Newer
Older
This document gives a brief, abstract description of main components that may be found in this repository.
If you want to see how we're using them to build Rococo <> Wococo (Kusama <> Polkadot) bridge, please
refer to the [Polkadot <> Kusama Bridge](./polkadot-kusama-bridge-overview.md).
This repo contains all components required to build a trustless connection between standalone Substrate chains,
that are using GRANDPA finality, their parachains or any combination of those. On top of this connection, we
offer a messaging pallet that provides means to organize messages exchange.
On top of that layered infrastructure, anyone may build their own bridge applications - e.g. [XCM messaging](./polkadot-kusama-bridge-overview.md),
[encoded calls messaging](https://github.com/paritytech/parity-bridges-common/releases/tag/encoded-calls-messaging) and so on.
## Terminology
Even though we support (and require) two-way bridging, the documentation will generally talk about
a one-sided interaction. That's to say, we will only talk about syncing finality proofs and messages
from a _source_ chain to a _target_ chain. This is because the two-sided interaction is really just the
one-sided interaction with the source and target chains switched.
The bridge has both on-chain (pallets) and offchain (relayers) components.
## On-chain components
On-chain bridge components are pallets that are deployed at the chain runtime. Finality pallets require
deployment at the target chain, while messages pallet needs to be deployed at both, source
and target chains.
### Bridge GRANDPA Finality Pallet
A GRANDPA light client of the source chain built into the target chain's runtime. It provides a "source of truth"
about the source chain headers which have been finalized. This is useful for higher level applications.
The pallet tracks current GRANDPA authorities set and only accepts finality proofs (GRANDPA justifications),
generated by the current authorities set. The GRANDPA protocol itself requires current authorities set to
generate explicit justificaion for the header that enacts next authorities set. Such headers and their finality
proofs are called mandatory in the pallet and relayer pays no fee for such headers submission.
The pallet does not require all headers to be imported or provided. The relayer itself chooses which headers
he wants to submit (with the exception of mandatory headers).
More: [code](../modules/grandpa/).
### Bridge Parachains Finality Pallet
Parachains are not supposed to have their own finality, so we can't use bridge GRANDPA pallet to verify their
finality proofs. Instead, they rely on their relay chain finality. The parachain header is considered final,
when it is accepted by the [`paras` pallet](https://github.com/paritytech/polkadot/tree/1a034bd6de0e76721d19aed02a538bcef0787260/runtime/parachains/src/paras)
at its relay chain. Obviously, the relay chain block, where it is accepted, must also be finalized by the relay
chain GRANDPA gadget.
That said, the bridge parachains pallet accepts storage proof of one or several parachain heads, inserted to the
[`Heads`](https://github.com/paritytech/polkadot/blob/1a034bd6de0e76721d19aed02a538bcef0787260/runtime/parachains/src/paras/mod.rs#L642)
map of the [`paras` pallet](https://github.com/paritytech/polkadot/tree/1a034bd6de0e76721d19aed02a538bcef0787260/runtime/parachains/src/paras).
To verify this storage proof, the pallet uses relay chain header, imported earlier by the bridge GRANDPA pallet.
The pallet may track multiple parachains at once and those parachains may use different primitives. So the
parachain header decoding never happens at the pallet level. For maintaining the headers order, the pallet
uses relay chain header number.
More: [code](../modules/parachains/).
The pallet is responsible for queuing messages at the source chain and receiving the messages proofs at the
target chain. The messages are sent to the particular _lane_, where they are guaranteed to be received in the
same order they are sent. The pallet supports many lanes.
The lane has two ends. Outbound lane end is storing number of messages that have been sent and the number of
messages that have been received. Inbound lane end stores the number of messages that have been received and
also a map that maps messages to relayers that have delivered those messages to the target chain.
The pallet has three main entrypoints:
- the `send_message` may be used by the other runtime pallets to send the messages;
- the `receive_messages_proof` is responsible for parsing the messages proof and handing messages over to the
dispatch code;
- the `receive_messages_delivery_proof` is responsible for parsing the messages delivery proof and rewarding
relayers that have delivered the message.
Many things are abstracted by the pallet:
- the message itself may mean anything, the pallet doesn't care about its content;
- the message dispatch happens during delivery, but it is decoupled from the pallet code;
- the messages proof and messages delivery proof are verified outside of the pallet;
- the relayers incentivization scheme is defined outside of the pallet.
Outside of the messaging pallet, we have a set of adapters, where messages and delivery proofs are regular
storage proofs. The proofs are generated at the bridged chain and require bridged chain finality. So messages
pallet, in this case, depends on one of the finality pallets. The messages are XCM messages and we are using
XCM executor to dispatch them on receival. You may find more info in [Polkadot <> Kusama Bridge](./polkadot-kusama-bridge-overview.md)
document.
More: [code](../modules/messages/).
