Add keep-blocks archive params

[zjb0807]

a0ec652?diff=unified#diff-0e523389a21c9bd9fc1da47816eb73a5ad21637337bb938abecedfe7f936d824R1720

The logic of prune_blocks has changed, and the --keep-blocks=archive needs to be specified.

[bkchr]

The logic of prune_blocks has changed, and the --keep-blocks=archive needs to be specified.

Yes the logic has changed, that is right. However, why does keep-blofks=archive needs to be specified? Could you please go more into detail. You are also talking about benchmarks, which kind of benchmarks are you talking about and how are they broken?

[zjb0807]

substrate/client/db/src/lib.rs

Lines 1704 to 1738 in fbd7e5a

	fn prune_blocks(
	&self,
	transaction: &mut Transaction<DbHash>,
	finalized: NumberFor<Block>,
	displaced: &FinalizationOutcome<Block::Hash, NumberFor<Block>>,
	) -> ClientResult<()> {
	if let BlocksPruning::Some(blocks_pruning) = self.blocks_pruning {
	// Always keep the last finalized block
	let keep = std::cmp::max(blocks_pruning, 1);
	if finalized >= keep.into() {
	let number = finalized.saturating_sub(keep.into());
	self.prune_block(transaction, BlockId::<Block>::number(number))?;
	}
	// Also discard all blocks from displaced branches
	for h in displaced.leaves() {
	let mut number = finalized;
	let mut hash = *h;
	// Follow displaced chains back until we reach a finalized block.
	// Since leaves are discarded due to finality, they can't have parents
	// that are canonical, but not yet finalized. So we stop deleting as soon as
	// we reach canonical chain.
	while self.blockchain.hash(number)? != Some(hash) {
	let id = BlockId::<Block>::hash(hash);
	match self.blockchain.header(id)? {
	Some(header) => {
	self.prune_block(transaction, id)?;
	number = header.number().saturating_sub(One::one());
	hash = *header.parent_hash();
	},
	None => break,
	}
	}
	}
	}

The origin logic is that if self.blocks_pruning is some, the blocks will be pruned.

Now its default value is BlocksPruning::KeepFinalized. It only keeps the last finalized block. block history will be pruned.

--pruning=archive should be used with --keep-blocks=archive as RPC node. Otherwise, chain_getBlock cannot get historical blocks. I only want to tell the developers to keep this in mind.

[bkchr]

Now its default value is BlocksPruning::KeepFinalized. It only keeps the last finalized block. block history will be pruned.

KeepFinalized is the default now, yes. But, it will not only keep the last finalized block. The docs are correct:
/// Keep full finalized block history.

This means that it will keep all blocks of the finalized chain.

[bkchr]

It is also about block bodies, not state.

[lisa-parity]

@bkchr, if the docs are okay as-is, should we close this issue or would it be more useful to add some examples about the use of the --keep-blocks=archive or add some clarification around keeping finalized block bodies not the actual state?
What is your recommendation (specifically for documentation)?
Related issue opened for docs: polkadot-developers/substrate-docs#1705

[bkchr]

@bkchr, if the docs are okay as-is, should we close this issue or would it be more useful to add some examples about the use of the --keep-blocks=archive

We don't need this cli flag specified everywhere. So, I don't think we need to change any issue. Keeping non canonical block bodies is just wasting disk space.

[bkchr]

What is your recommendation (specifically for documentation)?

I don't know 100%, but I think you don't need to change anything. block-pruning is also not really used anywhere up to now.