Sebastian Kunert
9be655701a
* Worker * Reorganize and unpin onnotification drop * Pin in state-db, pass block number * Pin blocks in blockchain db * Switch to reference counted LRU * Disable pinning when we keep all blocks * Fix pinning hint for state-db * Remove pinning from backend layer * Improve readability * Add justifications to test * Fix justification behaviour * Remove debug prints * Convert channels to tracing_unbounded * Add comments to the test * Documentation and Cleanup * Move task start to client * Simplify cache * Improve test, remove unwanted log * Add tracing logs, remove expect for block number * Cleanup * Add conversion method for unpin handle to Finalitynotification * Revert unwanted changes * Improve naming * Make clippy happy * Fix docs Co-authored-by: Michal Kucharczyk <1728078+michalkucharczyk@users.noreply.github.com> * Use `NumberFor` instead of u64 in API * Hand over weak reference to unpin worker task * Unwanted * &Hash -> Hash * Remove number from interface, rename `_unpin_handle`, LOG_TARGET * Move RwLock one layer up * Apply code style suggestions * Improve comments * Replace lru crate by schnellru * Only insert values for pinned items + better docs * Apply suggestions from code review Co-authored-by: Bastian Köcher <git@kchr.de> * Improve comments, log target and test Co-authored-by: Michal Kucharczyk <1728078+michalkucharczyk@users.noreply.github.com> Co-authored-by: Bastian Köcher <git@kchr.de>
Integration of the GRANDPA finality gadget into substrate.
This crate is unstable and the API and usage may change.
This crate provides a long-running future that produces finality notifications.
Usage
First, create a block-import wrapper with the block_import function. The
GRANDPA worker needs to be linked together with this block import object, so
a LinkHalf is returned as well. All blocks imported (from network or
consensus or otherwise) must pass through this wrapper, otherwise consensus
is likely to break in unexpected ways.
Next, use the LinkHalf and a local configuration to run_grandpa_voter.
This requires a Network implementation. The returned future should be
driven to completion and will finalize blocks in the background.
Changing authority sets
The rough idea behind changing authority sets in GRANDPA is that at some point, we obtain agreement for some maximum block height that the current set can finalize, and once a block with that height is finalized the next set will pick up finalization from there.
Technically speaking, this would be implemented as a voting rule which says, "if there is a signal for a change in N blocks in block B, only vote on chains with length NUM(B) + N if they contain B". This conditional-inclusion logic is complex to compute because it requires looking arbitrarily far back in the chain.
Instead, we keep track of a list of all signals we've seen so far (across all forks), sorted ascending by the block number they would be applied at. We never vote on chains with number higher than the earliest handoff block number (this is num(signal) + N). When finalizing a block, we either apply or prune any signaled changes based on whether the signaling block is included in the newly-finalized chain.
License: GPL-3.0-or-later WITH Classpath-exception-2.0