-
Notifications
You must be signed in to change notification settings - Fork 1.6k
Add a Statement Gossip Subsystem section to the implementors guide #1151
Changes from 1 commit
File filter
Filter by extension
Conversations
Jump to
Diff view
Diff view
- Loading branch information
There are no files selected for viewing
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -989,6 +989,35 @@ Dispatch a `PovFetchSubsystemMessage(relay_parent, candidate_hash, sender)` and | |
|
|
||
| (TODO: send statements to Statement Distribution subsystem, handle shutdown signal from candidate backing subsystem) | ||
|
|
||
| ### Statement Gossip Subsystem | ||
|
|
||
| The statement gossip subsystem is responsible for gossiping out and receiving [Signed Statements](#Signed-statement-type) from validators. | ||
|
|
||
| This subsystem communicates with the overseer via [two message types](#Candidate-Backing-Subsystem-Messages), one for overseer -> subsystem and another for subsystem -> overseer. Currently this just differentiates between statements that the subsystem has been instructed to gossip out, and statements that the subsystem has received. | ||
|
|
||
| The subsystem needs to contain a handle to a gossiping engine to gossip and recieve messages. This will be cloned into each job that is spawned. Apart from that, it also contains the general structures that all subsystems contain, e.g. channels to communicate with the overseer and handles to spawned jobs in order to shut them down. | ||
|
Contributor
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Why clone the Does the above make sense? I am happy to expand further on this.
Contributor
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Regardless, it seems reasonable for the gossip engine to have a single owner (the subsystem). A job for this subsystem would only entail stringing together |
||
|
|
||
| On `OverseerSignal::StartWork` it should: | ||
| * Spawn a job and pass in `relay_parent`, a clone of the gossiping engine and a handle to a message validator. | ||
| * Send out a neighbour packet with the `relay_parent` added to the list of accepted chain heads. | ||
|
Contributor
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. In case you agree with me that the neighbor packet should be sent via the If there is no agreement I can expand on why this would help the architecture.
Contributor
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. One thing we need to be aware of is that the validator needs to be aware of the most recent neighbor packets we have sent to peers. I suppose we just need to be careful when the |
||
|
|
||
| On `OverseerSignal::StopWork` it should: | ||
| * Stop the job via the job handle. | ||
| * Send out a neighbour packet with the job's `relay_parent` removed. | ||
|
|
||
| On `StatementGossipSubsystemMessageIn::StatementToGossip` it should: | ||
| * Send the signed statement to the job running for the `relay_parent`. | ||
|
|
||
| The statement gossip job needs to poll two seperate futures (as well as the exit signal): | ||
|
Contributor
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Why not make the statement gossip job a pure state machine that gets events as input and outputs commands to either be forwarded to the Thus business logic (Polkadot specific) would stay pure within a single state machine and I/O logic would stay within the |
||
|
|
||
| * A future that takes the passed-in statements, validates them and gossips them along with the `relay_parent` hash using the gossip engine. | ||
| * A future that takes the messages that the gossip engine receives for the `relay_parent`, validates them and sends to the subsystem. | ||
|
|
||
| I have a basic implementation of this code on the [`ashley-test-statement-gossip-subsystem`](https://github.com/paritytech/polkadot/tree/ashley-test-statement-gossip-subsystem) branch. | ||
|
Contributor
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. In case you would like to have input on this, would you mind opening up a pull request? That would make discussions a lot easier. |
||
|
|
||
| (TODO: Do we need a message type for sending a statemen directly to a peer?) | ||
| (TODO: We probably need to account for backpressure) | ||
|
|
||
| --- | ||
|
|
||
| [TODO: subsystems for gathering data necessary for block authorship, for networking, for misbehavior reporting, etc.] | ||
|
|
@@ -1072,7 +1101,7 @@ enum OverseerSignal { | |
| ``` | ||
|
|
||
|
|
||
| #### Candidate Backing subsystem Message | ||
| #### Candidate Backing Subsystem Message | ||
|
|
||
| ```rust | ||
| enum CandidateBackingSubsystemMessage { | ||
|
|
@@ -1085,6 +1114,22 @@ enum CandidateBackingSubsystemMessage { | |
| } | ||
| ``` | ||
|
|
||
| ### Statement Gossip Subsystem Messages | ||
|
|
||
| ```rust | ||
| enum StatementGossipSubsystemIn { | ||
| /// A statement to be gossiped out to validators. | ||
| StatementToGossip { relay_parent: Hash, statement: SignedStatement } | ||
| } | ||
| ``` | ||
|
|
||
| ```rust | ||
| enum StatementGossipSubsystemOut { | ||
| /// A statement that we've received from a validator. | ||
| StatementReceived { relay_parent: Hash, statement: SignedStatement } | ||
| } | ||
| ``` | ||
|
|
||
| #### Host Configuration | ||
|
|
||
| The internal-to-runtime configuration of the parachain host. This is expected to be altered only by governance procedures. | ||
|
|
||
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Did you meant to reference the singular?