-
Notifications
You must be signed in to change notification settings - Fork 2.8k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Applying optimizations from 0.17.11
release
#1158
Conversation
- Replaced combined `column_key` with an array of hash maps. - Added RocksDB LRU cache for uncompressed rows. It speeds up the loading of objects from the database two times. <img width="1190" alt="image" src="https://user-images.githubusercontent.com/18346821/233048562-428ee4e5-39fa-4e97-8918-b764824814ba.png"> <img width="1190" alt="image" src="https://user-images.githubusercontent.com/18346821/233048615-4f4adb88-7736-471d-b5a9-048e24fcd81a.png"> - Optimized the deserialization of the contracts via the `OptimizedContract` wrapper type. Instead of 53% now it takes only 10% of the time. <img width="1195" alt="image" src="https://user-images.githubusercontent.com/18346821/233048817-85ae413f-888a-4522-9dde-287814514d5a.png"> - Replaced all values `Vec<u8>` with `Arc<Vec<u8>>` to avoid cloning because we use it only for deserialization, and we can't consume the actual `Vec<u8>`. It is helpful for `MemoryStore` and `MemoryTransactionView`. - Added a new e2e test for the contract with a large state. - Updated the `snapshot`command to do a snapshot only of the contract's state. - Updated the `MemoryStore` to use `BTreeMap` instead of `HashMap`. It significantly improved the performance of the `iter_all` without `RocksDB`. Without this optimization, a new e2e failed for `--no-default-features` build with a timeout(but worked with `RocksDB`). - Added a new e2e test that allows dry run any hex-encoded transaction. - Simplified the `PrometheusExtension`. - Use a new `fuel-vm 0.26.3` with disabled logging for `instruction` function. - Use `tokio_rayon::spawn_fifo` to run dry runs instead of manual semaphore and `tokio::spawn_blocking`. - Removed 5 second delay for dry runs caused by the relayer.
This seems like the same PR that I already reviewed. If so maybe it can just be merged? |
It is almost the same, but tuned for the If you think that it is ready to land |
# Conflicts: # Cargo.lock # bin/fuel-core/src/cli/snapshot.rs # crates/fuel-core/src/graphql_api/service.rs
)?; | ||
self.apply_coinbase(coinbase_tx, block, execution_data, block_db_transaction)?; | ||
if execution_kind != ExecutionKind::DryRun { | ||
coinbase_tx = self.check_coinbase( |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Do we still need to do this check in production mode?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We need only to apply coinbase, but not to check. But while the check_coinbase
function does simple check, I think it is better to keep it to be sure that everything is okay(and makes the code consistante)=)
crates/fuel-core/src/executor.rs
Outdated
*block.transactions_mut() = vec![script.into()]; | ||
|
||
let ExecutionResult { block, .. } = producer | ||
.execute_and_commit(ExecutionBlock::DryRun(block.into())) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
should we add any special logic here to ensure that dry run's won't actually commit? Or even better, enforce it at a type level?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We use execute_and_commit
only for tests, so I marked it with the corresponding compilation feature. Inside the method will panic if someone passes the ExecutionBlock ::DryRun
.select_new_da_height(previous_block_info.da_height) | ||
.await?; | ||
let mut block_header = self._new_header(height, block_time)?; | ||
let new_da_height = self.select_new_da_height(block_header.da_height).await?; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
maybe add some comments around this part of the code, to make it more clear that dry runs should reuse a previous da_height while new blocks should fetch da_height from the relayer. Since this was a big performance issue in the past, I think it's worth making the distinction very clear and explain why.
