-
Notifications
You must be signed in to change notification settings - Fork 248
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
add reconnecting tests for unstable_backend #1765
Conversation
|
||
type RpcResult<T> = Result<T, RpcError>; | ||
type Item = RpcResult<String>; | ||
pub type SubscriptionHandler = Box< |
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.
quite complicated type and a bit hard to read but I'm fine to remove the lifetimes and cloning the data for each call if it simplifies the code...
it just tests anyway
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.
i dont think cloning would help as we need mutable refs to mutate the data inside, but if you have an idea how to simplify it, please suggest
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.
It's fine to keep this, I thought it was possible to just clone the Arc<Mutex<_>>
but it's one layer above this...
|
||
struct MockDataTable { | ||
items: HashMap<Vec<u8>, VecDeque<Item>>, | ||
pub enum Message<T> { |
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.
dq: why is many Result<T, Error> and a single just T?
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.
when we load initial mock_data its concrete types like FollowEvent
but inside mocked rpc its just Item
which is Result<String,RpcError>
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.
nice work, I have mostly have comments to simplify the tests to make it a little easier to read.
I'll take a closer look at the test themselves in the next iteration
#[serde(deserialize_with = "unsigned_number_as_string::deserialize")] | ||
#[cfg_attr( | ||
test, | ||
serde(serialize_with = "unsigned_number_as_string::for_test::serialize") |
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.
aight, I see makes sense to have these in this module then
Missed the serialize_with
pub struct TransactionBlockDetails<Hash> { | ||
/// The block hash. | ||
pub hash: Hash, | ||
/// The index of the transaction in the block. | ||
#[serde(with = "unsigned_number_as_string")] | ||
#[serde(deserialize_with = "unsigned_number_as_string::deserialize")] |
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.
ok, deserialize is only implemented for this type and with
is a shortcut for both serialize and deserialize impl
("continue_response", Message::Single(Ok(()))), | ||
("continue_response", Message::Single(Ok(()))), | ||
]; | ||
let mock_data = vec![ |
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.
It's not really easy for me to follow how the responses is linked the "follow_event/subscription data`,
So I wonder whether we provide something like:
(
method_response(id1),
Message::Many(vec![Ok(storage_items("Id1", &[storage_result("ID1", "Data1")])), Ok(operation_continue("Id1"))])
)
where the "first item" is the method_response i.e, operationStarted and the "second item" is the subscription data i.e, follow_events on provided on that method response/operation ID.
That way it's simpler to understand the responses for each call but it may not work that well how you wrote the tests i.e, by pop the data unless it inserted with some unique id
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.
hmm, given we know the exact ordering of the requests/responses/subscription messages i can try to refactor it to be one single stack of responses that will be popped one by one. But maybe dong this in a followup PR would be more appropriate?
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.
Yeah it's fine to keep it out of this PR, let's hear what James and Alex thinks about this as well :)
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.
Yep it would be a bit easier to follow the test with Niklas suggestion. If it turns out too complicated to change it we can do a followup for sure :D
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.
Ok, so after a bit of experimenting it looks like it's not possible without adding an external dependency or adding mock data item by item to the mock data set. using dyn Serialize
is not possible(not object safe) and adding another dependency just to make tests seem a bit more pretty feels like an overkill.
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.
Great,
Minor thingy, whether we can improve the readability for the tests by linking responses and subscription messages together but it's not a blocker for this PR IMO.
"chainHead_v1_storage", | ||
Message::Many(Ok(vec![Ok(operation_error("Id1")), Ok(FollowEvent::Stop)])), | ||
)]; | ||
let rpc_client = setup_mock_rpc_client(false) |
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.
Nice! The API looks quite elegant!
subxt/src/backend/mod.rs
Outdated
#[tokio::test] | ||
/// Tests the error behaviour in streams based on chainHead_v1_continue calls |
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.
nit: new line
) | ||
.await | ||
.unwrap(); | ||
// operation returned FollowEvent::OperationError |
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.
dq: Does subxt retry on OperationError?
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.
No, all errors except DisconnectWillReconnect and Rejected
are handed back to the user to deal with.
