* subsystems have an unbounded channel to the overseer
* Update node/overseer/src/lib.rs
Co-authored-by: Bernhard Schuster <bernhard@ahoi.io>
* bump Cargo.lock
Co-authored-by: Bernhard Schuster <bernhard@ahoi.io>
* Do not send empty view updates to peers
It happened that we send empty view updates to our peers, because we
only updated our finalized block. This could lead to situations where we
overwhelmed sub systems with too many messages. On Rococo this lead to
constant restarts of our nodes, because some node apparently was
finalizing a lot of blocks.
To prevent this, the pr is doing the following:
1. If a peer sends us an empty view, we report this peer and decrease it
reputation.
2. We ensure that we only send a view update when the `heads` changed
and not only the `finalized_number`.
3. We do not send empty `ActiveLeavesUpdates` from the overseer, as this
makes no sense to send these empty updates. If some subsystem is relying
on the finalized block, it needs to listen for the overseer signal.
* Update node/network/bridge/src/lib.rs
Co-authored-by: Peter Goodspeed-Niklaus <coriolinus@users.noreply.github.com>
* Don't work if they're are no added heads
* Fix test
* Ahhh
* More fixes
Co-authored-by: Peter Goodspeed-Niklaus <coriolinus@users.noreply.github.com>
* Fuse receive stream in Context
* Revert "Fuse receive stream in Context"
This reverts commit ddd26fa98f0ca1afbc22064e93010e4193a058b2.
* Exit on node shutdown from av-store loop
* Filter only context error
* Store all chunks and in a single transaction
* Adds chunks LRU to store
* Add pruning records metrics
* Use honest cache instead of LRU
* Remove unnecessary optional cache
* Fix review nits that are fixable
* Companion PR for refactoring priority groups
* Fix non reserved node
* Try fix tests
* Missing import
* Fix warning
* Change protocols order
* Fix test
* Renames
* Update syn dependency to make it compile again after merging master
* "Update Substrate"
Co-authored-by: parity-processbot <>
* add timing setup to OverseerSubsystemContext
* figure out how to initialize the rng
* attach a timer to a portion of the messages traveling to the Overseer
This timer only exists / logs a fraction of the time (configurable
by `MESSAGE_TIMER_METRIC_CAPTURE_RATE`). When it exists, it tracks
the span between the `OverSubsystemContext` receiving the message
and its receipt in `Overseer::run`.
* propagate message timing to the start of route_message
This should be more accurate; it ensures that the timer runs
at least as long as that function. As `route_message` is async,
it may not actually run for some time after it is called (or ever).
* fix failing test
* rand_chacha apparently implicitly has getrandom feature
* change rng initialization
The previous impl using `from_entropy` depends on the `getrandom`
crate, which uses the system entropy source, and which does not
work on `wasm32-unknown-unknown` because it wants to fall back to
a JS implementation which we can't assume exists.
This impl depends only on `rand::thread_rng`, which has no documentation
stating that it's similarly limited.
* remove randomness in favor of a simpler 1 of N procedure
This deserves a bit of explanation, as the motivating issue explicitly
requested randomness. In short, it's hard to get randomness to compile
for `wasm32-unknown-unknown` because that is explicitly intended to be
as deterministic as practical. Additionally, even though it would never
be used for consensus purposes, it still felt offputting to intentionally
introduce randomness into a node's operations. Except, it wasn't really
random, either: it was a deterministic PRNG varying only in its state,
and getting the state to work right for that target would have required
initializing from a constant.
Given that it was a deterministic sequence anyway, it seemed much simpler
and more explicit to simply select one of each N messages instead of
attempting any kind of realistic randomness.
* reinstate randomness for better statistical properties
This partially reverts commit 0ab8594c328b3f9ce1f696fe405556d4000630e9.
`oorandom` is much lighter than the previous `rand`-based implementation,
which makes this easier to work with.
This implementation gives each subsystem and each child RNG a distinct
increment, which should ensure they produce distinct streams of values.
* don't modify inherent data on heavy block
* write up current thinking on block weight detection
* extract inherent inclusion check into its own function
* put heavy block check into runtime
* the `inclusion` inherent call is Operational, not Mandatory
This resolves a lot of the trickiness about this issue, because
we no longer need to override or supplant any existing proposer
logic; the existing logic should exhibit these behaviors:
- the `inclusion` inherent is prioritized over standard transactions
- but if it's too heavy, i.e. in case of runtime upgrade, it'll be
dropped in favor of that.
It is my belief that allowing the proposer to just not include
this data won't have any adverse effects: it's equivalent to replacing
them with empty versions of themselves, which the `ProvideInherent`
impl already does.
* Revert "the `inclusion` inherent call is Operational, not Mandatory"
This reverts commit e58858d109b18b84e7af3ac47981c6900b2d9a3e.
* Revert "write up current thinking on block weight detection"
This reverts commit fd587b80c46761b2a2b62448193348237863f99f.
* Revert "don't modify inherent data on heavy block"
This reverts commit 38299d3c23e9efb5a354d8cfa658e62a5c8c7ddf.
