`example_bitcoind_rpc_polling` sync command ends with stack overflow

[ValuedMammal]

Describe the bug
I ran a sync from genesis using a local signet node. It took around 25 minutes to reach the new tip but ended abruptly before finishing.

[  1442.151s] committed to db (took 0.11852204s)
[  1443.706s] synced to 0000003dee9b4362728c83d684d8490ddf4ffeb6ad7fe5ccda3f2203bffc24f2 @ 210819 | total: 0.00020300 BTC
[ 1449.7129s] synced to 000000a55e6a22bfd21c68f46774abda1e403ea95983f7d26981ccb24984ac60 @ 211577 | total: 0.00020300 BTC
[ 1455.7239s] synced to 00000071de5c82e264071bd7f0a8a91eab1e1b5461d44fce3b0fa99092e3f314 @ 212323 | total: 0.00020300 BTC
[ 1461.7284s] synced to 000000eeb493d04aeb69f4363bd01eb30540ff125a978c25ba082ad5f405071f @ 212890 | total: 0.00020300 BTC
[ 1467.7336s] synced to 0000002fdc5b95799d03f8c150fe7ead118bb9348f532edbdd5eda1bd02c1de7 @ 213657 | total: 0.00020300 BTC
[  1473.861s] synced to 000000b71b4299131fec5052912e975a10058a0c5c7532af945da9e081c76275 @ 214637 | total: 0.00019500 BTC
[   1479.87s] synced to 0000009c12399ad6344fe09264e4292f33b8db09fd26b79ad0f5bae102e9b880 @ 215572 | total: 0.00069500 BTC

thread 'main' has overflowed its stack
fatal runtime error: stack overflow
zsh: abort      cargo run --bin example_bitcoind_rpc_polling -- sync

To Reproduce

DESCRIPTOR=tr([7d94197e/86'/1'/0']tpubDCyQVJj8KzjiQsFjmb3KwECVXPvMwvAxxZGCP9XmWSopmjW3bCV3wD7TgxrUhiGSueDS1MU5X1Vb1YjYcp8jitXc5fXfdC1z68hDDEyKRNr/0/*)
CHANGE_DESCRIPTOR=tr([7d94197e/86'/1'/0']tpubDCyQVJj8KzjiQsFjmb3KwECVXPvMwvAxxZGCP9XmWSopmjW3bCV3wD7TgxrUhiGSueDS1MU5X1Vb1YjYcp8jitXc5fXfdC1z68hDDEyKRNr/1/*)
FALLBACK_HEIGHT=1

cargo run --bin example_bitcoind_rpc_polling -- sync

Expected behavior

Build environment

• BDK tag/commit: v1.0.0-beta.4
• OS+version: macOS Sonoma 14.5
• Rust/Cargo version: 1.81.0
• Rust/Cargo target: aarch64-apple-darwin

Additional context

[ValuedMammal]

New information: the database file is ~7.9M. Interestingly I also get a stack overflow when calling the balance command, indicating that there could be a problem loading the file contents into memory (as a side point this balance also appears wrong to me).

     Running `target/debug/example_bitcoind_rpc_polling balance`
confirmed:
    total             69500 sats
    spendable         69500 sats
    immature              0 sats
unconfirmed:
    total                 0 sats
    trusted               0 sats
    untrusted             0 sats

thread 'main' has overflowed its stack
fatal runtime error: stack overflow
zsh: abort      cargo run --bin example_bitcoind_rpc_polling -- balance

I commented out the line here in example_cli where we apply the chain changeset

bdk/example-crates/example_cli/src/lib.rs

Lines 810 to 815 in 139d971

	let chain = Mutex::new({
	let (mut chain, _) =
	LocalChain::from_genesis_hash(constants::genesis_block(network).block_hash());
	chain.apply_changeset(&changeset.local_chain)?;
	chain
	});

and while no balance is returned (as expected), it does prevent hitting a stack overflow

     Running `target/debug/example_bitcoind_rpc_polling balance`
confirmed:
    total                 0 sats
    spendable             0 sats
    immature              0 sats
unconfirmed:
    total                 0 sats
    trusted               0 sats
    untrusted             0 sats

[luisschwab]

Interesting, I also had this problem a while ago. Will try to reproduce it again and report back.