@@ -745,11 +769,16 @@ pub enum TransactionStatus<Hash> { | |||
|
|||
/// Details of a block that a transaction is seen in. | |||
#[derive(Debug, Clone, PartialEq, Eq, Deserialize)] | |||
#[cfg_attr(test, derive(Serialize))] |
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.
Should we just expose a derive(Serialize)
instead of placing it under a cfg(test)? That might turn out to be useful for some developers, wdyt?
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.
I'm not super opposed to this! If we do choose to do it though then we should make sure to be consistent and apply it on all related types and not just those used in tests (if any were missed!)
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.
LGTM! Tiny comments around test styles. Maybe we can expose the derive(Serialize)
implementation for some structures and we can think about grouping test data, but those can be a followup 👍
subxt/src/backend/mod.rs
Outdated
Box::new(|data, _sub, params| { | ||
Box::pin(async move { | ||
let params = params.map(|p| p.get().to_string()); | ||
let rpc_params = jsonrpsee::types::Params::new(params.as_deref()); | ||
let key: sp_core::Bytes = rpc_params.sequence().next().unwrap(); | ||
let value = data.pop(key.0).unwrap_single(); | ||
value.map(|v| serde_json::value::RawValue::from_string(v).unwrap()) | ||
}) | ||
}), |
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.
I wonder how easily we could remove the Box's here to make the API cleaner.
Eg if add_method
and add_subscription
took in some generic like:
fn add_subscription(sub: &str, f: F)
where
Fn: Send + Fn(
&'a mut MockDataTable,
&'a mut Option<Subscription>,
Option<Box<serde_json::value::RawValue>>,
) -> RawRpcFuture<'a, Box<serde_json::value::RawValue>>,
Then you could do the boxing internally and keep the API cleaner (maybe you also return some future that is then boxed or whatever instead of returning RawRpcFuture<'a, Box<serde_json::value::RawValue>>
so that both boxes are removed?)
The downside is you probably end up having to copy these bounds in both the Box
and in the where
though could maybe avoid duplication with a tiny macro or trait like trait SubscriptionFn: Fn(...)
.
Not a big deal really because this is just test code but might be a nice little ergonomic win :)
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.
removed boxing form the callsite
) | ||
}); | ||
|
||
(*sub)(&mut self.data_table, &mut self.subscription_channel, params) |
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.
I geuss this means that once you call
or subscribe
, the returned fut has a mutable borrow on the RPC client and then you can't do another call or subscribe while thefirst one is open, which I geuss is working out ok!
One thing I was wondering was whether we can keep MockDataTable
separate from MockRpcClient
so that the two were separate and one could eg pick a different way to pass data in to MockRpcClient
stuff.
One approach would be to have the closure capture whatever you wanted eg:
let mock_data = MockDataTable::new();
MockRpcBuilder::default()
.add_method(
"state_getStorage",
Box::new(|_sub, params| {
Box::pin(async move {
let params = params.map(|p| p.get().to_string());
let rpc_params = jsonrpsee::types::Params::new(params.as_deref());
let key: sp_core::Bytes = rpc_params.sequence().next().unwrap();
let value = mock_data.pop(key.0).unwrap_single();
value.map(|v| serde_json::value::RawValue::from_string(v).unwrap())
})
});
But I think that you need to have the mock data embedded in order for the lifetimes to work out nicely (though one could add some easy way to clone it or whatever but that's more annoying).
Another way to keep the two separate might be to have MockRpcClient<Data>
so that one can be generic over the structure of data provided etc, allowing for different testing places to use different things. This would lead to the generic being added in a bunch of places though as a tradeoff for decoupling the two things.
Not sure it's a big deal either way, but it's the sort of thing I'd probably consider in this sort of design to make the MockRpcClient
more generic in terms of how it can be used :)
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.
Ok, so a few methods/subscriptions do in fact require access to the same data point, so capturing stuff in a closure ends up more cumbersome that the current approach in some cases
Description
As in the title. This PR adds tests for
UnstableBackend
implementation in subxt/backend.The tests are mostly about testing
StorageResponses
in different cases of reconnections.Pr uses a mocked
RpcClient
implementation for providing expected responses.Additional changes
MockRpcClient
to be usable across legacy backend and unstable_backend following suggestion from last PRPart of the work on #1677