* Clarifying/commenting FCU setup condition & small fixes, comments etc.
* Update some logging
* Reorg metrics updater and activation
* Better `async` responsiveness
why:
Block import does not allow `async` task activation while
executing. So allow potential switch after each imported
block (rather than a group of 32 blocks.)
* Handle resuming after previous sync followed by import
why:
In this case the ledger state is more recent than the saved
sync state. So this is considered a pristine sync where any
previous sync state is forgotten.
This fixes some assert thrown because of inconsistent internal
state at some point.
* Provide option for clearing saved beacon sync state before starting syncer
why:
It would resume with the last state otherwise which might be undesired
sometimes.
Without RPC available, the syncer typically stops and terminates with
the canonical head larger than the base/finalised head. The latter one
will be saved as database/ledger state and the canonical head as syncer
target. Resuming syncing here will repeat itself.
So clearing the syncer state can prevent from starting the syncer
unnecessarily avoiding useless actions.
* Allow workers to request syncer shutdown from within
why:
In one-trick-pony mode (after resuming without RPC support) the
syncer can be stopped from within soavoiding unnecessary polling.
In that case, the syncer can (theoretically) be restarted externally
with `startSync()`.
* Terminate beacon sync after a single run target is reached
why:
Stops doing useless polling (typically when there is no RPC available)
* Remove crufty comments
* Tighten state reload condition when resuming
why:
Some pathological case might apply if the syncer is stopped while the
distance between finalised block and head is very large and the FCU
base becomes larger than the locked finalised state.
* Verify that finalised number from CL is at least FCU base number
why:
The FCU base number is determined by the database, non zero if
manually imported. The finalised number is passed via RPC by the CL
node and will increase over time. Unless fully synced, this number
will be pretty low.
On the other hand, the FCU call `forkChoice()` will eventually fail
if the `finalizedHash` argument refers to something outside the
internal chain starting at the FCU base block.
* Remove support for completing interrupted sync without RPC support
why:
Simplifies start/stop logic
* Rmove unused import
* prefer the spec-derived name where possible
* don't pass stateRoot to LedgerRef and friends (it doesn't do anything)
* add deprecation warning in graphql - it needs updating to use
forkedchain instead
When `nimbus import` runs, we end up with a database without MPT roots
leading to long startup times the first time one is needed.
Computing the state root is slow because the on-disk order based on
VertexID sorting does not match the trie traversal order and therefore
makes lookups inefficent.
Here we introduce a helper that speeds up this computation by traversing
the trie in on-disk order and computing the trie hashes bottom up
instead - even though this leads to some redundant reads of nodes that
we cannot yet compute, it's still a net win as leaves and "bottom"
branches make up the majority of the database.
This PR also addresses a few other sources of inefficiency largely due
to the separation of AriKey and AriVtx into their own column families.
Each column family is its own LSM tree that produces hundreds of SST
filtes - with a limit of 512 open files, rocksdb must keep closing and
opening files which leads to expensive metadata reads during random
access.
When rocksdb makes a lookup, it has to read several layers of files for
each lookup. Ribbon filters to skip over files that don't have the
requested data but when these filters are not in memory, reading them is
slow - this happens in two cases: when opening a file and when the
filter has been evicted from the LRU cache. Addressing the open file
limit solves one source of inefficiency, but we must also increase the
block cache size to deal with this problem.
* rocksdb.max_open_files increased to 2048
* per-file size limits increased so that fewer files are created
* WAL size increased to avoid partial flushes which lead to small files
* rocksdb block cache increased
All these increases of course lead to increased memory usage, but at
least performance is acceptable - in the future, we'll need to explore
options such as joining AriVtx and AriKey and/or reducing the row count
(by grouping branch layers under a single vertexid).
With this PR, the mainnet state root can be computed in ~8 hours (down
from 2-3 days) - not great, but still better.
Further, we write all keys to the database, also those that are less
than 32 bytes - because the mpt path is part of the input, it is very
rare that we actually hit a key like this (about 200k such entries on
mainnet), so the code complexity is not worth the benefit really, in the
current database layout / design.
