2023-06-20 13:26:25 +00:00
|
|
|
# nimbus-eth1
|
2024-03-14 22:17:43 +00:00
|
|
|
# Copyright (c) 2023-2024 Status Research & Development GmbH
|
2023-06-20 13:26:25 +00:00
|
|
|
# Licensed under either of
|
|
|
|
# * Apache License, version 2.0, ([LICENSE-APACHE](LICENSE-APACHE) or
|
|
|
|
# http://www.apache.org/licenses/LICENSE-2.0)
|
|
|
|
# * MIT license ([LICENSE-MIT](LICENSE-MIT) or
|
|
|
|
# http://opensource.org/licenses/MIT)
|
|
|
|
# at your option. This file may not be copied, modified, or distributed
|
|
|
|
# except according to those terms.
|
|
|
|
|
|
|
|
## Rocksdb backend for Aristo DB
|
|
|
|
## =============================
|
|
|
|
##
|
|
|
|
## The iterators provided here are currently available only by direct
|
|
|
|
## backend access
|
|
|
|
## ::
|
|
|
|
## import
|
|
|
|
## aristo/aristo_init,
|
|
|
|
## aristo/aristo_init/aristo_rocksdb
|
|
|
|
##
|
|
|
|
## let rc = AristoDb.init(BackendRocksDB, "/var/tmp")
|
|
|
|
## if rc.isOk:
|
|
|
|
## let be = rc.value.to(RdbBackendRef)
|
|
|
|
## for (n, key, vtx) in be.walkVtx:
|
|
|
|
## ...
|
|
|
|
##
|
|
|
|
{.push raises: [].}
|
|
|
|
|
|
|
|
import
|
|
|
|
eth/common,
|
|
|
|
rocksdb,
|
2023-09-12 18:45:12 +00:00
|
|
|
results,
|
2023-06-22 11:13:24 +00:00
|
|
|
../aristo_desc,
|
2023-08-25 22:53:59 +00:00
|
|
|
../aristo_desc/desc_backend,
|
2023-11-08 12:18:32 +00:00
|
|
|
../aristo_blobify,
|
2023-08-25 22:53:59 +00:00
|
|
|
./init_common,
|
2024-06-05 15:08:29 +00:00
|
|
|
./rocks_db/[rdb_desc, rdb_get, rdb_init, rdb_put, rdb_walk],
|
|
|
|
../../opts
|
2023-06-20 13:26:25 +00:00
|
|
|
|
2024-04-16 20:39:11 +00:00
|
|
|
const
|
|
|
|
extraTraceMessages = false
|
|
|
|
## Enabled additional logging noise
|
2023-06-20 13:26:25 +00:00
|
|
|
|
|
|
|
type
|
2023-06-30 22:22:33 +00:00
|
|
|
RdbBackendRef* = ref object of TypedBackendRef
|
2023-06-20 13:26:25 +00:00
|
|
|
rdb: RdbInst ## Allows low level access to database
|
|
|
|
|
|
|
|
RdbPutHdlRef = ref object of TypedPutHdlRef
|
|
|
|
|
2024-04-16 20:39:11 +00:00
|
|
|
when extraTraceMessages:
|
|
|
|
import chronicles
|
2023-06-20 13:26:25 +00:00
|
|
|
|
2024-04-16 20:39:11 +00:00
|
|
|
logScope:
|
|
|
|
topics = "aristo-backend"
|
2023-06-20 13:26:25 +00:00
|
|
|
|
|
|
|
# ------------------------------------------------------------------------------
|
|
|
|
# Private helpers
|
|
|
|
# ------------------------------------------------------------------------------
|
|
|
|
|
|
|
|
proc newSession(db: RdbBackendRef): RdbPutHdlRef =
|
|
|
|
new result
|
|
|
|
result.TypedPutHdlRef.beginSession db
|
|
|
|
|
|
|
|
proc getSession(hdl: PutHdlRef; db: RdbBackendRef): RdbPutHdlRef =
|
|
|
|
hdl.TypedPutHdlRef.verifySession db
|
|
|
|
hdl.RdbPutHdlRef
|
|
|
|
|
|
|
|
proc endSession(hdl: PutHdlRef; db: RdbBackendRef): RdbPutHdlRef =
|
|
|
|
hdl.TypedPutHdlRef.finishSession db
|
|
|
|
hdl.RdbPutHdlRef
|
|
|
|
|
|
|
|
# ------------------------------------------------------------------------------
|
|
|
|
# Private functions: interface
|
|
|
|
# ------------------------------------------------------------------------------
|
|
|
|
|
|
|
|
proc getVtxFn(db: RdbBackendRef): GetVtxFn =
|
|
|
|
result =
|
2024-09-20 05:43:53 +00:00
|
|
|
proc(rvid: RootedVertexID, flags: set[GetVtxFlag]): Result[VertexRef,AristoError] =
|
2023-06-20 13:26:25 +00:00
|
|
|
|
|
|
|
# Fetch serialised data record
|
2024-09-20 05:43:53 +00:00
|
|
|
let vtx = db.rdb.getVtx(rvid, flags).valueOr:
|
2024-04-16 20:39:11 +00:00
|
|
|
when extraTraceMessages:
|
2024-09-04 13:48:38 +00:00