The change bumps the version to `0.18.0` and exposes `sync_max_get_header` and `sync_max_get_txns` in the helm chart. # Release 0.18.0 ## Overview A new release brings: - Optimization for the execution based on the performance from beta 3 and on internal benchmarks. Improved metrics gathering. - Stabilization and better test coverage of the `fuel-vm`. We removed almost all unsafe code and added test cases for each opcode. Fixed some edge cases with memory in the `fuel-vm`. - Fully integrated Merkle trees and filled all malleable fields in the transactions. - Added retryable messages, removed redundant fields from it, and updated the API to support a new commitment schema. ## What's Changed ### Breaking - All unsafe functions were replaced with safe analog in the `fuel-crypto` - FuelLabs/fuel-vm#346 - `$hp` holds the address of the last available byte in a heap, while previously it was `$hp - 1` - FuelLabs/fuel-vm#377 - Each variant in the `fuel_tx::Input` enum now has its own type - FuelLabs/fuel-vm#364 - Message nonce is unified and now `Bytes32` everywhere - FuelLabs/fuel-vm#394 - Removed the `message_id` field from all places - FuelLabs/fuel-vm#397, FuelLabs/fuel-vm#373, - Unified the block height in all places with the introduction of a new `BlockHeigh` - FuelLabs/fuel-vm#410 - Make SMO instruction take data ptr as an argument - FuelLabs/fuel-vm#404 - Now the chain id affects the signature and predicate's owner and should be passed into the `sign` function - FuelLabs/fuel-vm#406 - Updated the `produce_blocks` endpoint to accept the start time and the number of blocks. All new blocks will use the previous timestamp as a base - #1059 - The `fuel-core` stores only unspent coins and messages, so all API that previously returned spent coins is affected - Prune owned coin idx when inputs are spent by @Voxelot in #1055 - The message proof API has been changed to be compatible with a new version - #1071 - The `fuel-core` now has retryable messages and coin messages. Retryable messages can only be consumed during successful transaction execution. The coin message acts as common coins. `resouces_to_spend` API was replaced with `coins_to_spend` that returns a new `CoinType` type. - #1067 ## All changes * adding fuel-core service monitor to helm chart by @rfuelsh in #1037 * Adding specific app selector by @rfuelsh in #1039 * use sticky sessions for GQL requests to sentries by @Voxelot in #1042 * Add ingress service name to helm chart by @rfuelsh in #1043 * Added test to verify the iteration over owned transactions by @xgreenx in #1041 * Change sentry lb to use clusterIP by @Voxelot in #1045 * Fixed Tempfile dependency by @ControlCplusControlV in #1048 * Write contract code in raw by @freesig in #994 * attempt to use buildjet runners by @Voxelot in #1017 * Set tx pointer during block production by @Voxelot in #1054 * Used `BASE_AMOUNT` for test with bob to pay for fee by @xgreenx in #1057 * Prepare GraphQL Crate for extraction by @ControlCplusControlV in #1012 * Support large contract deployments over p2p by @Voxelot in #1062 * fix yaml and add linting job by @Voxelot in #1063 * Actualized the ABI for message to be compatible with last version of the Solidity contracts. by @xgreenx in #1060 * feat: Contract state and assets merkle data storage by @bvrooman in #1008 * Take into account the previous block timestamp during block production by @xgreenx in #1059 * Prune owned coin idx when inputs are spent by @Voxelot in #1055 * Retrayable messages fuel-core part by @xgreenx in #1067 * chore: Additional Tests for Contract States and Balances by @bvrooman in #1073 * Rust 1.68.1 & Sparse Registry by @Voxelot in #1077 * Nginx tuneup by @Voxelot in #1080 * chore: Update balance and state Merkleization by @bvrooman in #1084 * Support sticky session in the client by @xgreenx in #1079 * Added e2e test to run 1000 `dry_run` by @xgreenx in #1091 * Use docker.io auth credentials to overcome rate limiting by @Voxelot in #1092 * honeycomb tracing subscription by @Voxelot in #1083 * Update withdrawal proof API to support proving from a block header committed to L1 by @xgreenx in #1071 * Upstream v0.17.8 by @Voxelot in #1102 * use chainid for transactions and predicates by @Voxelot in #1107 * fix the dry-run e2e test to actually perform dry runs by @Voxelot in #1111 * peer reputation with separate app & gossipsub score by @leviathanbeak in #1028 * Adding sentry data synchronization by @rfuelsh in #1115 * Capture metrics for graphql api by @Voxelot in #1113 * Update cargo audit and add to commit by @freesig in #1152 * add task params by @leviathanbeak in #1159 * Took into account that block creation could take some time by @xgreenx in #1162 * Applying optimizations from `0.17.11` release by @xgreenx in #1158 **Full Changelog**: v0.17.3...v0.18.0
The change bumps the version to `0.18.0` and exposes `sync_max_get_header` and `sync_max_get_txns` in the helm chart. # Release 0.18.0 ## Overview A new release brings: - Optimization for the execution based on the performance from beta 3 and on internal benchmarks. Improved metrics gathering. - Stabilization and better test coverage of the `fuel-vm`. We removed almost all unsafe code and added test cases for each opcode. Fixed some edge cases with memory in the `fuel-vm`. - Fully integrated Merkle trees and filled all malleable fields in the transactions. - Added retryable messages, removed redundant fields from it, and updated the API to support a new commitment schema. ## What's Changed ### Breaking - All unsafe functions were replaced with safe analog in the `fuel-crypto` - FuelLabs/fuel-vm#346 - `$hp` holds the address of the last available byte in a heap, while previously it was `$hp - 1` - FuelLabs/fuel-vm#377 - Each variant in the `fuel_tx::Input` enum now has its own type - FuelLabs/fuel-vm#364 - Message nonce is unified and now `Bytes32` everywhere - FuelLabs/fuel-vm#394 - Removed the `message_id` field from all places - FuelLabs/fuel-vm#397, FuelLabs/fuel-vm#373, - Unified the block height in all places with the introduction of a new `BlockHeigh` - FuelLabs/fuel-vm#410 - Make SMO instruction take data ptr as an argument - FuelLabs/fuel-vm#404 - Now the chain id affects the signature and predicate's owner and should be passed into the `sign` function - FuelLabs/fuel-vm#406 - Updated the `produce_blocks` endpoint to accept the start time and the number of blocks. All new blocks will use the previous timestamp as a base - FuelLabs/fuel-core#1059 - The `fuel-core` stores only unspent coins and messages, so all API that previously returned spent coins is affected - Prune owned coin idx when inputs are spent by @Voxelot in FuelLabs/fuel-core#1055 - The message proof API has been changed to be compatible with a new version - FuelLabs/fuel-core#1071 - The `fuel-core` now has retryable messages and coin messages. Retryable messages can only be consumed during successful transaction execution. The coin message acts as common coins. `resouces_to_spend` API was replaced with `coins_to_spend` that returns a new `CoinType` type. - FuelLabs/fuel-core#1067 ## All changes * adding fuel-core service monitor to helm chart by @rfuelsh in FuelLabs/fuel-core#1037 * Adding specific app selector by @rfuelsh in FuelLabs/fuel-core#1039 * use sticky sessions for GQL requests to sentries by @Voxelot in FuelLabs/fuel-core#1042 * Add ingress service name to helm chart by @rfuelsh in FuelLabs/fuel-core#1043 * Added test to verify the iteration over owned transactions by @xgreenx in FuelLabs/fuel-core#1041 * Change sentry lb to use clusterIP by @Voxelot in FuelLabs/fuel-core#1045 * Fixed Tempfile dependency by @ControlCplusControlV in FuelLabs/fuel-core#1048 * Write contract code in raw by @freesig in FuelLabs/fuel-core#994 * attempt to use buildjet runners by @Voxelot in FuelLabs/fuel-core#1017 * Set tx pointer during block production by @Voxelot in FuelLabs/fuel-core#1054 * Used `BASE_AMOUNT` for test with bob to pay for fee by @xgreenx in FuelLabs/fuel-core#1057 * Prepare GraphQL Crate for extraction by @ControlCplusControlV in FuelLabs/fuel-core#1012 * Support large contract deployments over p2p by @Voxelot in FuelLabs/fuel-core#1062 * fix yaml and add linting job by @Voxelot in FuelLabs/fuel-core#1063 * Actualized the ABI for message to be compatible with last version of the Solidity contracts. by @xgreenx in FuelLabs/fuel-core#1060 * feat: Contract state and assets merkle data storage by @bvrooman in FuelLabs/fuel-core#1008 * Take into account the previous block timestamp during block production by @xgreenx in FuelLabs/fuel-core#1059 * Prune owned coin idx when inputs are spent by @Voxelot in FuelLabs/fuel-core#1055 * Retrayable messages fuel-core part by @xgreenx in FuelLabs/fuel-core#1067 * chore: Additional Tests for Contract States and Balances by @bvrooman in FuelLabs/fuel-core#1073 * Rust 1.68.1 & Sparse Registry by @Voxelot in FuelLabs/fuel-core#1077 * Nginx tuneup by @Voxelot in FuelLabs/fuel-core#1080 * chore: Update balance and state Merkleization by @bvrooman in FuelLabs/fuel-core#1084 * Support sticky session in the client by @xgreenx in FuelLabs/fuel-core#1079 * Added e2e test to run 1000 `dry_run` by @xgreenx in FuelLabs/fuel-core#1091 * Use docker.io auth credentials to overcome rate limiting by @Voxelot in FuelLabs/fuel-core#1092 * honeycomb tracing subscription by @Voxelot in FuelLabs/fuel-core#1083 * Update withdrawal proof API to support proving from a block header committed to L1 by @xgreenx in FuelLabs/fuel-core#1071 * Upstream v0.17.8 by @Voxelot in FuelLabs/fuel-core#1102 * use chainid for transactions and predicates by @Voxelot in FuelLabs/fuel-core#1107 * fix the dry-run e2e test to actually perform dry runs by @Voxelot in FuelLabs/fuel-core#1111 * peer reputation with separate app & gossipsub score by @leviathanbeak in FuelLabs/fuel-core#1028 * Adding sentry data synchronization by @rfuelsh in FuelLabs/fuel-core#1115 * Capture metrics for graphql api by @Voxelot in FuelLabs/fuel-core#1113 * Update cargo audit and add to commit by @freesig in FuelLabs/fuel-core#1152 * add task params by @leviathanbeak in FuelLabs/fuel-core#1159 * Took into account that block creation could take some time by @xgreenx in FuelLabs/fuel-core#1162 * Applying optimizations from `0.17.11` release by @xgreenx in FuelLabs/fuel-core#1158 **Full Changelog**: FuelLabs/fuel-core@v0.17.3...v0.18.0
Closes #1137
The PR duplicates most of the changes from the #1157 but for the
master
branch.It adds two e2e tests, but they fail because of new changes in the
fuel-vm
(the change to thePC
register). It is not critical, but replacing them with the correct scripts later would be nice.state_root
calculation for theCreate
transaction takes an eternity #1143The change improves the
fuel-core snapshot
command to accept the arguments, and you can either snap the whole state of the blockchain via theeveryone
command, or the state only of the contract. In the follow-up PRs we can snapping of the coins and messages too.Optimizations:
max-database-cache-size
.DryRun
execution type that skips the coinbaseMint
transaction. In the future, we can skip some non-actual steps too.PrometheusExtension
to not use the lock.PrometheusExtension
to gather the statistic about requests. Update cargo audit and add to commit #1152 removes the honeycomb, so I didn't duplicate it here.tokio::task::spawn_blocking
+Semaphore
bytokio_rayon::spawn_fifo
.column_key
with the array for separate hash maps per column in theMemoryStore
andMemoryTransactionView
.Vec<u8>
in the KV API by theArc<Vec<u8>>
, because in most cases, we use it only for deserialization, and the memory database can continue to store the type. It helps reduce the number of clones.BTreeMap
in theMemoryStore
, because it allows to have a more optimizediter_all
method. Without this change, a new e2e test fails with a timeout.