-
Notifications
You must be signed in to change notification settings - Fork 1.7k
Fix performance issue importing Kovan blocks #9914
Conversation
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.
Nice find! 👍 Since we used the wrong hash when building the subchain the caching wouldn't work properly and we'd have to rebuild the subchain on every block.
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.
Amazing!
@@ -815,10 +816,11 @@ impl AuthorityRound { | |||
None | |||
} | |||
}) | |||
.while_some(); | |||
.while_some() | |||
.take_while(|&(h, _)| h != epoch_transition_hash); |
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.
So that limits the iterator, can you elaborate a little bit why it's valid to do so?
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.
I discussed this with @ngotchac. Aura only used to track finality for safely triggering validator set changes, therefore we were only interested in finalizing blocks within the current epoch, so this take_while
was the original behavior. Recently we added finality tracking to the client so we updated Aura to track finality until the last finalized block. On chains where there isn't a majority of validators online the unfinalized subchain just keeps growing and this can take some time to build (although if the subchain is properly cached this cost is avoided).
#9692 is in beta, so should we backport this PR to beta as well? |
@ordian yes we should 👍 |
* Bump beta to version 2.2.1 * fix: Intermittent failing CI due to addr in use (#9885) Allow OS to set port at runtime * Use Weak reference in PubSubClient (#9886) * Fix json tracer overflow (#9873) * Fix json tracer overflow * Replace trace_executed with a direct trace push * Remove unused variable * Add test for 5a51 * Remove duplicate json! * Fix docker script (#9854) * Dockerfile: change source path of the newly added check_sync.sh (#9869) * Allow to seal work on latest block (#9876) * Allow to seal work on latest block. * Test from @todr to check sealing conditions. * gitlab-ci: make android release build succeed (#9743) * use docker cargo config file for android builds * make android build succeed * ethcore: use Machine::verify_transaction on parent block (#9900) * ethcore: use Machine::verify_transaction on parent block also fixes off-by-one activation of transaction permission contract * ethcore: clarify call to verify_transaction * foundation: #6692865, ropsten: #4417537, kovan: #9363457 * Remove rust-toolchain file (#9906) * EIP-712 implementation (#9631) * EIP-712 impl * added more tests * removed size parsing unwrap * corrected TYPE_REGEX to disallow zero sized fixed length arrays, replaced LinkedHashSet with IndexSet, added API spec to docs, fixed Type::Byte encoding branch * use Option<u64> instead of u64 for Type::Array::Length * replace `.iter()` with `.values()` Co-Authored-By: seunlanlege <[email protected]> * tabify eip712.rs * use proper comments for docs * Cargo.lock: revert unrelated changes * tabify encode.rs * EIP 191 (#9701) * added sign_191 rpc method * fixed hash_structured_data return type * added ConfirmationPayload::SignMessage for non-prefixed signatures, added tests for sign191 * renamed WithValidator -> PresignedTransaction * rename applicationData to data in test * adds docs for EIP191Version, renamed SignRequest to EIP191SignRequest * light-fetch: Differentiate between out-of-gas/manual throw and use required gas from response on failure (#9824) * fix start_gas, handle OOG exceptions & NotEnoughGas * Change START_GAS: 50_000 -> 60_000 * When the `OutOfGas exception` is received then try to double the gas until it succeeds or block gas limit is reached * When `NotEnoughBasGas error` is received then use the required gas provided in the response * fix(light-fetch): ensure block_gas_limit is tried Try the `block_gas_limit` before regard the execution as an error * Update rpc/src/v1/helpers/light_fetch.rs Co-Authored-By: niklasad1 <[email protected]> * simplify cargo audit * Use block header for building finality (#9914) * ci: nuke the gitlab caches (#9855)
The
block
passed toself.engine.ancestry_actions
didn't have all the Header's data, thus theblock.header.hash()
was wrong, and the conditionfinality_checker.subchain_head() != Some(*chain_head.parent_hash())
inAuthorityRound::build_finality
would never befalse
, so we would check all ancestry for every imported blocks, until a finalised one was to be found. This was very costly in same cases, eg. when first syncing the chain: on currentmaster
I would sync at 50 blocks/s, with this PR (as on stable) it goes at about 1_000 blocks/s.