The pallet is quite simple. It just registers relayer rewards and has an entrypoint to collect them. When
the rewards are registered and the reward amount is configured outside of the pallet.
More: [code](../modules/relayers/).
Offchain bridge components are separate processes, called relayers. Relayers are connected both to the
source chain and target chain nodes. Relayers are reading state of the source chain, compare it to the
state of the target chain and, if state at target chain needs to be updated, submits target chain
transaction.
The task of relay is to submit source chain GRANDPA justifications and their corresponding headers to
the Bridge GRANDPA Finality Pallet, deployed at the target chain. For that, the relay subscribes to
the source chain GRANDPA justifications stream and submits every new justification it sees to the
target chain GRANDPA light client. In addition, relay is searching for mandatory headers and
submits their justifications - without that the pallet will be unable to move forward.
Svyatoslav Nikolsky
committed
More: [GRANDPA Finality Relay Sequence Diagram](./grandpa-finality-relay.html), [pallet level documentation and code](../relays/finality/).
The relay connects to the source _relay_ chain and the target chain nodes. It doesn't need to connect to the
tracked parachain nodes. The relay looks at the [`Heads`](https://github.com/paritytech/polkadot/blob/1a034bd6de0e76721d19aed02a538bcef0787260/runtime/parachains/src/paras/mod.rs#L642)
map of the [`paras` pallet](https://github.com/paritytech/polkadot/tree/1a034bd6de0e76721d19aed02a538bcef0787260/runtime/parachains/src/paras)
in source chain, and compares the value with the best parachain head, stored in the bridge parachains pallet at
the taget chain. If new parachain head appears at the relay chain block `B`, the relay process **waits**
until header `B` or one of its ancestors appears at the target chain. Once it is available, the storage
proof of the map entry is generated and is submitted to the target chain.
As its on-chain component (which requires bridge GRANDPA pallet to be deployed nearby), the parachains
finality relay requires GRANDPA finality relay to be running in parallel. Without it, the header `B` or
any of its children's finality at source won't be relayed at target, and target chain
won't be able to verify generated storage proof.
More: [Parachains Finality Relay Sequence Diagram](./parachains-finality-relay.html), [code](../relays/parachains/).
Messages relay is actually two relays that are running in a single process: messages delivery relay and
delivery confirmation relay. Even though they are more complex and have many caveats, the overall algorithm
is the same as in other relays.
Message delivery relay connects to the source chain and looks at the outbound lane end, waiting until new
messages are queued there. Once they appear at the source block `B`, the relay start waiting for the block
`B` or its descendant appear at the target chain. Then the messages storage proof is generated and submitted
to the bridge messages pallet at the target chain. In addition, the transaction may include the storage proof
of the outbound lane state - that proves that relayer rewards have been paid and this data (map of relay
accounts to the delivered messages) may be pruned from the inbound lane state at the target chain.
Delivery confirmation relay connects to the target chain and starts watching the inbound lane end. When new
messages are delivered to the target chain, the corresponding _source chain account_ is inserted to the
map in the inbound lane data. Relay detects that, say, at the target chain block `B` and waits until that
block or its descendant appears at the source chain. Once that happens, the relay crafts a storage proof of
that data and sends it to the messages pallet, deployed at the source chain.
As you can see, the messages relay also requires finality relay to be operating in parallel. Since messages
relay submits transactions to both source and target chains, it requires both _source-to-target_ and
_target-to-source_ finality relays. They can be GRANDPA finality relays or GRANDPA+parachains finality relays,
depending on the type of connected chain.
Svyatoslav Nikolsky
committed
More: [Messages Relay Sequence Diagram](./messages-relay.html), [pallet level documentation and code](../relays/messages/).
Every relay transaction has its cost. The only transaction, that is "free" to relayer is when the mandatory
GRANDPA header is submitted. The relay that feeds the bridge with every relay chain and/or parachain head it
sees, will have to pay a (quite large) cost. And if no messages are sent through the bridge, that is just
waste of money.
We have a special relay mode, called _complex relay_, where relay mostly sleeps and only submits transactions
that are required for the messages/confirmations delivery. This mode starts two message relays (in both
directions). All required finality relays are also started in a special _on-demand_ mode. In this mode they
do not submit any headers without special request. As always, the only exception is when GRANDPA finality
relay sees the mandatory header - it is submitted without such request.
The message relays are watching their lanes and when, at some block `B`, they see new messages/confirmations
to be delivered, they are asking on-demand relays to relay this block `B`. On-demand relays does that and
then message relay may perform its job. If on-demand relay is a parachain finality relay, it also runs its
own on-demand GRANDPA relay, which is used to relay required relay chain headers.
More: [Complex Relay Sequence Diagram](./complex-relay.html), [code](../relays/bin-substrate/src/cli/relay_headers_and_messages/).