This change introduces garbage collection for the light client. Historical
chain data is deleted periodically. If you want to disable the GC, use
the --light.nopruning flag.
This fixes two issues with state sync restarts:
When sync restarts with a new root, some peers can have in-flight requests.
Since all peers with active requests were marked idle when exiting sync,
the new sync would schedule more requests for those peers. When the
response for the earlier request arrived, the new sync would reject it and
mark the peer idle again, rendering the peer useless until it disconnected.
The other issue was that peers would not be marked idle when they had
delivered a response, but the response hadn't been processed before
restarting the state sync. This also made the peer useless because it
would be permanently marked busy.
Co-authored-by: Felix Lange <fjl@twurst.com>
This adds a test suite for discovery v4. The test suite is a port of the Hive suite for
discovery, and will replace the current suite on Hive soon-ish. The tests can be
run locally with this command:
devp2p discv4 test -remote enode//...
Co-authored-by: Felix Lange <fjl@twurst.com>
This change further improves the performance of RLP encoding by removing
allocations for big.Int and [...]byte types. I have added a new benchmark
that measures RLP encoding of types.Block to verify that performance is
improved.
Exposing /debug/metrics and /debug/metrics/prometheus was dependent
on --pprof, which also exposes other HTTP APIs. This change makes it possible
to run the metrics server on an independent endpoint without enabling pprof.
This PR reduces the bandwidth used by the light client to compute the
recommended gas price. The current mechanism for suggesting the price is:
- retrieve recent 20 blocks
- get the lowest gas price of these blocks
- sort the price array and return the middle(60%) one
This works for full nodes, which have all blocks available locally.
However, this is very expensive for the light client because the light
client needs to retrieve block bodies from the network.
The PR changes the default options for light client. With the new config,
the light client only retrieves the two latest blocks, but in order to
collect more sample transactions, the 3 lowest prices are collected from
each block.
This PR also changes the behavior for empty blocks. If the block is empty,
the lastest price is reused for sampling.
* core: use uint64 for total tx costs instead of big.Int
* core: added local tx pool test case
* core, crypto: various allocation savings regarding tx handling
* Update core/tx_list.go
* core: added tx.GasPriceIntCmp for comparison without allocation
adds a method to remove unneeded allocation in comparison to tx.gasPrice
* core: handle pools full of locals better
* core/tests: benchmark for tx_list
* core/txlist, txpool: save a reheap operation, avoid some bigint allocs
Co-authored-by: Martin Holst Swende <martin@swende.se>
List headers made up 11% of all allocations during sync. This change
removes most of those allocations by keeping the list header values
cached in the encoder buffer instead. Since encoder buffers are pooled,
list headers are no longer allocated in the common case where an
encoder buffer is available for reuse.
Co-authored-by: Felix Lange <fjl@twurst.com>
* eth/downloaded: fixed datarace between synchronize and Progress
There was a race condition between `downloader.synchronize()` and `Progress` `syncWithPeer` `fetchHeight` `findAncestors` and `processHeaders`
This PR changes the behavior of the downloader a bit.
Previously the functions `Progress` `syncWithPeer` `fetchHeight` `findAncestors` and `processHeaders` read the syncMode anew within their loops. Now they read the syncMode at the start of their function and don't change it during their runtime.
* eth/downloaded: comment
* eth/downloader: added comment
* core, crypto: various allocation savings regarding tx handling
* core: reduce allocs for gas price comparison
This change reduces the allocations needed for comparing different transactions to each other.
A call to `tx.GasPrice()` copies the gas price as it has to be safe against modifications and
also needs to be threadsafe. For comparing and ordering different transactions we don't need
these guarantees
* core: added tx.GasPriceIntCmp for comparison without allocation
adds a method to remove unneeded allocation in comparison to tx.gasPrice
* core/types: pool legacykeccak256 objects in rlpHash
rlpHash is by far the most used function in core that allocates a legacyKeccak256 object on each call.
Since it is so widely used it makes sense to add pooling here so we relieve the GC.
On my machine these changes result in > 100 MILLION less allocations and > 30 GB less allocated memory.
* reverted some changes
* reverted some changes
* trie: use crypto.KeccakState instead of replicating code
Co-authored-by: Martin Holst Swende <martin@swende.se>
This PR implements the EVM state transition tool, which is intended
to be the replacement for our retesteth client implementation.
Documentation is present in the cmd/evm/README.md file.
Co-authored-by: Felix Lange <fjl@twurst.com>
Without use of global, these flags didn't actually modify
miner configuration, since we weren't grabbing from the
proper context scope, which should be global (vs. subcommand).
Signed-off-by: meows <b5c6@protonmail.com>
The import command should not return a 0 status
code if the import finishes prematurely becaues
of an import error.
Returning the error causes the program to exit with 1
if the err is non nil.
Signed-off-by: meows <b5c6@protonmail.com>
* core/vm: fix incorrect computation of discount
During testing on Yolov1 we found that the way geth calculates the discount
is not in line with the specification. Basically what we did is calculate
128 * Bls12381GXMulGas * discount / 1000 whenever we received more than 128 pairs
of values. Correct would be to calculate k * Bls12381... for k > 128.
* core/vm: better logic for discount calculation
* core/vm: better calculation logic, added worstcase benchmarks
* core/vm: better benchmarking logic
The ancients variable in the freezer is a list of hashes, which
identifies all of the hashes to be frozen. The slice is being allocated
with a capacity of `limit`, which is the number of the last block
this batch will attempt to add to the freezer. That means we are
allocating memory for all of the blocks in the freezer, not just
the ones to be added.
If instead we allocate `limit - f.frozen`, we will only allocate
enough space for the blocks we're about to add to the freezer. On
mainnet this reduces usage by about 320 MB.
