We know it already, but with current loading code VM will hash it once
more. It doesn't help a lot and still it costs nothing to avoid this
overhead.
name old time/op new time/op delta
VerifyWitness-8 93.4µs ± 3% 92.7µs ± 2% ~ (p=0.353 n=10+10)
name old alloc/op new alloc/op delta
VerifyWitness-8 3.43kB ± 0% 3.40kB ± 0% -0.70% (p=0.000 n=9+9)
name old allocs/op new allocs/op delta
VerifyWitness-8 67.0 ± 0% 66.0 ± 0% -1.49% (p=0.000 n=10+10)
Block processing consists of:
* saving block/transactions to the DB
* executing blocks/transactions
* processing notifications/saving AERs
* updating MPT
* atomically updating Blockchain state
Of these the first one is completely independent of others, it can be done in
a separate routine easily. The third one technically depends on the second,
it just doesn't have data until something is executed. At the same time it
doesn't affect future executions in any way, so we can offload
AER/notification processing to separate goroutine (while the main thread
proceeds with other transactions).
MPT update depends on all executions, so it can't be offloaded, but it can be
done concurrently to AER processing. And only the last thing actually needs
all previous ones to be finished, so it's a natural synchronization point.
So we spawn two additional routines and let the main one execute transactions
and update MPT as fast as it can. While technically all of these routines
could share single DAO (they are working with different KV sets) benchmarking
shows that using separate DAOs and then persisting them to lower one actually
works about 7-8%% better. At the same time we can simplify DAOs used, Cached
one is only relevant for AER processing because it caches NEP-17 tracking
data, everything else can do just fine with Simple.
The change was tested for performance with neo-bench (single node, 10 workers,
LevelDB) on two machines and block dump processing (RC4 testnet up to 50825
with VerifyBlocks set to false) on i7-8565U. neo-bench creates huge blocks
with lots of transactions while RC4 dump mostly consists of empty blocks.
Reference results (06c3dda5d1):
Ryzen 9 5950X:
RPS ≈ 20059.569 21186.328 20158.983 ≈ 20468 ± 3.05%
TPS ≈ 19544.993 20585.450 19658.338 ≈ 19930 ± 2.86%
CPU ≈ 18.682% 23.877% 22.852% ≈ 21.8 ± 12.62%
Mem ≈ 618.981MB 559.246MB 541.539MB ≈ 573 ± 7.08%
Core i7-8565U:
RPS ≈ 5927.082 6526.739 6372.115 ≈ 6275 ± 4.96%
TPS ≈ 5899.531 6477.187 6329.515 ≈ 6235 ± 4.81%
CPU ≈ 56.346% 61.955% 58.125% ≈ 58.8 ± 4.87%
Mem ≈ 212.191MB 224.974MB 205.479MB ≈ 214 ± 4.62%
DB restore:
real 0m12.683s 0m13.222s 0m13.382s ≈ 13.096 ± 2.80%
user 0m18.501s 0m19.163s 0m19.489s ≈ 19.051 ± 2.64%
sys 0m1.404s 0m1.396s 0m1.666s ≈ 1.489 ± 10.32%
After the change:
Ryzen 9 5950X:
RPS ≈ 23056.899 22822.015 23006.543 ≈ 22962 ± 0.54%
TPS ≈ 22594.785 22292.071 22800.857 ≈ 22562 ± 1.13%
CPU ≈ 24.262% 23.185% 25.921% ≈ 24.5 ± 5.65%
Mem ≈ 614.254MB 613.204MB 555.491MB ≈ 594 ± 5.66%
Core i7-8565U:
RPS ≈ 6378.702 6423.927 6363.788 ≈ 6389 ± 0.49%
TPS ≈ 6327.072 6372.552 6311.179 ≈ 6337 ± 0.50%
CPU ≈ 57.599% 58.622% 59.737% ≈ 58.7 ± 1.82%
Mem ≈ 198.697MB 188.746MB 200.235MB ≈ 196 ± 3.18%
DB restore:
real 0m13.576s 0m13.334s 0m12.757s ≈ 13.222 ± 3.18%
user 0m19.113s 0m19.490s 0m20.197s ≈ 19.600 ± 2.81%
sys 0m2.211s 0m1.558s 0m1.559s ≈ 1.776 ± 21.21%
On Ryzen 9 we've got 12% better RPS, 13% better TPS with 12% CPU and 3% RAM
more used. Core i7-8565U changes don't seem to be statistically significant:
1.8% more RPS, 1.6% more TPS with about the same CPU and 8.5% less RAM
used. It also is 1% worse in DB restore time.
