hyperlane-monorepo

The home for Hyperlane core contracts, sdk packages, and other infrastructure

blockchain hyperlane interchain monorepo

History

Daniel Savu 6e39cf01af Relayer retry endpoints (#3476 ) ### Description Adds a `/message_retry` relayer endpoint, which supports retrying OpQueue operations either by ID or by destination domain. Such a request will be sent to an `mpmc` channel's receiving end in the OpQueue, which is emptied when `OpQueue::pop` is called. Example calls: ``` GET http://127.0.0.1:60843/message_retry?destination_domain=42 GET http://127.0.0.1:60843/message_retry?message_id=0x46910b1329ee53c86a023b322e9ca1c17e5f9f0bee789c77b0abced0a173d714 ``` If the endpoint is called specifying both filters, two requests will be sent across the channel, one for each condition. Efficiency note: The entire queue is iterated over if there's at least one retry request in the channel. The good thing is that rebuilding the heap is `O(n)` so not too bad. ### Drive-by changes - Removed the usage of `enum_dispatch`, because it would have made unit tests messy. The drawback is that we have to work around object safety restrictions in `PendingOperation` - for instance it can't be cloned (but can be stored behind an `Arc` if we ever need this). - Added new methods to `PendingOperation`, either to match by ID, or to implement heap element prioritization without depending on the concrete `PendingMessage` type (as was done before). Additions: `id()`, `origin_domain()`, `priority()`. Don't think this conflicts with any operations we may add in the future (e.g. gas oracle updates). - Although `tokio`'s broadcast channel is multi-producer-multi-consumer by default, the consumer end isn't `Clone` - you instead need to have a producer to call `.subscribe()` on to get a new consumer, so I added an `MpmcChannel` struct to encapsulate keeping a producer around to get new consumers. - OpQueue is moved into its own file and has unit tests added for retries (covering broadcast to multi-queue, and both retry types). - Did a bit of agent `server `cleanup, including `EigenNodeApi` ### Related issues - Fixes https://github.com/hyperlane-xyz/hyperlane-monorepo/issues/3098 ### Backward compatibility Yes ### Testing Unit tests for the server route (channel transmitter logic) and for the OpQueue retries (channel receiver logic).	8 months ago
..
src	Relayer retry endpoints (#3476 )	8 months ago
Cargo.toml	Cosmos block-by-block indexing (#3245 )	9 months ago

### Description

Adds a `/message_retry` relayer endpoint, which supports retrying
OpQueue operations either by ID or by destination domain. Such a request
will be sent to an `mpmc` channel's receiving end in the OpQueue, which
is emptied when `OpQueue::pop` is called.

Example calls:
```
GET http://127.0.0.1:60843/message_retry?destination_domain=42
GET http://127.0.0.1:60843/message_retry?message_id=0x46910b1329ee53c86a023b322e9ca1c17e5f9f0bee789c77b0abced0a173d714
```

If the endpoint is called specifying both filters, **two** requests will
be sent across the channel, one for each condition.


Efficiency note: The entire queue is iterated over if there's at least
one retry request in the channel. The good thing is that rebuilding the
heap is `O(n)` so not too bad.

### Drive-by changes

- Removed the usage of `enum_dispatch`, because it would have made unit
tests messy. The drawback is that we have to work around object safety
restrictions in `PendingOperation` - for instance it can't be cloned
(but can be stored behind an `Arc` if we ever need this).
- Added new methods to `PendingOperation`, either to match by ID, or to
implement heap element prioritization without depending on the concrete
`PendingMessage` type (as was done before). Additions: `id()`,
`origin_domain()`, `priority()`. Don't think this conflicts with any
operations we may add in the future (e.g. gas oracle updates).
- Although `tokio`'s broadcast channel is multi-producer-multi-consumer
by default, the consumer end isn't `Clone` - you instead need to have a
producer to call `.subscribe()` on to get a new consumer, so I added an
`MpmcChannel` struct to encapsulate keeping a producer around to get new
consumers.
- OpQueue is moved into its own file and has unit tests added for
retries (covering broadcast to multi-queue, and both retry types).
- Did a bit of agent `server `cleanup, including `EigenNodeApi`


### Related issues

- Fixes https://github.com/hyperlane-xyz/hyperlane-monorepo/issues/3098

### Backward compatibility

Yes

### Testing

Unit tests for the server route (channel transmitter logic) and for the
OpQueue retries (channel receiver logic).

8 months ago

src

Relayer retry endpoints (#3476 )

8 months ago

Cargo.toml

Cosmos block-by-block indexing (#3245 )

9 months ago