metrics: incorrect write_time counter for insert_decided_block in store.rs

[ikwuoz]

arc-node/crates/consensus-db/src/store.rs

Lines 457 to 490 in a85368c

	fn insert_decided_block(
	&self,
	decided_block: DecidedBlock,
	proposer: Address,
	) -> Result<(), StoreError> {
	let start = Instant::now();
	let mut write_bytes = 0;
	let height = decided_block.height();
	let tx = self.db.begin_write()?;
	{
	let mut blocks = tx.open_table(DECIDED_BLOCKS_TABLE)?;
	let block_bytes = encode_execution_payload(&decided_block.execution_payload);
	#[allow(clippy::arithmetic_side_effects)]
	{
	write_bytes += block_bytes.len();
	}
	blocks.insert(height, block_bytes)?;
	}
	self.insert_certificate(
	&tx,
	decided_block.certificate,
	CommitCertificateType::Minimal,
	Some(proposer),
	)?;
	tx.commit()?;
	self.update_write_metrics(write_bytes, start.elapsed());
	Ok(())
	}

In insert_decided_block above, start.elapsed() spans the entire function body including insert_certificate(), which itself calls update_write_metrics with its own internal timer, then:

1. The write time in insert_decided_block is for block bytes, but it ends up including the write time for certificate bytes,
2. However insert_certificate also reports its own write time internally.

Therefore, the certificate insertion time is double counted in the aggregated write_time metric because both calls accumulate into the same counter.

Impact: The write latency metric overstates actual DB write time by roughly one cert encoding + insert cycle per insert_decided_block call, causing metric inaccuracy.