* Add a `send_request` to `NetworkService`.
This function delivers responses via a provided sender and also allows
for sending requests to currently not connected peers.
* Document caveats of send_request better.
* Fix compilation in certain cases.
* Update docs + introduce IfDisconnected enum
for more readable function calls.
* Doc fix.
* Rename send_request to detached_request.
* Whitespace fix - arrrgh
* Update client/network/src/service.rs
spaces/tabs
Co-authored-by: Pierre Krieger <pierre.krieger1708@gmail.com>
* Update client/network/src/request_responses.rs
Documentation fix
Co-authored-by: Roman Borschel <romanb@users.noreply.github.com>
* Update client/network/src/service.rs
Typo.
Co-authored-by: Roman Borschel <romanb@users.noreply.github.com>
* Update client/network/src/service.rs
Better docs.
Co-authored-by: Roman Borschel <romanb@users.noreply.github.com>
* Update client/network/src/service.rs
Typo.
Co-authored-by: Roman Borschel <romanb@users.noreply.github.com>
* Update client/network/src/service.rs
Doc improvements.
Co-authored-by: Roman Borschel <romanb@users.noreply.github.com>
* Remove error in logs on dialing a peer.
This is now valid behaviour.
* Rename detached_request to start_request.
As suggested by @romanb.
* Fix merged master.
* Fix too long lines.
Co-authored-by: Pierre Krieger <pierre.krieger1708@gmail.com>
Co-authored-by: Roman Borschel <romanb@users.noreply.github.com>
* client/network: Re-enable light_client_handler.rs unit tests
* client/network: Add scaffolding for light client using req-resp
* client/network: Make it compile
* client/network: Rename OutEvent SendRequest
* client/network: Restructure light client request client and handler
* client/network: Rename light client request client to sender
* client/network: Remove light client prepare_request
* client/network/src/light: Rework configuration
* client/network: Formatting
* client/network/light: Remove RequestId
* client/network/light: Make request functions methods
* client/network/light: Refactor request wrapping
* client/network/light: Fix warnings
* client/network/light: Serialize request in method
* client/network/light: Make returning response a method
* client/network/light: Depend on request response to timeout requests
* client/network: Fix test compilation
* client/network/light: Re-enable connection test
* client/network/light: Re-enable timeout test
* client/network/light: Re-enable incorrect_response test
* client/network/light: Re-enable wrong_response_type test
* client/network/light: Re-enable retry_count_failures test
* client/network/light: Re-enable issue_request tests
* client/network/light: Re-enable send_receive tests
* client/network/light: Deduplicate test logic
* client/network/light: Remove unused imports
* client/network/light: Handle request failure
* client/network/light: Move generate_protocol_config
* client/network: Fix test compilation
* client/network: Rename light client request client to sender
* client/network: Handle too-many-requests error
* client/network: Update outdated comments
* client/network/light: Choose any peer if none has best block defined
* .maintain: Replace sentry-node with local-docker-test-network
Sentry nodes are deprecated. Thus there is no need for
`.maintain/sentry-node` to spin up a sentry node test environment.
Instead this commit rewrites the setup to contain two full-connected
validators and one light client.
With the steps below one can now spin up a local test network with
two validators, one light-client, Prometheus and Grafana.
- cargo build --release
- sudo docker-compose -f .maintain/local-docker-test-network/docker-compose.yml up
* client/network/light: Handle oneshot cancellation
* client/network/light: Do not reduce retry count on missing peer
* client/network/request-response: Assert in debug request id to be unique
* client/network/light: Choose same limit as block request protocol
* client/network: Report reputation changes via response
Allow request response protocol handlers to issue reputation changes, by
sending them back along with the response payload.
* client/network: Remove resolved TODOs
* Sync: Fix issue of not freeing a block announcement slot
There was a bug that when the block announcement validation returned an
error, the slot reserved for this validation wasn't freed. This could
lead to a situation where we rejected any block announcement from such a
peer for that the block announcement returned an error multiple times.
* Better logging
* Fuck I'm dumb
* 🤦
* client/network: Report reputation changes via response
When handling a request by a remote peer in a request response handler,
one might want to in- or de-crease the reputation of the peer. E.g. one
might want to decrease the reputation slightly for each request, given
that it forces the local node to do work, or one might want to issue a
larger reputation change due to a malformed request by the remote peer.
Instead of having to pass a peerset handle to each request response
handler, this commit suggests to allow handlers to isssue reputation
changes via the provided `pending_response` `oneshot` channel.
A reputation change issued by a request response handler via the
`pending_response` channel is received by the
`RequestResponsesBehaviour` which passes the reputation change up as an
event to eventually be send to a peerset via a peerset handle.