The result is somewhat expected, on a powerful machine with lots of spare
cores we get 10%+ better results while on average resource-constrained laptop it
doesn't change much (the machine is already saturated). Overall, this seems to
be worthwhile.
Request NEP17 balances from a set of NEP17 contracts instead of getting
them from storage. LastUpdatedBlock tracking remains untouched, because
there's no way to retrieve it dynamically.
Balances are to be removed from state.NEP17TransferInfo, so the remnant
fields are NextTransferBatch, NewBatch and a map of LastUpdatedBlocks.
These fields are more staff-related.
Also rename dao.[Get, Put, put]NEP17Balances and STNEP17Balances
preffix.
Also rename NEP17TransferInfo.Trackers to LastUpdatedBlockTrackers
because NEP17TransferInfo.Balances are to be removed.
We have a lot of native contract types that are converted to stack items
before serialization, then deserialized as stack items and converted back to
regular structures. stackitem.Convertible allows to remove a lot of repetitive
io.Serializable code.
This also introduces to/from converter in testserdes which unfortunately
required to change util tests to avoid circular references.
Problem: with StateRootInHeader setting on only one header of height N+1
can be added to the chain of height N, because we need local stateroot
to verify headers (which is calculated for the last stored block N).
Thus, adding chunk of headers starting from the current chain's heigh
is impossible and (*Blockchain).AddHeaders doesn't have much sense.
Solution: verify header.PrevStateRoot only for header N+1. Rest of the
headers should be added without PrevStateRoot verification.
It was completely ruined by bf20db09e0. MPT was
updating bc.dao directly which shouldn't ever happen, it must write into the
same cache and then Persist these KVs as usual.
1. `System.Contract.CallNative` expects version on stack.
2. Actual method is determined based on current
instruction pointer.
3. Native hashes don't longer depend on NEF checksum.
We have additional logic for getting BaseExecFee policy value. This
logic should be moved to interop context instead of being in Policer,
because Policer is just an interface over Policy contract.
After moving this logic to interop context, we need to use it to define
BaseExecFee instead of (Policer).BaseExecFee. Thus, moving
(*Blockchain).GetPrice to (*Context).GetPrice is necessary.
C# does it this way now:
callFlags = !witness.VerificationScript.IsStandardContract() ? CallFlags.ReadStates : CallFlags.None
So non-standard scripts _always_ have access to state and standards ones just
don't care (their code is known and it doesn't touch state).
1. Initialization is performed via `Blockchain` methods.
2. Native Oracle contract updates list of oracle nodes
and in-fly requests in `PostPersist`.
3. RPC uses Oracle module directly.
Move oracleScript from global context to Oracle itself. We have the hash
already computed by NewContractMD, there is no need to repeat this
calculation.
Prices are defined in as a coefficients to `BaseExecFee` which
is defined by Policy contract (TBD later).
Native method prices are defined without need to multiply.
Previous commit sets AllowCall flag as required for Neo.Native.Call, but
invocation script was loaded with ReadStates flag => native contracts
verification failed.
Other contracts can also make use of AllowCall call flag.
We must be sure that stack has no other items before returning `false`
verification result. It is an error in both cases, but by preserving the
order we know exactly that it was correct `false` on stack.
When using contract-based verification it's important to load contract's hash
along with the script, otherwise it won't be valid.
Simplify things along the way.
We should return verbose transaction in case if it is in the mempool
from `getrawtransaction`. We also shouldn't return height from
`gettransactionheight` in case if transaction is in the mempool.
Attributes check should be done before adding transaction to
the pool, otherwise there might be a case when transaction with invalid
attributes is in the pool.
Follow missed change from neo-project/neo#1816 .
`None` may be used for any signer. Currently it is used
for sender to only pay fees, or to sign tx attributes.
It can't ever happen. We're guaranteed to have a consistent chain of headers
(we're verifying them above, if we're not verifying --- it's not our fault)
that starts at HeaderHeight that was actual when we were asking for it
previously. HeaderHeight can only move forward, so if that happened that would
be filtered out by the condition below and the first one can't happen. Though
to be absolutely sure change the second check to only pass "+1" headers (which
is what we want).