* add backed candidate block weight assumption to configuration
* Limit backed candidates according to a candidate weight heuristic.
This approach replaces making the inclusion inherent non-mandatory.
It's still not ideal in that we have to configure a heuristic for
how much each backed candidate 'weighs', instead of directly
measuring it somehow.
This approach also never truncates the signed bitfields. The
rationale for that depends on some assumptions:
- processing the signed bitfields is cheap compared to the
backed candidates
- it is beneficial to the progress of the relay chain
to update the signed bitfields even if not all backed candidates
are updated
* simplify limit_backed_candidates and weight assumption
* don't trust the provisioner to fairly distribute candidates
* use saturating subtraction
* empty commit to restart ci
* use new mechanism for getting max block weight
* apply weight refunds to the inclusion inherent
This makes some assumptions about fundamental weights, which are
encapsulated as constants. From there, it lets Substrate know
what the actual computed weight of the inherent is.
* use a correct fixed weight for the inclusion inherent
Co-authored-by: Guillaume Thiolliere <gui.thiolliere@gmail.com>
* use dynamic inclusion weight so we reduce calculated weight when excluding candidates
* don't double-count this intrinsic's weight in the block weight
* add unit tests of fn limit_backed_candidates
* add tests that the inclusion inherent's weight correctly updates
Co-authored-by: Guillaume Thiolliere <gui.thiolliere@gmail.com>
* Add one Jaeger span per relay parent
This adds one Jaeger span per relay parent, instead of always creating
new spans per relay parent. This should improve the UI view, because
subsystems are now grouped below one common span.
* Fix doc tests
* Replace `PerLeaveSpan` to `PerLeafSpan`
* More renaming
* Moare
* Update node/subsystem/src/lib.rs
Co-authored-by: Andronik Ordian <write@reusable.software>
* Skip the spans
* Increase `spec_version`
Co-authored-by: Andronik Ordian <write@reusable.software>
* Add well_known_keys
* Reorder HrmpChannel and HostConfiguration members
* abridged versions and well known keys tests
* Add some comments
* Add a note on generation of the prefixes
and other magic values
* Recommend accessing the well known values through abridged structs
* Cont.: Implement the state root obtaining during inclusion
During inclusion now we obtain the storage root by passing it through
the inclusion_inherent.
* Fix tests
* Bump rococo spec version
* Reorder the parent header into the end
of the inclusion inherent.
When the parent header is in the beginning, it shifts the other two
fields, so that a previous version won't be able to decode that. If
we put the parent header in the end, the other two fields will stay
at their positions, thus make it possible to decode with the previous
version.
That allows us to perform upgrade of rococo runtime without needing of
simultanuous upgrade of nodes and runtime, or restart of the network.
* Squash a stray tab
* point to branch
* update babe API implementations
* build
* Revert "point to branch"
This reverts commit 669fde1e58cac1eef1847d5e6b39520fa4953f33.
* update Cargo.lock
* bump
This pr adds support to change the session length of a Rococo chain at
genesis. This is rather useful because Rococo has a session length of
1 hour, while on rococo-local you will now get 1 minute. This improves
the dev experience, because a parachain is only going live at the
start of a new session.
* Don't initialize block when calling runtime APIs
* Adapt check_validation_outputs
We split the code path for the inclusion and for the commitments checking.
* Slap #[skip_initialize_block] on safe runtime APIs
That is, those that should not be affected by this attribute
* Make `Scheduled` not ephemeral
So that it is persisted in the storage and ready to be inspected
by the runtime APIs. This is in contrast to what was before, where we
would remove the storage entry and then rely on the scheduling performed
by `on_initialize` again.
* Add a big fat comment
* Typos
Co-authored-by: Robert Habermeier <rphmeier@gmail.com>
* Move session change to the end of the current block
Previously, it was the beginning of the next block. This allows us to
put #[skip_initialize_block]
* Update tests
* Fix a test in paras registrar
Also refactor it a bit so the next time there are more chances this kind
of issue is diagnosed quicker.
* Add for_runtime_api to inclusion's check_validation_outputs
Co-authored-by: Robert Habermeier <rphmeier@gmail.com>
* add candidate hash statement circulation span
* add relay-parent to hash-span
* Some typos and misspellings in docs I found, during my studies. (#2144)
* Fix stale link to overseer docs
* Some typos and mispellings in docs/comments
I found during studying how Polkadot works.
* Rococo V1 (#2141)
* Update to latest master and use 30 minutes sessions
* add bootnodes to chainspec
* Update Substrate
* Update chain-spec
* Update Cargo.lock
* GENESIS
* Change session length to one hour
* Bump spec_version to not fuck anything up ;)
Co-authored-by: Erin Grasmick <erin@parity.io>
* avoid creating duplicate unbacked spans when we see extra statements (#2145)
* improve jaeger spans for statement distribution
* tweak and add failing test for repropagation
* make a change that gets the test passing
* guide: clarify
* remove semicolon
Co-authored-by: Robert Klotzner <eskimor@users.noreply.github.com>
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
Co-authored-by: Erin Grasmick <erin@parity.io>
Robert Habermeier