* remove redundant abstraction
* fix misleading raises - the implementation actually swallows errors or
panics (depending on how many other layers of abstraction we penetrate
before detecting it)
* blocks can be bigger than the default 1mb when json-rpc-encoded - this
happens on sepolia for example
* json-rpc bump improves debug logging and fixes a number of bugs
* json-serialization bump fixes a crash on invalid arrays in json data
At some point, it would probably be better to compute the maximum block
size from actual block constraints, though this is somewhat tricky and
depends on gas limits etc. Until then, 16mb should be plenty.
With this, sepolia can be synced :)
* Update comments & logs
* Do not start beacon sync unless there is possibly something to do
why:
It would continue polling without having any effect other than
logging. Now it will not start unless there is RPC available
or there was a previously interrupted sync to be resumed.
* Accept finalised hash from RPC with the canon header as well
* Reorg internal sync descriptor(s)
details:
Update target from RPC to provide the `consensus header` as well as
the `finalised` block number
why:
Prepare for using `importBlock()` instead of `persistBlocks()`
* Cosmetic updates
details:
+ Collect all pretty printers in `helpers.nim`
+ Remove unused return codes from function prototype
* Use `importBlock()` + `forkChoice()` rather than `persistBlocks()`
* Update logging and metrics
* Update docu
* Update `ForkedChainRef` constructor
why:
Initialisation is based on the canonical head which is always zero
after resuming a stopped `ForkedChainRef` based import.
* Update new-base calculator
why:
There is some ambiguous code which might not do what the comment
implies. In short, an unsigned condition like `2u - 3u < 1u => false`
is coded where the comment suggests that `2 - 3 < 1 => true` is meant.
This patch fixes notorious crashes when resuming import after a stop.
* partial commit
* fixes
* remove converters too
* revert changes on nimbus_verified_proxy
* revert changes in converter
* revert changes(re-xport) in rpc_types
* update copyright year
* replace types in other binaries
* chain config bug
* fix rebase conflict imcomplete buffer
* fix more rebase buffers
* remove ditto types and converters
* fix the tests
* update copyright year
* rename nimbus binary to nimbus_execution_client
* additional replacements
* makefile and dockerfile
* fix ci building errors
* github workflows
* improved Makefile target
---------
Co-authored-by: Pedro Miranda <pedro.miranda@nimbus.team>
* Fix fringe condition clarifying how to handle an empty range
why:
The `interval_set` module would treat an undefined interval construct
`[2,1]` as`[2,2]` (the right bound being `max(2,1)`.)
* Use the `consensus head` rather than the `finalised` block as sync target
why:
The former is ahead of the `finalised` block.
* In ctx descriptor rename `final` field to `target`
* Update docu, rename `F` -> `T`
* bump nimbus-build-system to use Nim v2.0.10
* 2.0.10 fixes
* fluffy linting
* make trivial change which should trigger whole-nimbus+fluffy rebuild/ci
* Nim v2.0.10 chronicles.error/macros.error ambiguity workaround
* another contentType enum specifier
* fluffy linting
* Fix eth/common & web3 related deprecation warnings for fluffy
This commit uses the new types in the new eth/common/ structure
to remove deprecation warnings.
It is however more than just a mass replace as also all places
where eth/common or eth/common/eth_types or eth/common/eth_types_rlp
got imported have been revised and adjusted to a better per submodule
based import.
There are still a bunch of toMDigest deprecation warnings but that
convertor is not needed for fluffy code anymore so in theory it
should not be used (bug?). It seems to still get imported via export
leaks ffrom imported nimbus code I think.
* Address review comments
* Remove two more unused eth/common imports
* fix: rpc can't serve blocks in db
* shift db access to forkedchainref
* cleanup
* kurtosis test fix, should fail + eth_getTransactionReceipt
* remove kurtosis not + cleanup
* alter CI check to pass
* optimize impl
* cleanup
* fix loop case
* Rename `base` -> `coupler`, `B` -> `C`
why:
Glossary: The jargon `base` is used for the `base state` block number
which can be smaller than what is now the `coupler`.
