feat!: improve block add where many orphan chain tips existed #5763

hansieodendaal · 2023-09-11T16:02:46Z

Description

Improved block add due to previous inefficient retrieval of the strongest orphan chain header.

The previous algo retrieved all orphan blocks for all orphan chain tips, without evaluating the total accumulated difficulty. This PR added total accumulated difficulty to the orphan chain tips db so that efficient retrieval of only the strongest orphan chain headers can be done.

Block add times improved drastically by multiple orders of magnitude in the case where many orphan chain tips existed as only one or very seldomly two orphan chain tips would have the same total accumulated difficulty, thereby minimizing the retrieval of orphan blocks to enable the construction of orphan chain headers.

These symptoms were aggravated when candidate blocks were solved quicker than the base node could process them, resulting in a negative spiral of adding even more orphan chain tips when corresponding slowing down of blocks that were added to the blockchain.

Motivation and Context

See above

How Has This Been Tested?

New unit test added
System level testing on Igor

What process can a PR reviewer use to test or verify this change?

Code walk through and system level testing

Breaking Changes

[] None
Requires data directory on base node to be deleted
Requires hard fork
Other - Please specify

BREAKING CHANGE: The blockchain database should be deleted and then re-synced from the network.

github-actions · 2023-09-11T16:17:22Z

Test Results (CI)

1 201 tests 1 201 ✔️ 10m 19s ⏱️
    38 suites       0 💤
      1 files       0 ❌

Results for commit f145b1b.

github-actions · 2023-09-11T16:27:58Z

Test Results (Integration tests)

  2 files +  2 11 suites +11 22m 39s ⏱️ + 22m 39s
27 tests +27 26 ✔️ +26 0 💤 ±0 1 ❌ +1
28 runs +28 27 ✔️ +27 0 💤 ±0 1 ❌ +1

For more details on these failures, see this check.

Results for commit f145b1b. ± Comparison against base commit 45c20a3.

SWvheerden

Having read-write-read-write locks is going to be worse for performance than just having a single write lock.

While you can have multiple read locks, as soon as you ask for a single write lock, you hold on all new locks and then continue one with that single write lock after all reads are done. The read lock instructions are soo fast its not a real problem to just use the write logic.
The code is now much more complex for I would argue slower performance, as we specifically care about single block add performance, this will not be slower on average as you pause processing of the block to await locks.

base_layer/core/src/chain_storage/blockchain_database.rs

base_layer/core/src/chain_storage/lmdb_db/lmdb_db.rs

stringhandler · 2023-09-12T07:08:54Z

base_layer/core/src/chain_storage/blockchain_database.rs

@@ -1891,8 +1931,21 @@ fn handle_possible_reorg<T: BlockchainBackend>(
    chain_strength_comparer: &dyn ChainStrengthComparer,
    candidate_block: Arc<Block>,
 ) -> Result<BlockAddResult, ChainStorageError> {


Maybe not related to this PR, but BlockAddREsult doesn't make sense here to return. For example, we return OrphanBlock when we are on the strongest chain

stringhandler · 2023-09-12T07:09:14Z

base_layer/core/src/chain_storage/blockchain_database.rs

-    let all_orphan_tips = db.fetch_all_orphan_chain_tips()?;
-    if all_orphan_tips.is_empty() {
+    let strongest_orphan_tips = db.fetch_strongest_orphan_chain_tips()?;
+    if strongest_orphan_tips.is_empty() {
        // we have no orphan chain tips, we have trimmed remaining headers, we are on the best tip we have, so lets
        // return ok
        return Ok(BlockAddResult::OrphanBlock);


See previous comment, but this is a strange thing to return IMO

I think its the correct thing to return.
If all the orphans tips are empty. It means the block you added is an orphan, hence the result here.
New blocks are first added as an orphan, and then from there, an orphan chain tip is created if you can link it to the main chain. Then do swap to the highest chain. So any new block will always be an orphan chain tip first.

I agree with @SWvheerden on this

I don't think you understand my point. The method name "swap_to_highest_chain" suggests it can be called at any time, but the value returned is tightly coupled to the caller context.... i.e. it is returning a critical result that the caller "handle_possible_reorg" requires. this is a private method, so it's probably fine, but it's a code smell that could potentially cause a bug in future.