|
|
|
trace logTxt "getVtxFn() failed", rvid, error=error[0], info=error[1]
|
2024-04-16 20:39:11 +00:00
|
|
|
return err(error[0])
|
2023-06-20 13:26:25 +00:00
|
|
|
|
2024-06-10 12:04:22 +00:00
|
|
|
if vtx.isValid:
|
|
|
|
return ok(vtx)
|
2023-06-20 13:26:25 +00:00
|
|
|
|
|
|
|
err(GetVtxNotFound)
|
|
|
|
|
|
|
|
proc getKeyFn(db: RdbBackendRef): GetKeyFn =
|
|
|
|
result =
|
Pre-allocate vids for branches (#2882)
Each branch node may have up to 16 sub-items - currently, these are
given VertexID based when they are first needed leading to a
mostly-random order of vertexid for each subitem.
Here, we pre-allocate all 16 vertex ids such that when a branch subitem
is filled, it already has a vertexid waiting for it. This brings several
important benefits:
* subitems are sorted and "close" in their id sequencing - this means
that when rocksdb stores them, they are likely to end up in the same
data block thus improving read efficiency
* because the ids are consequtive, we can store just the starting id and
a bitmap representing which subitems are in use - this reduces disk
space usage for branches allowing more of them fit into a single disk
read, further improving disk read and caching performance - disk usage
at block 18M is down from 84 to 78gb!
* the in-memory footprint of VertexRef reduced allowing more instances
to fit into caches and less memory to be used overall.
Because of the increased locality of reference, it turns out that we no
longer need to iterate over the entire database to efficiently generate
the hash key database because the normal computation is now faster -
this significantly benefits "live" chain processing as well where each
dirtied key must be accompanied by a read of all branch subitems next to
it - most of the performance benefit in this branch comes from this
locality-of-reference improvement.
On a sample resync, there's already ~20% improvement with later blocks
seeing increasing benefit (because the trie is deeper in later blocks
leading to more benefit from branch read perf improvements)
```
blocks: 18729664, baseline: 190h43m49s, contender: 153h59m0s
Time (total): -36h44m48s, -19.27%
```
Note: clients need to be resynced as the PR changes the on-disk format
R.I.P. little bloom filter - your life in the repo was short but
valuable
2024-12-04 10:42:04 +00:00
|
|
|
proc(rvid: RootedVertexID, flags: set[GetVtxFlag]): Result[(HashKey, VertexRef),AristoError] =
|
2023-06-20 13:26:25 +00:00
|
|
|
|
|
|
|
# Fetch serialised data record
|
Pre-allocate vids for branches (#2882)
Each branch node may have up to 16 sub-items - currently, these are
given VertexID based when they are first needed leading to a
mostly-random order of vertexid for each subitem.
Here, we pre-allocate all 16 vertex ids such that when a branch subitem
is filled, it already has a vertexid waiting for it. This brings several
important benefits:
* subitems are sorted and "close" in their id sequencing - this means
that when rocksdb stores them, they are likely to end up in the same
data block thus improving read efficiency
* because the ids are consequtive, we can store just the starting id and
a bitmap representing which subitems are in use - this reduces disk
space usage for branches allowing more of them fit into a single disk
read, further improving disk read and caching performance - disk usage
at block 18M is down from 84 to 78gb!
* the in-memory footprint of VertexRef reduced allowing more instances
to fit into caches and less memory to be used overall.
