Before we only updated the chain info of sync when we have imported
something using the import queue. However, if you import your own
blocks, this is not done using the import queue and so sync is not
updated. If we don't do this, it can lead to sync switching to "major
sync" mode because sync is not informed about new blocks. This
especially happens on Cumulus, where a collator is selected multiple
times to include its block into the relay chain and thus, sync switches
to major sync mode while the node is still building blocks.
* Seperate out the complexity in ServiceBuilder::build_common into seperate functions
* Fix line widths
* Move some functions to their respective crates
* client: use appropriate ExecutionContext for sync/import
* client: remove dead code
* client: ExecutionContext: distinguish between own and foreign imports
* client: fix cli parameter doc
* Revert "client: ExecutionContext: distinguish between own and foreign imports"
This reverts commit 0fac11520704c364a82432c5b927e987ba043cdb.
* primitives: add docs for ExecutionContext
* cli: execution strategy docs
* cli: use different execution context for importing block on validator
* cli: remove defaults from execution context flags
* transaction-pool: expose blocking api for tx submission
* service: separate ServiceBuilder::build for full and light
* service: add ServiceBuilder::build_common
* transaction-pool: extend docs
Co-authored-by: Tomasz Drwięga <tomusdrw@users.noreply.github.com>
Co-authored-by: Tomasz Drwięga <tomusdrw@users.noreply.github.com>
* Bump jsonrpc pubsub, core, http, and ws
Right now these are the packages which _need_ to be updated
so I can just the latest `jsonrpc-pubsub` code. Once a release
it cut upstream the rest of the dependencies should be updated
as well.
* Use jsonrpc-pubsub's SubscriptionManager
This places sc-rpc-api::Subscriptions
* Bump jsonrpc-core outside of sc-rpc-*
* Update client/rpc tests
Right now one of the `author` tests is failing, I
need to think a bit about how best to fix it.
* Remove Subscriptions manager
There's no need for this implementation since we're
using the one from `jsonrpc-pubsub` now
* Fix author RPC test
This test used to check for a numerial subscription ID,
whereas now it uses a string based ID which is the default
provided by `jsonrpc-pubsub`'s subscription manager.
* Remove unused NumericIdProvider
* Add missing bracket
Removed one too many with that last one, lol
* Bump `jsonrpc` to v14.2
There's an exception though. `jsonrpc-derive` cannot be bumped
past v14.0.5 just yet since it has a dependency on `quote` pinned
to v1.0.1. This means that at the moment it won't build on Substrate
since it's using v1.0.3.
* Track `jsonrpc-derive` master branch
* Bump `quote` version to v1.0.6
* Bump `jsonrpc-derive` to v14.2.1
This includes support for `quote` v1.0.6
* Use exact version for jsonrpc crates
Doing this to make sure any updates in jsonrpc don't
accidently trickle down to Polkadot.
Co-authored-by: André Silva <andre.beat@gmail.com>
* make all features explicit
* Change to -feature suffix to with- prefix
* Add newline at the end of the Cargo.toml file
* Remove rhd feature
* Remove some features from Cargo.toml
* Remove test-helpers feature in tx pool
* Return db_open_error("with-"..
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
* Rename subdb feature to with-subdb
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
* Remove 'strict' feature and cfg_attr
* Check for with-subdb feature instead of subdb
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
The `tree_route` generated by the import notification is only from the
old best block to the new best parent. This means, it does not contain
the new best block in `enacted()`. We need to prune the transactions of
the new best block "manually" to fix this bug.
Besides that, this pr also changed the `id` parameter of the `NewBlock`
chain event to `hash`. The hash of a block is unique in contrast to the
block number. (Block id can either be number or hash)
* Make tx pool aware of retracted fork blocks
* Make it compile
* Update client/transaction-pool/src/lib.rs
Co-authored-by: Nikolay Volf <nikvolf@gmail.com>
* Fix doc test
* Simplify the implementation
* Send tree route as arc to prevent heavy clones
* Switch to use `ExtrinsicHash` to make it more clear
* Fix benchmark
Co-authored-by: Nikolay Volf <nikvolf@gmail.com>
* Bump jsonrpc pubsub, core, http, and ws
Right now these are the packages which _need_ to be updated
so I can just the latest `jsonrpc-pubsub` code. Once a release
it cut upstream the rest of the dependencies should be updated
as well.
* Use jsonrpc-pubsub's SubscriptionManager
This places sc-rpc-api::Subscriptions
* Bump jsonrpc-core outside of sc-rpc-*
* Update client/rpc tests
Right now one of the `author` tests is failing, I
need to think a bit about how best to fix it.
* Remove Subscriptions manager
There's no need for this implementation since we're
using the one from `jsonrpc-pubsub` now
* Fix author RPC test
This test used to check for a numerial subscription ID,
whereas now it uses a string based ID which is the default
provided by `jsonrpc-pubsub`'s subscription manager.
* Remove unused NumericIdProvider
* Add missing bracket
Removed one too many with that last one, lol
* Bump `jsonrpc` to v14.2
There's an exception though. `jsonrpc-derive` cannot be bumped
past v14.0.5 just yet since it has a dependency on `quote` pinned
to v1.0.1. This means that at the moment it won't build on Substrate
since it's using v1.0.3.
