pevm: support parallel merge mode #233
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This PR implement the Parallel EVM engine Co-authored-by: setunapo Co-authored-by: sunny2022da Co-authored-by: galaio Co-authored-by: andyzhang2023
Fix an issue of incorrectly set the origin storage at parallel stateDB's GetState(). Remove this code because it is already solved by lightCopy PR: #2
rwset: support collect rwset from statedb; mvstates: support export DAG; dag: support travel all execution paths; dag: refactor versioned TxDAG; dag: support profile parallel execution path; protocol: support to transfer TxDAG in NewBLock msg; PR: #4
* fix several UT with racing issues
* fix incorrect nonce balance codehash issue
case: TestEIP1559 / TestDeleteThenCreate
* Fix ExecutionSpec tests
mainly root caused by balance not updated to dirty correctly.
also fix similar issue with nonce and codehash.
* fix TestBlockChain testcase issue
TestBlockchain/ValidBlocks/bcStateTests/refundReset.json
Co-authored-by: Sunny <[email protected]>
* dag: add merge execute path method; pevm: support dispatch with TxDAG; * dag: add merge execute path method; pevm: support dispatch with TxDAG; * dag: clean code; * statedb: fix some broken uts; * pevm: support disable slot steal; --------- Co-authored-by: galaio <[email protected]>
* fix several UT with racing issues
* fix incorrect nonce balance codehash issue
case: TestEIP1559 / TestDeleteThenCreate
* Fix ExecutionSpec tests
mainly root caused by balance not updated to dirty correctly.
also fix similar issue with nonce and codehash.
* fix TestBlockChain testcase issue
TestBlockchain/ValidBlocks/bcStateTests/refundReset.json
* fix concurrent racing issue of state.accounts.
fix incorrect use of s.accountStorageParallelLock, it is designed
to be used for dirty/pending/original storages, not the accounts and
storages.
Use statedb.AccountMux and statedb.StorageMux for accounts/storages
lock.
* fix issue of DAOTransactions
handle the issue of updateObject of mainDB object touched by DAO transaction.
---------
Co-authored-by: Sunny <[email protected]>
There can be a issue that the object updated by mainDB.GetNonce etc is obseleted. The fix use statedb.getStateObjectNoUpdate to avoid touching the stateObjects of mainDB. Case: TestBlockchain/ValidBlocks/bcEIP1559/intrinsic.json
…ng; (bnb-chain#10) * txdag: support txdag transfer in extra; * txdag: support txdag transfer in extra; --------- Co-authored-by: galaio <[email protected]>
txdag: record txdag metrics; txdag: opt txdag flag name; Co-authored-by: galaio <[email protected]>
…ain#12) Fix 3 issues: - re-execution happens only to new version of baseDB to remove redundancy execution. And remove the retry with same baseIndex that is conflicted. - incorrect merge dirty objects in addrStateChangeInSlot, which cause incorrect data.root copied with obseleted stateDB, this fix handle the created, stateChanged and deleted object separately. - stateObject.GetCommitedState check mainDB of the object delete. Co-authored-by: Sunny <[email protected]>
* pevm: support delay gas fee calculation; txdag: check gas fee receiver; tests: support PEVM+TxDAG UTs; * txdag: skip some cost time operation; tests: fix some broken UTs; --------- Co-authored-by: galaio <[email protected]>
The originStorage will miss some loading in txn execution,do merge rather than simple copy This fix also use stateObject specific lock for storage update, rather than the one in stateDB. Co-authored-by: Sunny <[email protected]>
Co-authored-by: galaio <[email protected]>
txdag: opt rw record flag; txdag: opt some logic; Co-authored-by: galaio <[email protected]>
Co-authored-by: Sunny <[email protected]>
* txdag: opt some logic; txdag: opt rw set collect logic; * pevm: opt logs; * txdag: opt txdag encoding, reduce rlp size; --------- Co-authored-by: galaio <[email protected]>
* remove finalise * fix: update maindb txIndex after merge slotDB otherwise there can be issue that txIndex is load before the change in mergeSlotDB. * Fix: avoid update mainDB nonce in executeInSlot It should use slotDB, otherwise it cause the stateObjects change in mainDB, which cause racing issue. * Fix: remove stateDB update during conflict check stateDB.getState() will update the stateDB's stateObjects, which should not be updated for the purpose of state read for conflict check. --------- Co-authored-by: Sunny <[email protected]>
* txdag: add excluded flag; mvstates: generate txdag with excluded flag; * pevm: support txdag with excluded tx; * blockchain: opt txdag file mode; * pevm: fix dispatch bugs; * pevm: opt txdag dispatch; --------- Co-authored-by: galaio <[email protected]>
* pevm: opt slot trigger mechanism; * txdag: opt execute stat; * pevm: opt slot trigger mechanism; * txdag: add txdag more validation logic; --------- Co-authored-by: galaio <[email protected]>
Make the change into the merged mainDB instead of slotDB to avoid the concurrency issue Co-authored-by: Sunny <[email protected]>
* txdag: support new txdep resolve method; pevm: avoid read txdag file when generating; * pevm: support read txdag file in const size; * txdag: reduce mem alloc and async resolve tx dependency; --------- Co-authored-by: galaio <[email protected]>
) * pevm: add some parallel tx metrics; * pevm: opt read large txdag logic; --------- Co-authored-by: galaio <[email protected]>
This change enable parallel execution when txs count > parallelTxNum/2 + 2, with lower bound as 4, as the parallel execution is slower than serial execution with single thread.
This change avoid the rewrite of reads cache in slotDB.
This change fix the issue that prevDestruct is not recorded correctly and the stateObjectDestruct is not recorded in slotDB's createObject. The incorrect record causes the GetState get obsoleted state as the stateObjectDestruct is not correct.
This reverts commit 65e3bc7.
# Conflicts: # eth/ethconfig/config.go
sunny2022da
force-pushed
the
feature/TxDAG-PEVM
branch
from
53f91e4 to
9e730b3
Compare
|
close this, pls use #238 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
Currently, in our PEVM code, only the execution phase is parallelized. After the parallel execution ends, the results need to be merged serially into the StateDB. I have modified the merging process to be executed in parallel as well, which further improves the performance of the PEVM and reduces the time consumption.
Rationale
By providing a new ParallelStateDB to replace the original StateDB, it supports concurrent reading and writing of data.
Provide ParallelGasPool instead of the original GasPool to support gas calculation in a concurrent environment.
Other similar changes.
Example
Enable parallel merge mode by adding startup parameters:
Note that this startup parameter conflicts with parallel.unordered-merge, and when both are set, parallel.unordered-merge will be invalid.
Changes
Notable changes:
state.StateDBerinterface replaces direct references toStateDB.TestBlockchainWithTxDAGunit test. I found that the previous unit test had issues, as the code did not enter the branch that uses the DAG files.