* Rename `global state` -> `base`, `T` -> `B`
why:
See glossary
* Rename `final` -> `end`, `F` -> `E`
why:
See glossary. Previously, `final` denoted some finalised block but not
`the finalised` block from the glossary (which is maximal.)
* Properly name finalised block as such, rename `Z` -> `F`
why:
See glossary
* Rename `least` -> `dangling`, `L` -> `D`
* Metrics update (variables not covered yet)
* Docu update and corrections
* Logger updates
* Remove obsolete `skeleton*Key` kvt columns from `storage_types` module
* Remove `--sync-mode` option from nimbus config
why:
Currently there is only one sync mode available.
* Rename `flare` -> `beacon`, but not base module folder and nim source
why:
The name `flare` was used do designate an alternative `beacon` mode that.
Leaving the base folder and source as-is for a moment, makes it easier
to read change diffs.
* Rename `flare` base module folder and nim source: `flare` -> `beacon`
* Dissolve legacy `sync/types.nim` into `*/eth/eth_types.nim`
* Flare sync: Simplify scheduler and remove `runSingle()` method
why:
`runSingle()` is not used anymore (main purpose was for negotiating
best headers in legacy full sync.)
Also, `runMulti()` was renamed `runPeer()`
* Flare sync: Move `chain` field from `sync_desc` -> `worker_desc`
* Flare sync: Remove handler descriptor lateral reference
why:
Not used anymore. It enabled to turn on/off eth handler activity with
regards to the sync state, i.e.from with in the sync worker.
* Flare sync: Update `Hash256` and other deprecated `std/eth` symbols
* Protocols: Update `Hash256` and other deprecated `std/eth` symbols
* Eth handler: Update `Hash256` and other deprecated `std/eth` symbols
* Update flare TODO
* Remove redundant `sync/type` import
why:
The import module `type` has been removed
* Remove duplicate implementation
This is a minimal set of changes to make things work with the new types
in nim-eth - this is the minimal PR that merely resolves
incompatibilities while the full change set would include more cleanup
and migration.
* Cosmetics, small fixes, add stashed headers verifier
* Remove direct `Era1` support
why:
Era1 is indirectly supported by using the import tool before syncing.
* Clarify database persistent save function.
why:
Function relied on the last saved state block number which was wrong.
It now relies on the tx-level. If it is 0, then data are saved directly.
Otherwise the task that owns the tx will do it.
* Extracted configuration constants into separate file
* Enable single peer mode for debugging
* Fix peer losing issue in multi-mode
details:
Running concurrent download peers was previously programmed as running
a batch downloading and storing ~8k headers and then leaving the `async`
function to be restarted by a scheduler.
This was unfortunate because of occasionally occurring long waiting
times for restart.
While the time gap until restarting were typically observed a few
millisecs, there were always a few outliers which well exceed several
seconds. This seemed to let remote peers run into timeouts.
* Prefix function names `unprocXxx()` and `stagedYyy()` by `headers`
why:
There will be other `unproc` and `staged` modules.
* Remove cruft, update logging
* Fix accounting issue
details:
When staging after fetching headers from the network, there was an off
by 1 error occurring when the result was by one smaller than requested.
Also, a whole range was mis-accounted when a peer was terminating
connection immediately after responding.
* Fix slow/error header accounting when fetching
why:
Originally set for detecting slow headers in a row, the counter
was wrongly extended to general errors.
* Ban peers for a while that respond with too few headers continuously
why:
Some peers only returned one header at a time. If these peers sit on a
farm, they might collectively slow down the download process.
* Update RPC beacon header updater
why:
Old function hook has slightly changed its meaning since it was used
for snap sync. Also, the old hook is used by other functions already.
* Limit number of peers or set to single peer mode
details:
Merge several concepts, single peer mode being one of it.
* Some code clean up, fixings for removing of compiler warnings
* De-noise header fetch related sources
why:
Header download looks relatively stable, so general debugging is not
needed, anymore. This is the equivalent of removing the scaffold from
the part of the building where work has completed.