Because of the increased locality of reference, it turns out that we no
longer need to iterate over the entire database to efficiently generate
the hash key database because the normal computation is now faster -
this significantly benefits "live" chain processing as well where each
dirtied key must be accompanied by a read of all branch subitems next to
it - most of the performance benefit in this branch comes from this
locality-of-reference improvement.
On a sample resync, there's already ~20% improvement with later blocks
seeing increasing benefit (because the trie is deeper in later blocks
leading to more benefit from branch read perf improvements)
```
blocks: 18729664, baseline: 190h43m49s, contender: 153h59m0s
Time (total): -36h44m48s, -19.27%
```
Note: clients need to be resynced as the PR changes the on-disk format
R.I.P. little bloom filter - your life in the repo was short but
valuable
2024-12-04 10:42:04 +00:00
|
|
|
let key = db.rdb.getKey(rvid, flags).valueOr:
|
2024-04-16 20:39:11 +00:00
|
|
|
when extraTraceMessages:
|
2024-09-04 13:48:38 +00:00
|
|
|
trace logTxt "getKeyFn: failed", rvid, error=error[0], info=error[1]
|
2024-04-16 20:39:11 +00:00
|
|
|
return err(error[0])
|
2023-06-20 13:26:25 +00:00
|
|
|
|
Pre-allocate vids for branches (#2882)
Each branch node may have up to 16 sub-items - currently, these are
given VertexID based when they are first needed leading to a
mostly-random order of vertexid for each subitem.
Here, we pre-allocate all 16 vertex ids such that when a branch subitem
is filled, it already has a vertexid waiting for it. This brings several
important benefits:
* subitems are sorted and "close" in their id sequencing - this means
that when rocksdb stores them, they are likely to end up in the same
data block thus improving read efficiency
* because the ids are consequtive, we can store just the starting id and
a bitmap representing which subitems are in use - this reduces disk
space usage for branches allowing more of them fit into a single disk
read, further improving disk read and caching performance - disk usage
at block 18M is down from 84 to 78gb!
* the in-memory footprint of VertexRef reduced allowing more instances
to fit into caches and less memory to be used overall.
Because of the increased locality of reference, it turns out that we no
longer need to iterate over the entire database to efficiently generate
the hash key database because the normal computation is now faster -
this significantly benefits "live" chain processing as well where each
dirtied key must be accompanied by a read of all branch subitems next to
it - most of the performance benefit in this branch comes from this
locality-of-reference improvement.
On a sample resync, there's already ~20% improvement with later blocks
seeing increasing benefit (because the trie is deeper in later blocks
leading to more benefit from branch read perf improvements)
```
blocks: 18729664, baseline: 190h43m49s, contender: 153h59m0s
Time (total): -36h44m48s, -19.27%
```
Note: clients need to be resynced as the PR changes the on-disk format
R.I.P. little bloom filter - your life in the repo was short but
valuable
2024-12-04 10:42:04 +00:00
|
|
|
if (key[0].isValid or key[1].isValid):
|
2024-06-10 12:04:22 +00:00
|
|
|
return ok(key)
|
2023-06-20 13:26:25 +00:00
|
|
|
|
|
|
|
err(GetKeyNotFound)