Issue created: #5765

base_layer/core/src/chain_storage/blockchain_database.rs

Improved block add due to previous inefficient retrieval of the strongest orphan chain header. The previous algo retrieved all orphan blocks for all orphan chain tips, without evaluating the total accumulated difficulty. This PR added total accumulated difficulty to the orphan chain tips db so that efficient retrieval of only the strongest orphan chain headers can be done. Block add times improved drastically by multiple orders of magnitude in the case where many orphan chain tips existed as only one or very seldomly two orphan chan tips would have the same total accumulated difficulty, thereby minimizing the retrival of orphan blocks to enable construction of orphan chain headers.

stringhandler · 2023-09-12T12:00:04Z

base_layer/core/src/chain_storage/blockchain_database.rs

-    let all_orphan_tips = db.fetch_all_orphan_chain_tips()?;
-    if all_orphan_tips.is_empty() {
+    let strongest_orphan_tips = db.fetch_strongest_orphan_chain_tips()?;
+    if strongest_orphan_tips.is_empty() {
        // we have no orphan chain tips, we have trimmed remaining headers, we are on the best tip we have, so lets
        // return ok
        return Ok(BlockAddResult::OrphanBlock);


I don't think you understand my point. The method name "swap_to_highest_chain" suggests it can be called at any time, but the value returned is tightly coupled to the caller context.... i.e. it is returning a critical result that the caller "handle_possible_reorg" requires. this is a private method, so it's probably fine, but it's a code smell that could potentially cause a bug in future.

stringhandler · 2023-09-12T12:05:24Z

base_layer/core/src/proof_of_work/monero_rx/helpers.rs

@@ -552,6 +552,7 @@ mod test {
        };
        let hash = block_header.merge_mining_hash();
        append_merge_mining_tag(&mut block, hash).unwrap();
+        #[allow(clippy::redundant_clone)]
        let mut block_header2 = block_header.clone();


Suggested change

let mut block_header2 = block_header.clone();

let mut block_header2 = block_header;

The clone and #allow is to satisfy cargo clippy --all-targets --all-features -- -D warnings

stringhandler · 2023-09-12T12:05:46Z

base_layer/core/src/proof_of_work/monero_rx/helpers.rs

@@ -552,6 +552,7 @@ mod test {
        };
        let hash = block_header.merge_mining_hash();
        append_merge_mining_tag(&mut block, hash).unwrap();
+        #[allow(clippy::redundant_clone)]


Suggested change

#[allow(clippy::redundant_clone)]

ghpbot-tari-project added P-acks_required Process - Requires more ACKs or utACKs P-reviews_required Process - Requires a review from a lead maintainer to be merged labels Sep 11, 2023

SWvheerden requested changes Sep 12, 2023

View reviewed changes

base_layer/core/src/chain_storage/blockchain_database.rs Outdated Show resolved Hide resolved

base_layer/core/src/chain_storage/blockchain_database.rs Outdated Show resolved Hide resolved

base_layer/core/src/chain_storage/blockchain_database.rs Outdated Show resolved Hide resolved

stringhandler requested changes Sep 12, 2023

View reviewed changes

hansieodendaal added 2 commits September 12, 2023 11:19

Review comments

5887069

hansieodendaal force-pushed the ho_improve_block_add branch from b266ea6 to 5887069 Compare September 12, 2023 09:19

hansieodendaal requested review from SWvheerden and stringhandler September 12, 2023 09:20

SWvheerden approved these changes Sep 12, 2023

View reviewed changes

stringhandler approved these changes Sep 12, 2023

View reviewed changes

hansieodendaal mentioned this pull request Sep 12, 2023

Code smell with "fn swap_to_highest_pow_chain" #5765

Open

SWvheerden merged commit 19b3f21 into tari-project:development Sep 13, 2023

hansieodendaal deleted the ho_improve_block_add branch September 14, 2023 08:35

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat!: improve block add where many orphan chain tips existed #5763

feat!: improve block add where many orphan chain tips existed #5763

hansieodendaal commented Sep 11, 2023 •

edited

Loading

github-actions bot commented Sep 11, 2023

github-actions bot commented Sep 11, 2023

SWvheerden left a comment

stringhandler Sep 12, 2023

stringhandler Sep 12, 2023

SWvheerden Sep 12, 2023

hansieodendaal Sep 12, 2023

stringhandler Sep 12, 2023

hansieodendaal Sep 12, 2023

stringhandler Sep 12, 2023

stringhandler Sep 12, 2023

hansieodendaal Sep 12, 2023

stringhandler Sep 12, 2023

hansieodendaal Sep 12, 2023

	let mut block_header2 = block_header.clone();
	let mut block_header2 = block_header;

feat!: improve block add where many orphan chain tips existed #5763

feat!: improve block add where many orphan chain tips existed #5763

Conversation

hansieodendaal commented Sep 11, 2023 • edited Loading

Description

Motivation and Context

How Has This Been Tested?

What process can a PR reviewer use to test or verify this change?

Breaking Changes

github-actions bot commented Sep 11, 2023

Test Results (CI)

github-actions bot commented Sep 11, 2023

Test Results (Integration tests)

SWvheerden left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

hansieodendaal commented Sep 11, 2023 •

edited

Loading