node_id_2_index
: CompactSize
(index of node id within the previously sent sequence)For the purpose of rapid syncing, we have deviated from the channel update format specified in BOLT 7 significantly. Our custom channel updates are structured as follows:
short_channel_id
: CompactSize
(incremental CompactSize
deltas starting at 0)custom_channel_flags
: u8
update_data
Specifically, our custom channel flags break down like this:
128 | 64 | 32 | 16 | 8 | 4 | 2 | 1 |
---|---|---|---|---|---|---|---|
Incremental update? | Disable channel? | Direction |
If the most significant bit is set to 1
, indicating an incremental update, the intermediate bit
flags assume the following meaning:
64 | 32 | 16 | 8 | 4 |
---|---|---|---|---|
cltv_expiry_delta has changed |
htlc_minimum_msat has changed |
fee_base_msat has changed |
fee_proportional_millionths has changed |
htlc_maximum_msat has changed |
If the most significant bit is set to 0
, the meaning is almost identical, except instead of a
change, the flags now represent a deviation from the defaults sent at the beginning of the update
sequence.
In both cases, update_data
only contains the fields that are indicated by the channel flags to be
non-default or to have mutated.
The way a server is meant to calculate this rapid gossip sync data is by taking the latest time any change, be it either an announcement or an update, was seen. That timestamp is included in each rapid sync message, so all the client needs to do is cache one variable.
If a particular channel update had never occurred before, the full update is sent. If a channel has had updates prior to the provided timestamp, the latest update prior to the timestamp is taken as a reference, and the delta is calculated against it.
Depending on whether the rapid sync message is calculated on the fly or a snapshotted version is returned, intermediate changes between the latest update seen by the client and the latest update broadcast on the network may be taken into account when calculating the delta.
Given the primary purpose of this utility is a faster graph sync, we thought it might be helpful to provide some examples of various delta sets. These examples were calculated as of May 19th 2022 with a network graph comprised of 80,000 channel announcements and 160,000 directed channel updates.
Full sync | |
---|---|
Message Length | 4.7 MB |
Gzipped Message Length | 2.0 MB |
Client-side Processing Time | 1.4 s |
Week-old sync | |
---|---|
Message Length | 2.7 MB |
Gzipped Message Length | 862 kB |
Client-side Processing Time | 907 ms |
Day-old sync | |
---|---|
Message Length | 191 kB |
Gzipped Message Length | 92.8 kB |
Client-side Processing Time | 196 ms |