Load sync states from the chain spec on the light client #7329
Conversation
…-state-WIP-loading
…-state-WIP-loading
…-state-WIP-loading
…rove-sync-state-WIP-loading
This reverts commit a8f0966.
expenses
added
A0-please_review
We should probably try and find out what the difference between the two is 🤔 |
|
There was a problem hiding this comment.
I do not know if 'allow_missing_parent' can be avoided, if it can't, it could also think of naming it 'snapshot_syncing' or something that refer to the actual action it is used for.
Still 'allow_missing_parent' is similar to existing 'allow_missing_state' (probably for light client), so maybe that is not a good idea at all place.
Generally loading from a snapshot do not use the same trust assumption, so I do not find this that surprising it is needed (maybe some specific api or trait could make it looks less hacky, I think making code generic may help here but that should be next steps).
| @@ -110,7 +110,7 @@ pub(crate) fn write_epoch_changes<Block: BlockT, F, R>( | |||
| } | |||
|
|
|||
| /// Write the cumulative chain-weight of a block ot aux storage. | |||
| pub(crate) fn write_block_weight<H: Encode, F, R>( | |||
| pub fn write_block_weight<H: Encode, F, R>( | |||
There was a problem hiding this comment.
why not have a specific 'load_snapshot_state' method that is also public but makes it clear that in should not be use lighty?
Could be part of a trait.
| @@ -1275,10 +1275,12 @@ impl<Block, Client, Inner> BlockImport<Block> for BabeBlockImport<Block, Client, | |||
| // used by pruning may not know about the block that is being | |||
| // imported. | |||
| let prune_and_import = || { | |||
| /* | |||
There was a problem hiding this comment.
Wonder if light pruning method from https://github.com/paritytech/substrate/pull/6851/files would work here, really just a wild guess.
| let hash = header.hash(); | ||
| let number = header.number(); | ||
|
|
||
| // Note: this may lock storage, so it must happen before obtaining storage | ||
| // write lock. | ||
| let best_tree_route = { | ||
| let best_hash = self.storage.read().best_hash; | ||
| if &best_hash == header.parent_hash() { | ||
| if &best_hash == header.parent_hash() || allow_missing_parent { |
There was a problem hiding this comment.
I wonder if modifying tree_route to not fail on missing parent may be better, or if it could break some assumption and make existing code unsecure.
gnunicorn
added
the
A5-stale
|
Closing due to the PR being stale for a while. |
A follow-up to #7225 and part of #6804.
Some questions about the PR that we should:
allow_missing_parentcode?EpochChangestree? See Add missing fields to the light sync state #7225 (comment).