* Track `jsonrpc-derive` master branch
* Bump `quote` version to v1.0.6
* Bump `jsonrpc-derive` to v14.2.1
This includes support for `quote` v1.0.6
Instead of having the tx pool and offchain worker being feed from the
same import notification stream, this pr splits them to use two
different streams. The first advantage of this split is that the tx pool
will not be spawned anymore in another task and instead will directly
process the notification in the same task. This has the advantage of
being faster when the system is being under load, as the tx pool will
not be waiting for being scheduled to handle the notification.
* Add Delay and info logging
* Switch from Duration to Delay in enum declaration
* Remove sleep from import_blocks fn
* Set back constans and remove unnecessary code
* Fix hot loop
* Reset timer once poll is ready, not when it's pending
* Don't clone values when calculating storage root
Instead of cloning all the keys and values of the overlay when
calculating the storage root, we pass all the values by reference. This
should probably bring some performance improvements when calculating the
storage root.
* no cow version (#6113)
Co-authored-by: cheme <emericchevalier.pro@gmail.com>
* Add BlockStream Enum and utility fn
* WIP: Modify import closure to work with BlockStream
* Fix trait bounds
* Working prototype
* Revamp block importing
* Add export_import_flow tests
* Add comments and clean code
* Add more comments in the import fn
* Add link code to import function
* Add condition when returning Ready(Ok(()) to make sure we've imported every block
* Add check for imported blocks in JSON case
* Use rest pattern
* Fix compilation error for undeclared variable
* Add polling and waker before pending
* Print read_block_count instead of count
* Simplify binary cli option with structopt
* Update test to reflect changes in CLI api
* Change Stream to take SignedBlock<B> instead of B
* Add comments to BlockStream
* Move out logic to smaller functions for clearer code
* Remove result over import_blocks return type
* Check for error in command output rather than simply checking command exit status
* Revamp export/import/revert testing
* Fix minor typos and formatting errors
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
* Remove unnecessary if condition in terminating condition
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
* Explicit error instead of returning it as a string
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
* Pass BlockStream to log_importing_status_updates and simplify matching arms for block stream
* Use .contains() instead of regex match
* Line break in match block; return future::ready instead of poll_fn
* Update Cargo.lock
* Add check so that queue doesn't grow too big
* Use Iterator instead of Stream
* Remove allow dead_code
* Remove outdated comments
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
* Return Errors instead of logging them
* Simplify match arms
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
* Remove check before terminating block import
* Apply suggestions from code review
* Check that queue is not full BEFORE calling
* Revert "Remove check before terminating block import"
This reverts commit 377823c0a648a3eb2e61185a257a61023067893d.
* Improve unit tests to make sure we actually import blocks
* Remove Unpin implementation for BlockIter
* Add prototype of enum for ImportStates
* Add working prototype for StateMachine
* Add comments for clearer code
* Add sleep before calling Waker when waiting for import queue
* Add Speedometer
* add dbg!(&log) for test debugging
* Fix lines with more than 100 cols
* Fix regex capture for test
* Update regexes to take to capture the whole number
* Rename Cmd to Command
Co-authored-by: Gavin Wood <gavin@parity.io>
* Actually rename Cmd to Command
* Apply suggestions from code review
Co-authored-by: Gavin Wood <gavin@parity.io>
* Fix compilation errors for tests
* Fix compilation errors from code review suggestion
* Update bin/node/cli/tests/export_import_flow.rs
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
Co-authored-by: Gavin Wood <gavin@parity.io>
Co-authored-by: Benjamin Kampmann <ben@gnunicorn.org>
Instead of informing the telemetry about each block that is finalized,
we only need to send the last finalized block. This removes log spam on
initial sync.
* client: Replace `unsafe_rpc_expose` with an `RpcMethods` enum
which can be either Default, Safe or Unsafe. The idea is to have the
following:
| | --rpc-external=false | --rpc-external=true |
|--------------------- |------------------- |----------------- |
| --rpc-methods=Default | | unsafe calls denied |
| --rpc-methods=Safe | unsafe calls denied | unsafe calls denied |
| --rpc-methods=Unsafe | | |
Since the previous `unsafe-rpc-expose` option was confusing.
* client: Only warn against exposing externally unsafe RPC method set
* Apply suggestions from code review
Co-Authored-By: Cecile Tonglet <cecile.tonglet@cecton.com>
* cli: Rephrase doc comment for rpc_methods config
* Improve debuggability of build_spec_works
...by printing to stderr the stderr of the command. This is normally
suppressed for succesful tests but not for failing ones - if that's the
case then it's useful to see the test failure reason inline rather than
having to execute the command separately ourselves.
* Rename RpcMethods::{Default => Auto} variant
* Update bin/node/cli/tests/build_spec_works.rs
Co-authored-by: Benjamin Kampmann <ben.kampmann@googlemail.com>
Co-authored-by: Cecile Tonglet <cecile.tonglet@cecton.com>
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
* Expose that `BasicQueue` expects blocking spawn
Up to now `BasicQueue` expected a closure that to spawn a `Future`.
This was expected to be a closure that spawns a blocking future.
However, this wasn't documented anywhere. This pr introduces a new trait
`SpawnBlocking` that exposes this requirement to the outside.
* Feedback
Arkadiy Paronyan