* More clean up and code prettification for headers stuff
* Implement body fetch and block import
details:
Available headers are used stage blocks by combining existing headers
with newly fetched blocks. Then these blocks are imported/executed via
`persistBlocks()`.
* Logger cosmetics and cleanup
* Remove staged block queue debugging
details:
Feature still available, just not executed anymore
* Docu, logging update
* Update/simplify `runDaemon()`
* Re-calibrate block body requests and soft config for import blocks batch
why:
* For fetching, larger fetch requests are mostly truncated anyway on
MainNet.
* For executing, smaller batch sizes reduce the memory needed for the
price of longer execution times.
* Update metrics counters
* Docu update
* Some fixes, formatting updates, etc.
* Update `borrowed` type: uint -. uint64
also:
Always convert to `uint64` rather than `uint` where appropriate
* ForkedChainRef.forkchoice: Skip newBase calculation and skip chain finalization if finalizedHash is zero
* Fix ForkedChainRef.forkChoice: do nothing if headHash is the same with cursorHash
* Fix stupid bug in engine API FCU when calling ForkedChainRef.forkChoice
* Wire RPC server API to nimbus RPC manager
* Add test case
* Use default(Hash256) in ForkedChainRef
* init style for Hash256
https://github.com/status-im/nim-eth/pull/733 updates `Hash256` to
become an array instead of an object - unfortunately, nim does not allow
constructing arrays with `name()`, so this PR changes it to `default`
which works with both.
* lint
* fix: nimbus state ahead of era history
* comments
* fix: suggestions
* fix: messages
* fix edge case resume
* check from last file
* formatting
* fix: typo
* fix: unwanted quit before rlp import
* batch database key writes during `computeKey` calls
* log progress when there are many keys to update
* avoid evicting the vertex cache when traversing the trie for key
computation purposes
* avoid storing trivial leaf hashes that directly can be loaded from the
vertex
* Add missing leaf cache update when a leaf turns to a branch with two
leaves (on merge) and vice versa (on delete) - this could lead to stale
leaves being returned from the cache causing validation failures - it
didn't happen because the leaf caches were not being used efficiently :)
* Replace `seq` with `ArrayBuf` in `Hike` allowing it to become
allocation-free - this PR also works around an inefficiency in nim in
returning large types via a `var` parameter
* Use the leaf cache instead of `getVtxRc` to fetch recent leaves - this
makes the vertex cache more efficient at caching branches because fewer
leaf requests pass through it.
The storage leaf cache was being circumvented when actually fetching
leaves and was instead only being filled with items :/
Also avoids an expensive copy when fetching account data (broadly,
variant objects are comparatively expensive to copy and fetching
accounts is a hotspot)
* move pfx out of variant which avoids pointless field type panic checks
and copies on access
* make `VertexRef` a non-inheritable object which reduces its memory
footprint and simplifies its use - it's also unclear from a semantic
point of view why inheritance makes sense for storing keys
Compared to `keyed_queue`, `minilru` uses significantly less memory, in
particular for the 32-byte hash keys where `kq` stores several copies of
the key redundantly.
detail:
For practical reasons, ifsuch an account is asked for a slot, an empty
proof list is returned. It is up to the user to provide an account
proof that shows that there is no storage tree.
* Reverse order in staged blob lists
why:
having the largest block number with the least header list index `0`
makes it easier to grow the list with parent headers, i.e. decreasing
block numbers.
* Set a header response threshold when to ditch peer
* Refactor extension of staged header chains record
why:
Was cobbled together as a proof of concept after several approaches of
how to run the download.
* TODO update
* Make debugging code independent of `release` flag
* Update import from jacek
* Block header download starting at Beacon down to Era1
details:
The header download implementation is intended to be completed to a
full sync facility.
Downloaded block headers are stored in a `CoreDb` table. Later on they
should be fetched, complemented by a block body, executed/imported,
and deleted from the table.
The Era1 repository may be partial or missing. Era1 headers are neither
downloaded nor stored on the `CoreDb` table.