It's used in two places now:
* Blockchain.AddBlock()
This one does transaction duplication check of its own, doing it in
Verify() is just a waste of time. Merkle tree root hash value check is
still relevant though
* Block.DecodeBinary()
We're decoding blocks for the following purposes:
- on restore from dump
The block will be added to the chain via AddBlock() and that will do a
full check of it (if configured to do so)
- on retrieving the block from the DB (DAO)
We trust the DB, if it's gone wild, this check won't really help
- on receiving the block via P2P
It's gonna be put into block queue and then end up in AddBlock() which
will check it
- on receiving the block via RPC (submitblock)
It is to be passed into AddBlock()
- on receiving the block via RPC in a client
That's the only problematic case probably, but RPC client has to trust
the server and it can check for the signature if it really
cares. Or a separate in-client check might be added.
As we can see nothing really requires this verification to be done the way it
is now, AddBlock can just have a Merkle check and DecodeBinary can do fine
without it at all.
It's a no-op and there is nothing we can do about it, header contents could
only be checked against chain state, there is nothing to check for internal
consistency.
Now we have VerifyTx() and PoolTx() APIs that either verify transaction in
isolation or verify it against the mempool (either the primary one or the one
given) and then add it there. There is no possibility to check against the
mempool, but not add a transaction to it, but I doubt we really need it.
It allows to remove some duplication between old PoolTx and verifyTx where
they both tried to check transaction against mempool (verifying first and then
adding it). It also saves us utility token balance check because it's done by
the mempool anyway and we no longer need to do that explicitly in verifyTx.
It makes AddBlock() and verifyBlock() transaction's checks more correct,
because previously they could miss that even though sender S has enough
balance to pay for A, B or C, he can't pay for all of them.
Caveats:
* consensus is running concurrently to other processes, so things could
change while verifyBlock() is iterating over transactions, this will be
mitigated in subsequent commits
Improves TPS value for single node by at least 11%.
Fixes#667, fixes#668.
New transactions are added to the chain with blocks. If there is no
transaction X at height N in DAO, it could only be added with block N+1, so
it has to be present there. Therefore we can replace `dao.HasTransaction()`
check with a search through in-block transactions. HasTransaction() is nasty
in that it may add useless load the DB and this code is being run with a big
Blockchain lock held, so we don't want to be delayed here at all.
Improves single-node TPS by ~2%.
The end effect is almost as if `VerifyTransactions: false` was set in the
config, but without actually compromising the guarantees provided by it.
It almost doubles performance for single-mode benchmarks and makes block
processing smoother (more smaller blocks are being produced).
C# node is quite picky as it expects there to be exactly one value returned,
but our testchain actually adds 4 signatures for multisig cases instead of 3
which makes it technically incompatible with C# node.
We were checking blocked accounts twice which is obviously excessive. We also
have our accounts sorted, so we can rely on that in CheckPolicy(). It also
doesn't make much sense to check MaxBlockSystemFee in Blockchain code, policy
contract can handle that.
It no longer depends on blockchain state and there can't ever be an error, in
fact we can always iterate over signers, so copying these hashes doesn't make
much sense at all as well as sorting arrays in verifyTxWitnesses (witnesses
order must match signers order).
It's not needed any more with Go 1.13 as we have wrapping/unwrapping in base
packages. All errors.Wrap calls are replaced with fmt.Errorf, some strings are
improved along the way.
We need to compact our in-memory MPT from time to time, otherwise it quickly
fills up all available memory. This raises two obvious quesions --- when to do
that and to what level do that.
As for 'when', I think it's quite easy to use our regular persistence interval
as an anchor (and it also frees up some memory), but we can't do that in the
persistence routine itself because of synchronization issues (adding some
synchronization primitives would add some cost that I'd also like to avoid),
so do it indirectly by comparing persisted and current height in `storeBlock`.
Choosing proper level is another problem, but if we're to roughly estimate one
full branch node to use 1K of memory (usually it's way less than that) then we
can easily store 1K of these nodes and that gives us a depth of 10 for our
trie.
Signed-off-by: Evgenii Stratonikov <evgeniy@nspcc.ru>
This was differing from C# notion of PrevHash. It's not a previous root, but
rather a hash of the previous serialized MPTRoot structure (that is to be
signed by CNs).
Signed-off-by: Evgenii Stratonikov <evgeniy@nspcc.ru>