* client/network/req-resp: Use Vec::new instead of None::<Vec<_>>
* client/network: Rename Response to OutgoingResponse
Given that a request-response request is not called `Request` but
`InomingRequest`, rename a request-response response to
`OutgoingResponse`.
* client/finality-grandpa-warp: Send empty rep change via response
* client/network/req-resp: Add unit test for request id collision
* client/network/req-resp: Prevent request id collision
`RequestId` is a monotonically increasing integer, starting at
`1`. A `RequestId` is unique for a single `RequestResponse`
behaviour, but not across multiple `RequestResponse` behaviours. Thus
when handling `RequestId` in the context of multiple
`RequestResponse` behaviours, one needs to couple the protocol name
with the `RequestId` to get a unique request identifier.
This commit ensures that pending requests (`pending_requests`) and
pending responses (`pending_response_arrival_time`) are tracked both by
their protocol name and `RequestId`.
* client/network/req-resp: Remove unused import
* client/network/req-resp: Introduce ProtocolRequestId struct
* client/network/req-resp: Update test doc comment
Treat `RequestId` as an opaque type.
* client/network/req-resp: Improve expect proof
* Sync: Propagate block announcement data
This pr adds a feature to the sync protocol to propagate the data that
we received alongside a block announcement. This is done by adding a
cache that caches the last X block announcement data where X is set to
the number of `in_peers` (giving every peer the chance to send us a
different block). This will be required by parachains to ensure that
even peers who are not connected to a collator receive the data
alongside the block announcement to properly validate it and request the
block.
* Review comment
* Bring back the code and add new variant to ensure we don't insert block
announce data when something wasn't checked
* Also use out_peers
* Rework priority groups
* Broken tests fix
* Fix warning causing CI to fail
* [Hack] Try restore backwards-compatibility
* Fix peerset bug
* Doc fixes and clean up
* Error on state mismatch
* Try debug CI
* CI debugging
* [CI debug] Can I please see this line
* Revert "[CI debug] Can I please see this line"
This reverts commit 4b7cf7c1511f579cd818b21d46bd11642dfac5cb.
* Revert "CI debugging"
This reverts commit 9011f1f564b860386dc7dd6ffa9fc34ea7107623.
* Fix error! which isn't actually an error
* Fix Ok() returned when actually Err()
* Tweaks and fixes
* Fix build
* Peerset bugfix
* [Debug] Try outbound GrandPa slots
* Another bugfix
* Revert "[Debug] Try outbound GrandPa slots"
This reverts commit d175b9208c088faad77d9f0ce36ff6f48bd92dd3.
* [Debug] Try outbound GrandPa slots
* Apply suggestions from code review
Co-authored-by: Max Inden <mail@max-inden.de>
* Use consts for hardcoded peersets
* Revert "Try debug CI"
This reverts commit 62c4ad5e79c03d561c714a008022ecac463a597e.
* Renames
* Line widths
* Add doc
Co-authored-by: Max Inden <mail@max-inden.de>
* *: Update to libp2p v0.33.0
* client/network: Consistently track request arrival time
With https://github.com/libp2p/rust-libp2p/pull/1886/ one is guaranteed
to receive either a `ResponseSent` or a `InboundFailure` event for each
received inbound request via `RequestResponseEvent::Message`. Given this
guarantee there is no need to track arrival times in a best-effort
manner and thus there is no need to use a LRU cache for arrival times.
* client/offchain: Adjust to PeerId API changes
* Update common block in sync after importing blocks of a peer
This updates the sync code to update the common block of a peer, after
we have imported blocks from this peer. This fixes a bug for when we are
connected to one or more nodes that are doing a full sync as our node.
Nodes in full sync will not announce new blocks, as we don't send import
notifications on full sync. The problem as now that we were connected to
some peer that reported some low number as its best and we tried to sync
these blocks. But, as we did not update the common block of this peer,
we would sync these blocks over and over again. Being captured in some
time warp.
The solution to this problem is that we increase the common number as we
import blocks from this peer.
* Test
* Test name..
* Fix test
* Cleanup some code and write some new regression test
* Implement the ancestor search
* Check that the common number is smaller than the last finalized block
* Update client/network/src/protocol/sync.rs
Co-authored-by: André Silva <123550+andresilva@users.noreply.github.com>
* Update client/network/src/protocol/sync.rs
Co-authored-by: André Silva <123550+andresilva@users.noreply.github.com>
* Update client/network/src/protocol/sync.rs
Co-authored-by: André Silva <123550+andresilva@users.noreply.github.com>
* Change the way we build the status messages
* Start some new test...