Headers are downloaded top down (largest block number first) using the
hash of the block header by one peer. Other peers fetch headers
opportunistically using block numbers
Observed download times for 14m `MainNet` headers varies between 30min
and 1h (Era1 size truncated to 66m blocks.), full download 52min
(anectdotal.) The number of peers downloading concurrently is crucial
here.
* Activate `flare` by command line option
* Fix copyright year
Saving both memory and processing, we can move entries from one
savepoint to another, specially when the target is empty as it often is
during transaction processing
* replace rocksdb row cache with larger rdb lru caches - these serve the
same purpose but are more efficient because they skips serialization,
locking and rocksdb layering
* don't append fresh items to cache - this has the effect of evicting
the existing items and replacing them with low-value entries that might
never be read - during write-heavy periods of processing, the
newly-added entries were evicted during the store loop
* allow tuning rdb lru size at runtime
* add (hidden) option to print lru stats at exit (replacing the
compile-time flag)
pre:
```
INF 2024-09-03 15:07:01.136+02:00 Imported blocks
blockNumber=20012001 blocks=12000 importedSlot=9216851 txs=1837042
mgas=181911.265 bps=11.675 tps=1870.397 mgps=176.819 avgBps=10.288
avgTps=1574.889 avgMGps=155.952 elapsed=19m26s458ms
```
post:
```
INF 2024-09-03 13:54:26.730+02:00 Imported blocks
blockNumber=20012001 blocks=12000 importedSlot=9216851 txs=1837042
mgas=181911.265 bps=11.637 tps=1864.384 mgps=176.250 avgBps=11.202
avgTps=1714.920 avgMGps=169.818 elapsed=17m51s211ms
```
9%:ish import perf improvement on similar mem usage :)
* Cosmetics, spelling, etc.
* Aristo: make sure that a save cycle always commits even when empty
why:
If `Kvt` is tied to the `Aristo` DB save cycle, then this save cycle
must also be committed if there is no data to save for `Aristo`.
Otherwise this will lead to excessive core memory use with some fringe
condition where Eth headers (or blocks) are downloaded while syncing
and not really stored on disk.
* CoreDb: Correct persistent save mode
why:
Saving `Kvt` first is seen as a harbinger (or canary) for `Aristo` as
both run in sync. If `Kvt` succeeds saving first, so must be `Aristo`
next. Other than this is a defect.
* Wiring ForkedChainRef to other components
- Disable majority of hive simulators
- Only enable pyspec_sim for the moment
- The pyspec_sim is using a smaller RPC service wired to ForkedChainRef
- The RPC service will gradually grow
* Addressing PR review
* Fix test_beacon/setup_env
* Enable consensus_sim (#2441)
* Enable consensus_sim
* Remove isFile check
* Enable Engine API jwt auth tests and exchange cap tests
* Enable engine api in build_sim.sh
* Wire ForkedChainRef to Engine API newPayload
* Wire Engine API getBodies to ForkedChainRef
* Wire Engine API api_forkchoice to ForkedChainRef
* Wire more RPC methods to ForkedChainRef
* Implement eth_syncing
* Implement eth_call and eth_getlogs
* TxPool: simplify smartHead
* Fix smartHead usage
* Fix txpool headDiff
* Remove hasBlockHeader and use headerExists
* Addressing review
This is a first step towards measuring the efficiency of the LRU caches
over time - metrics can be collected during import or when running
regulary.
Since `nim-metrics` carries some overhead for its default way of
reporting metrics, this PR implements a custom collector over atomic
counters, given that this is one of the hottest spots in the block
processing pipeline.
Using a compile-time flag, the same metrics can be printed on exit which
is useful when comparing different strategies for caching - here's a
recent run over blocks 16000001-1616384 - this is a good candidate to
expose in a better way in the future, maybe:
```
state vtype miss hit total hitrate
Account Leaf 4909417 4466215 9375632 47.64%
Account Branch 20742574 72015123 92757697 77.64%
World Leaf 940483 1140946 2081429 54.82%
World Branch 8224151 131496580 139720731 94.11%
all all 34816625 209118864 243935489 85.73%
```
* pre-allocate `blobify` data and remove redundant error handling
(cannot fail on correct data)
* use threadvar for temporary storage when decoding rdb, avoiding
closure env
* speed up database walkers by avoiding many temporaries
~5% perf improvement on block import, 100x on database iteration (useful
for building analysis tooling)
Tested up to block ~14m, zstd uses ~12% less space which seems to result
in a small:ish (2-4%) performance improvement on block import speed -
this seems like a better baseline for more extensive testing in the
future.