|
|
|
|
|
2024-06-04 15:05:13 +00:00
|
|
|
proc getTuvFn(db: RdbBackendRef): GetTuvFn =
|
2023-06-20 13:26:25 +00:00
|
|
|
result =
|
2024-06-04 15:05:13 +00:00
|
|
|
proc(): Result[VertexID,AristoError]=
|
2023-06-20 13:26:25 +00:00
|
|
|
|
2024-04-22 19:02:22 +00:00
|
|
|
# Fetch serialised data record.
|
2024-06-10 12:04:22 +00:00
|
|
|
let data = db.rdb.getAdm(AdmTabIdTuv).valueOr:
|
2024-04-16 20:39:11 +00:00
|
|
|
when extraTraceMessages:
|
2024-06-04 15:05:13 +00:00
|
|
|
trace logTxt "getTuvFn: failed", error=error[0], info=error[1]
|
2024-04-16 20:39:11 +00:00
|
|
|
return err(error[0])
|
2023-06-20 13:26:25 +00:00
|
|
|
|
2024-04-16 20:39:11 +00:00
|
|
|
# Decode data record
|
|
|
|
if data.len == 0:
|
2024-06-04 15:05:13 +00:00
|
|
|
return ok VertexID(0)
|
2023-06-22 11:13:24 +00:00
|
|
|
|
2023-06-20 13:26:25 +00:00
|
|
|
# Decode data record
|
2024-06-10 12:04:22 +00:00
|
|
|
result = data.deblobify VertexID
|
2023-06-20 13:26:25 +00:00
|
|
|
|
2024-05-31 17:32:22 +00:00
|
|
|
proc getLstFn(db: RdbBackendRef): GetLstFn =
|
|
|
|
result =
|
|
|
|
proc(): Result[SavedState,AristoError]=
|
|
|
|
|
|
|
|
# Fetch serialised data record.
|
2024-06-10 12:04:22 +00:00
|
|
|
let data = db.rdb.getAdm(AdmTabIdLst).valueOr:
|
2024-05-31 17:32:22 +00:00
|
|
|
when extraTraceMessages:
|
|
|
|
trace logTxt "getLstFn: failed", error=error[0], info=error[1]
|
|
|
|
return err(error[0])
|
|
|
|
|
|
|
|
# Decode data record
|
|
|
|
data.deblobify SavedState
|
|
|
|
|
2023-06-20 13:26:25 +00:00
|
|
|
# -------------
|
|
|
|
|
|
|
|
proc putBegFn(db: RdbBackendRef): PutBegFn =
|
|
|
|
result =
|
2024-06-13 18:15:11 +00:00
|
|
|
proc(): Result[PutHdlRef,AristoError] =
|
2024-04-16 20:39:11 +00:00
|
|
|
db.rdb.begin()
|
2024-06-13 18:15:11 +00:00
|
|
|
ok db.newSession()
|
2023-06-20 13:26:25 +00:00
|
|
|
|
|
|
|
proc putVtxFn(db: RdbBackendRef): PutVtxFn =
|
|
|
|
result =
|
Store keys together with node data (#2849)
Currently, computed hash keys are stored in a separate column family
with respect to the MPT data they're generated from - this has several
disadvantages:
* A lot of space is wasted because the lookup key (`RootedVertexID`) is
repeated in both tables - this is 30% of the `AriKey` content!
* rocksdb must maintain in-memory bloom filters and LRU caches for said
keys, doubling its "minimal efficient cache size"
* An extra disk traversal must be made to check for existence of cached
hash key
* Doubles the amount of files on disk due to each column family being
its own set of files
Here, the two CFs are joined such that both key and data is stored in
`AriVtx`. This means:
* we save ~30% disk space on repeated lookup keys
* we save ~2gb of memory overhead that can be used to cache data instead
of indices
* we can skip storing hash keys for MPT leaf nodes - these are trivial
to compute and waste a lot of space - previously they had to present in
the `AriKey` CF to avoid having to look in two tables on the happy path.
* There is a small increase in write amplification because when a hash
value is updated for a branch node, we must write both key and branch
data - previously we would write only the key
* There's a small shift in CPU usage - instead of performing lookups in
the database, hashes for leaf nodes are (re)-computed on the fly
* We can return to slightly smaller on-disk SST files since there's
fewer of them, which should reduce disk traffic a bit
Internally, there are also other advantages:
* when clearing keys, we no longer have to store a zero hash in memory -
instead, we deduce staleness of the cached key from the presence of an
updated VertexRef - this saves ~1gb of mem overhead during import
* hash key cache becomes dedicated to branch keys since leaf keys are no
longer stored in memory, reducing churn
* key computation is a lot faster thanks to the skipped second disk
traversal - a key computation for mainnet can be completed in 11 hours
instead of ~2 days (!) thanks to better cache usage and less read
amplification - with additional improvements to the on-disk format, we
can probably get rid of the initial full traversal method of seeding the
key cache on first start after import
All in all, this PR reduces the size of a mainnet database from 160gb to
110gb and the peak memory footprint during import by ~1-2gb.