* Finish test
* Rename test
* Update client/network/src/protocol.rs
Co-authored-by: André Silva <123550+andresilva@users.noreply.github.com>
Co-authored-by: André Silva <123550+andresilva@users.noreply.github.com>
* *: Update to libp2p v0.32.0
* Cargo.lock: Update async-tls to 0.10.2
* client/network/request_response: Adjust to new request response events
* client/network/request_response.rs: Clean up silently failing responses
* client/network/discovery: Lazily instantiate mdns
* client/network/discovery: Exclude MdnsWrapper for target_os unknown
* client/network/discovery: Fix indentation
* client/network/request-response: Use LruCache to track pending resp time
* client/network/request_responses: Fix early connection closed error
* client/network/request-response: Replace debug_assert with debug
* client/network/request-response: Fix typo
* client/network/request-response: Don't emit event on send_response fail
* client/network/request-response: Revert waker.wake_by_ref()
* client/network/request-resp: Make duration in InboundRequest optional
* client/network/req-resp: Don't emit two events for busy builder
When a response builder is busy incoming requests are dropped.
Previously this was reported both via a `ResponseFailure::Busy` and a
`ReponseFailure::Network(InboundFailure::Omisssion)` event.
With this commit the former is removed, leaving only the latter in
place.
* Use block requests to check if block responses are correct
Before this pr sync relied on recently announced blocks to check if a
given peer response is correct. However this could lead to situations
where we requested a block from a peer and it gave us the requested, but
we rejected the response because this peer never send us an announcement
for the given block. See the added tests for a reproduction of the
problem.
With this pr, we now take the block request to check if a given response
matches the request. A node should not send us a block response
without a request anyway.
Essentially there is still a bug, because as you see in the test, we are
requesting block 2, while we already have this block imported. It even
happens that we request a block from the network that we have authored.
However a fix for this would require some more refactoring of the sync code.
* Revert change
* Give the test a proper name
* Add moar logging
* Move cheaper checks
* Move checks to common place
* Remove sc_network::NetworkService::register_notifications_protocol
* Missing calls to .into()
* Wrong crate name
* [WIP] Fix Grandpa tests
* One more passing
* One more. Two to go.
* This one was actually already passing 🎉
* Last one compiles
* Progress
* grandpa: fix voter_persists_its_votes test
* Restore other tests
* Try spawn future later
Co-authored-by: André Silva <andrerfosilva@gmail.com>
* Log outgoing notifications too
* Update client/network/src/protocol/generic_proto/handler.rs
Co-authored-by: Max Inden <mail@max-inden.de>
Co-authored-by: Addie Wagenknecht <addie@nortd.com>
Co-authored-by: Max Inden <mail@max-inden.de>
* chore/error: remove from str conversion and add deprecation notifications
* fixup changes
* fix test looking for gone ::Msg variant
* another test fix
* one is duplicate, the other is not, so duplicates reported are n-1
* darn spaces
Co-authored-by: Andronik Ordian <write@reusable.software>
* remove pointless doc comments of error variants without any value
* low hanging fruits (for a tall person)
* moar error type variants
* avoid the storage modules for now
They are in need of a refactor, and the pain is rather large
removing all String error and DefaultError occurences.
* chore remove pointless error generic
* fix test for mocks, add a bunch of non_exhaustive
* max line width
* test fixes due to error changes
* fin
* error outputs... again
* undo stderr adjustments
* Update client/consensus/slots/src/lib.rs
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
* remove closure clutter
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
* more error types
* introduce ApiError
* extract Mock error
* ApiError refactor
* even more error types
* the last for now
* chore unused deps
* another extraction
* reduce should panic, due to extended error messages
* error test happiness
* shift error lines by one
* doc tests
* white space
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
* Into -> From
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
* remove pointless codec
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
* avoid pointless self import
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
Co-authored-by: Bernhard Schuster <bernhard@parity.io>
Co-authored-by: Andronik Ordian <write@reusable.software>
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
* Inform sync explicitly about new best block
Instead of "fishing" the new best block out of the processed blocks, we
now tell sync directly that there is a new best block. It also makes
sure that we update the corresponding sync handshake to the new best
block. This is required for parachains as they first import blocks and
declare the new best block after being made aware of it by the relay chain.
* Adds test
* Make sure async stuff had time to run
* client/authority-discovery: Publish and query on exponential interval
When a node starts up publishing and querying might fail due to various
reasons, for example due to being not yet fully bootstrapped on the DHT.
Thus one should retry rather sooner than later. On the other hand, a
long running node is likely well connected and thus timely retries are
not needed. For this reasoning use an exponentially increasing interval
for `publish_interval`, `query_interval` and
`priority_group_set_interval` instead of a constant interval.
* client/authority-discovery/src/interval.rs: Add license header
* .maintain/gitlab: Ensure adder collator tests are run on CI
* Remove necessity to pass ConsensusEngineId when registering notifications protocol
* Line width
* Fix tests protocol name
* Other renames
* Doc update
* Change issue in TODO
Benjamin Kampmann