[ValuedMammal]

Seems to be isolated to the CheckPoint type, but the question remains whether something should be done in bdk_bitcoind_rpc to mitigate it

    #[test]
    fn apply_massive_changeset() {
        let (mut chain, _) = LocalChain::from_genesis_hash(genesis_block(Network::Bitcoin).block_hash());
        let blocks = (1u32..850_000)
            .into_iter()
            .map(|height| {
                let hash = bitcoin::hashes::Hash::hash(height.to_string().as_bytes());
                (height, Some(hash))
            })
            .collect();
        chain.apply_changeset(&ChangeSet { blocks }).unwrap();
    }

[luisschwab]

I'm having some trouble reproducing this. Can you provide the command sequence you ran @ValuedMammal ?

btw, the crate's README needs some updating

[ValuedMammal]

@luisschwab I usually add my descriptors as environment variables and then just follow the help text of the CLI. I'm interested to see if someone can reproduce it, otherwise it could be something specific to my architecture. You're right the README is outdated

# setup environment
export DESCRIPTOR="tr([7d94197e/86'/1'/0']tpubDCyQVJj8KzjiQsFjmb3KwECVXPvMwvAxxZGCP9XmWSopmjW3bCV3wD7TgxrUhiGSueDS1MU5X1Vb1YjYcp8jitXc5fXfdC1z68hDDEyKRNr/0/*)"
export CHANGE_DESCRIPTOR="tr([7d94197e/86'/1'/0']tpubDCyQVJj8KzjiQsFjmb3KwECVXPvMwvAxxZGCP9XmWSopmjW3bCV3wD7TgxrUhiGSueDS1MU5X1Vb1YjYcp8jitXc5fXfdC1z68hDDEyKRNr/1/*)"
# specific for example_bitcoind_rpc_polling
# assumes a full node setup locally on signet
export RPC_URL="127.0.0.1:38332"
export RPC_COOKIE=<path-to-your-cookie-file> # e.g. /home/satoshi/.bitcoin/signet/.cookie
export FALLBACK_HEIGHT=1

# initialize database
cargo run --bin example_bitcoind_rpc_polling -- init

# sync
cargo run --bin example_bitcoind_rpc_polling -- sync

[luisschwab]

Reproduced in v1.0.0-beta.5, Ubuntu 24.04.1 LTS x86_64.

[ 2517.2751s] synced to 00000067cf51b1e213711205dc8828a7933bd6bc0c51b4eecf8ca1921e7e3182 @ 215513 | total: 0.00069500 BTC
[ 2522.3696s] committed to db (took 0.013493876s)
[ 2523.3176s] synced to 0000007467101167136fcc53596a2ee6789cc9da74f43e3d1212c0ec778a117c @ 215775 | total: 0.00069500 BTC
[ 2529.3455s] synced to 0000008ef21a51707a38677076b5380950800af75fd35c8ac0cbe63ac5d87d89 @ 216012 | total: 0.00069500 BTC

thread 'main' has overflowed its stack
fatal runtime error: stack overflow
Aborted (core dumped)

$ cargo run --bin example_bitcoind_rpc_polling -- balance
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.34s
     Running `target/debug/example_bitcoind_rpc_polling balance`
confirmed:
    total             69500 sats
    spendable         69500 sats
    immature              0 sats
unconfirmed:
    total                 0 sats
    trusted               0 sats
    untrusted             0 sats

thread 'main' has overflowed its stack
fatal runtime error: stack overflow
Aborted (core dumped)

My DB is similarly large at 8MB.

$ du -sh .bdk_example_rpc.db 
8.0M	.bdk_example_rpc.db

[LLFourn]

Seems to be isolated to the CheckPoint type, but the question remains whether something should be done in bdk_bitcoind_rpc to mitigate it

    #[test]
    fn apply_massive_changeset() {
        let (mut chain, _) = LocalChain::from_genesis_hash(genesis_block(Network::Bitcoin).block_hash());
        let blocks = (1u32..850_000)
            .into_iter()
            .map(|height| {
                let hash = bitcoin::hashes::Hash::hash(height.to_string().as_bytes());
                (height, Some(hash))
            })
            .collect();
        chain.apply_changeset(&ChangeSet { blocks }).unwrap();
    }

This doesn't really isolate it to the CheckPoint type. It isolates it to the LocalChain type. The most likely explanation in my experience is that somehow this has become recursive. I guess the right thing to do is to use rust-gdb to figure out where this happens.

[ValuedMammal]

I'm using this also as a more targeted litmus test

    let mut cp = CheckPoint::new(BlockId {
        height: 0,
        hash: hash!("g"),
    });
    let end = 10_000;
    for height in 1u32..end {
        let hash = bitcoin::hashes::Hash::hash(height.to_be_bytes().as_slice());
        let block = BlockId { height, hash };
        cp = cp.push(block).unwrap();
    }
    assert_eq!(cp.iter().count() as u32, end);

[ValuedMammal]

Screen of my debugger showing a lot of drop calls. I experimented with writing a custom Drop impl for CheckPoint which did fix the test mentioned above, but when I tested the change on example_bitcoind_rpc_polling the performance was significantly slower to the point that I gave up syncing on more than one occasion, so I guess the jury is still out.