Pre: 57383308 kb
Post: 50831236 kb
* bump metrics
* Remove cruft
* Cosmetics, update some logging, noise control
* Renamed `CoreDb` function `hasKey` => `hasKeyRc` and provided `hasKey`
why:
Currently, `hasKey` returns a `Result[]` rather than a `bool` which
is what one would expect from a function prototype of this name.
This was a bit of an annoyance and cost unnecessary attention.
In the current VM opcode dispatcher, a two-level case statement is
generated that first matches the opcode and then uses another nested
case statement to select the actual implementation based on which fork
it is, causing the dispatcher to grow by `O(opcodes) * O(forks)`.
The fork does not change between instructions causing significant
inefficiency for this approach - not only because it repeats the fork
lookup but also because of code size bloat and missed optimizations.
A second source of inefficiency in dispatching is the tracer code which
in the vast majority of cases is disabled but nevertheless sees multiple
conditionals being evaluated for each instruction only to remain
disabled throughout exeuction.
This PR rewrites the opcode dispatcher macro to generate a separate
dispatcher for each fork and tracer setting and goes on to pick the
right one at the start of the computation.
This has many advantages:
* much smaller dispatcher
* easier to compile
* better inlining
* fewer pointlessly repeated instruction
* simplified macro (!)
* slow "low-compiler-memory" dispatcher code can be removed
Net block import improvement at about 4-6% depending on the contract -
synthetic EVM benchmnarks would show an even better result most likely.
Because EthBlock is quite large, the stack usage that results from the
multiple copies (temporary and not) present in the import command is
larger than it should be - this PR moves some of that data to a closure
environment allocated once per EthBlock - a larger restructuring of the
code is due but in the meantime, this simple change speeds up garbage
collection a little bit.
* Remove redundant `eth/68` message and clean up docu
details:
There is only eth/68 available at the moment
* Allow to turn on chronicles line number logging in `Makefile`
* Accept (and forget) tx hashes announcements
why:
Does no harm to just ignore it at the moment
* Bump nim-eth (rlp fix)
When the stack has an empty layer on top, there's no need to copy the
contents of `top` to it since it would be the same.
~13% processing saved (!)
pre
```
INF 2024-08-17 19:11:31.748+02:00 Imported blocks
blockNumber=18667648 blocks=12000 importedSlot=7860043 txs=1797812
mgas=181135.177 bps=8.763 tps=1375.062 mgps=132.125 avgBps=6.798
avgTps=1018.501 avgMGps=102.617 elapsed=29m25s154ms
```
post
```
INF 2024-08-17 18:22:52.513+02:00 Imported blocks
blockNumber=18667648 blocks=12000 importedSlot=7860043 txs=1797812
mgas=181135.177 bps=9.648 tps=1513.961 mgps=145.472 avgBps=7.876
avgTps=1179.998 avgMGps=118.888 elapsed=25m23s572ms
```
The reverse slot hash mechanism causes quite a bit of database traffic
but is broadly not useful except for iterating the storage of an
account, something that a validator never does (it's used by the
tracers).
This flag adds one more thing that is not stored in the database, to be
explored more comprehensively when designing full, validator and archive
modes with different pruning options in the future.
`ldb` says this is 60gb of data (!):
```
ldb --db=. --ignore_unknown_options --column_family=KvtGen approxsize
--hex --from=0x05
--to=0x05ffffffffffffffffffffffffffffffffffffffffffffff
66488353954
```
andri lim
tersec
Jacek Sieka
Jordan Hrycaj
Advaita Saha
Chirag Parmar
Pedro Miranda
Kim De Mey
web3-developer