Bug: mempool sync failed: batch request failed

[antonilol]

Describe the bug
after updating to the newest version today electrs spits out a bunch of mempool sync failed: batch request failed warnings
one time after about 20 of them electrs exited, this could be unrelated, it said something like receiving on an unconnected channel (i lost that part of the log)

Electrs version
84c9a79 (v0.10.2)

To Reproduce
Steps to reproduce the behavior:

1. Configure and start electrs
2. Connect with electrum client mempool (mempool.space local instance) server
3. Wait
4. See warnings (log level must be at least warn), maybe it exits

Expected behavior
not getting warnings that say something failed (mempool server does not use that part of electrs, mempool functions fine, including address lookup)

Configuration
no config file or env
Arguments: --log-filters DEBUG --db-dir /home/node/.bitcoin/electrs --electrum-rpc-addr=0.0.0.0:50002 --index-lookup-limit=500

System running electrs

• Deployment method: manual
• OS name and version: archlinux (uname -r: 6.1.69-1-lts)

Electrum client
Client name (if not upstream desktop Electrum) and version: mempool/mempool bf4a1fcd7ad04a9ec626e82b36668916a600bee9

Additional context
it happened just after updating 0.9.14

if it is not directly clear where this comes from i can set up a way to test the mempool sync part of electrs (i think using electrum wallet should be fine)

[romanz]

Many thanks for reporting this issue!

[eugenesan]

Same issue here.
Running master as of today [f1be85f] and Core 26.0.

[antonilol]

just to add, i run Core v25.0.0

[romanz]

IIUC, the current mempool sync algorithm may fail to retrieve some transactions (most probably due to RBF):

electrs/src/mempool.rs

Lines 54 to 98 in f1be85f

	let txids = daemon.get_mempool_txids()?;
	debug!("loading {} mempool transactions", txids.len());
	let new_txids = HashSet::<Txid>::from_iter(txids);
	let to_add = &new_txids - &old_txids;
	let to_remove = &old_txids - &new_txids;
	let to_add: Vec<Txid> = to_add.into_iter().collect();
	let mut new_entries = Vec::with_capacity(to_add.len());
	for txids_chunk in to_add.chunks(1000) {
	exit_flag.poll().context("mempool update interrupted")?;
	let entries = daemon.get_mempool_entries(txids_chunk)?;
	ensure!(
	txids_chunk.len() == entries.len(),
	"got {} mempools entries, expected {}",
	entries.len(),
	txids_chunk.len()
	);
	let txs = daemon.get_mempool_transactions(txids_chunk)?;
	ensure!(
	txids_chunk.len() == txs.len(),
	"got {} mempools transactions, expected {}",
	txs.len(),
	txids_chunk.len()
	);
	let chunk_entries: Vec<Entry> = txids_chunk
	.iter()
	.zip(entries.into_iter().zip(txs.into_iter()))
	.filter_map(|(txid, (entry, tx))| {
	let tx = tx.ok()?;
	let entry = entry.ok()?;
	Some(Entry {
	txid: *txid,
	tx,
	vsize: entry.vsize,
	fee: entry.fees.base,
	has_unconfirmed_inputs: !entry.depends.is_empty(),
	})
	})
	.collect();
	new_entries.extend(chunk_entries);
	}

In that case, it's not a real problem - unless it fails consistently.

[antonilol]

i now regularly get these warnings about twice a day (max 10 on 6th of january), but before i would get them spammed in the log every 2 or so seconds, restarting electrs seemed to fix that

[romanz]

Please try #997 - it should help with this issue.

[antonilol]

got master + #997 running on my node right now, will report later on the results as these messages did not show very often, also, no new mempool sync failed messages have been logged since january 24th which is a bit strange to me as i did not do anything to it then (no restart or update etc)

i will report bugs and logged warnings related to that pr in the thread there