2024-11-20 08:56:27 +00:00
|
|
|
proc(hdl: PutHdlRef; rvid: RootedVertexID; vtx: VertexRef, key: HashKey) =
|
2023-06-20 13:26:25 +00:00
|
|
|
let hdl = hdl.getSession db
|
2023-06-30 22:22:33 +00:00
|
|
|
if hdl.error.isNil:
|
Store keys together with node data (#2849)
Currently, computed hash keys are stored in a separate column family
with respect to the MPT data they're generated from - this has several
disadvantages:
* A lot of space is wasted because the lookup key (`RootedVertexID`) is
repeated in both tables - this is 30% of the `AriKey` content!
* rocksdb must maintain in-memory bloom filters and LRU caches for said
keys, doubling its "minimal efficient cache size"
* An extra disk traversal must be made to check for existence of cached
hash key
* Doubles the amount of files on disk due to each column family being
its own set of files
Here, the two CFs are joined such that both key and data is stored in
`AriVtx`. This means:
* we save ~30% disk space on repeated lookup keys
* we save ~2gb of memory overhead that can be used to cache data instead
of indices
* we can skip storing hash keys for MPT leaf nodes - these are trivial
to compute and waste a lot of space - previously they had to present in
the `AriKey` CF to avoid having to look in two tables on the happy path.
* There is a small increase in write amplification because when a hash
value is updated for a branch node, we must write both key and branch
data - previously we would write only the key
* There's a small shift in CPU usage - instead of performing lookups in
the database, hashes for leaf nodes are (re)-computed on the fly
* We can return to slightly smaller on-disk SST files since there's
fewer of them, which should reduce disk traffic a bit
Internally, there are also other advantages:
* when clearing keys, we no longer have to store a zero hash in memory -
instead, we deduce staleness of the cached key from the presence of an
updated VertexRef - this saves ~1gb of mem overhead during import
* hash key cache becomes dedicated to branch keys since leaf keys are no
longer stored in memory, reducing churn
* key computation is a lot faster thanks to the skipped second disk
traversal - a key computation for mainnet can be completed in 11 hours
instead of ~2 days (!) thanks to better cache usage and less read
amplification - with additional improvements to the on-disk format, we
can probably get rid of the initial full traversal method of seeding the
key cache on first start after import
All in all, this PR reduces the size of a mainnet database from 160gb to
110gb and the peak memory footprint during import by ~1-2gb.
2024-11-20 08:56:27 +00:00
|
|
|
db.rdb.putVtx(rvid, vtx, key).isOkOr:
|
2024-04-16 20:39:11 +00:00
|
|
|
hdl.error = TypedPutHdlErrRef(
|
|
|
|
pfx: VtxPfx,
|
2024-06-10 12:04:22 +00:00
|
|
|
vid: error[0],
|
2024-04-16 20:39:11 +00:00
|
|
|
code: error[1],
|
|
|
|
info: error[2])
|
2023-06-20 13:26:25 +00:00
|
|
|
|
2024-06-04 15:05:13 +00:00
|
|
|
proc putTuvFn(db: RdbBackendRef): PutTuvFn =
|
2023-06-20 13:26:25 +00:00
|
|
|
result =
|
2024-06-04 15:05:13 +00:00
|
|
|
proc(hdl: PutHdlRef; vs: VertexID) =
|
2023-06-20 13:26:25 +00:00
|
|
|
let hdl = hdl.getSession db
|
2023-06-30 22:22:33 +00:00
|
|
|
if hdl.error.isNil:
|
2024-06-04 15:05:13 +00:00
|
|
|
if vs.isValid:
|
2024-07-02 18:25:06 +00:00
|
|
|
db.rdb.putAdm(AdmTabIdTuv, vs.blobify.data()).isOkOr:
|
2024-06-04 15:05:13 +00:00
|
|
|
hdl.error = TypedPutHdlErrRef(
|
|
|
|
pfx: AdmPfx,
|
|
|
|
aid: AdmTabIdTuv,
|
|
|
|
code: error[1],
|
|
|
|
info: error[2])
|
2024-06-10 12:04:22 +00:00
|
|
|
return
|
|
|
|
|
2023-08-22 18:44:54 +00:00
|
|
|
|
2024-05-31 17:32:22 +00:00
|
|
|
proc putLstFn(db: RdbBackendRef): PutLstFn =
|
|
|
|
result =
|
|
|
|
proc(hdl: PutHdlRef; lst: SavedState) =
|
|
|
|
let hdl = hdl.getSession db
|
|
|
|
if hdl.error.isNil:
|
Pre-allocate vids for branches (#2882)
Each branch node may have up to 16 sub-items - currently, these are
given VertexID based when they are first needed leading to a
mostly-random order of vertexid for each subitem.
Here, we pre-allocate all 16 vertex ids such that when a branch subitem
is filled, it already has a vertexid waiting for it. This brings several
important benefits:
* subitems are sorted and "close" in their id sequencing - this means
that when rocksdb stores them, they are likely to end up in the same
data block thus improving read efficiency
* because the ids are consequtive, we can store just the starting id and
a bitmap representing which subitems are in use - this reduces disk
space usage for branches allowing more of them fit into a single disk
read, further improving disk read and caching performance - disk usage
at block 18M is down from 84 to 78gb!
* the in-memory footprint of VertexRef reduced allowing more instances
to fit into caches and less memory to be used overall.
Because of the increased locality of reference, it turns out that we no
longer need to iterate over the entire database to efficiently generate
the hash key database because the normal computation is now faster -
this significantly benefits "live" chain processing as well where each
dirtied key must be accompanied by a read of all branch subitems next to
it - most of the performance benefit in this branch comes from this
locality-of-reference improvement.
On a sample resync, there's already ~20% improvement with later blocks
seeing increasing benefit (because the trie is deeper in later blocks
leading to more benefit from branch read perf improvements)
```
blocks: 18729664, baseline: 190h43m49s, contender: 153h59m0s
Time (total): -36h44m48s, -19.27%
```
Note: clients need to be resynced as the PR changes the on-disk format
R.I.P. little bloom filter - your life in the repo was short but
valuable
2024-12-04 10:42:04 +00:00
|
|
|
let data = lst.blobify
|
2024-06-10 12:04:22 +00:00
|
|
|
db.rdb.putAdm(AdmTabIdLst, data).isOkOr:
|
2024-05-31 17:32:22 +00:00
|
|
|
hdl.error = TypedPutHdlErrRef(
|
|
|
|
pfx: AdmPfx,
|
|
|
|
aid: AdmTabIdLst,
|
|
|
|
code: error[1],
|
|
|
|
info: error[2])
|
|
|
|
|
2023-06-20 13:26:25 +00:00
|
|
|
proc putEndFn(db: RdbBackendRef): PutEndFn =
|
|
|
|
result =
|
2023-09-12 18:45:12 +00:00
|
|
|
proc(hdl: PutHdlRef): Result[void,AristoError] =
|
2023-06-20 13:26:25 +00:00
|
|
|
let hdl = hdl.endSession db
|
2023-06-30 22:22:33 +00:00
|
|
|
if not hdl.error.isNil:
|
2024-04-16 20:39:11 +00:00
|
|
|
when extraTraceMessages:
|
|
|
|
case hdl.error.pfx:
|
|
|
|
of VtxPfx, KeyPfx: trace logTxt "putEndFn: vtx/key failed",
|
2023-08-18 19:46:55 +00:00
|
|
|
pfx=hdl.error.pfx, vid=hdl.error.vid, error=hdl.error.code
|
2024-04-16 20:39:11 +00:00
|
|
|
of AdmPfx: trace logTxt "putEndFn: admin failed",
|
|
|
|
pfx=AdmPfx, aid=hdl.error.aid.uint64, error=hdl.error.code
|
|
|
|
of Oops: trace logTxt "putEndFn: oops",
|
2024-06-13 18:15:11 +00:00
|
|
|
pfx=hdl.error.pfx, error=hdl.error.code
|
2024-06-10 12:04:22 +00:00
|
|
|
db.rdb.rollback()
|
2023-09-12 18:45:12 +00:00
|
|
|
return err(hdl.error.code)
|
2024-04-16 20:39:11 +00:00
|
|
|
|
|
|
|
# Commit session
|
|
|
|
db.rdb.commit().isOkOr:
|
2023-06-20 13:26:25 +00:00
|
|
|
when extraTraceMessages:
|
2024-04-16 20:39:11 +00:00
|
|
|
trace logTxt "putEndFn: failed", error=($error[0]), info=error[1]
|
|
|
|
return err(error[0])
|
2023-09-12 18:45:12 +00:00
|
|
|
ok()
|
2023-06-20 13:26:25 +00:00
|
|
|
|
|
|
|
proc closeFn(db: RdbBackendRef): CloseFn =
|
|
|
|
result =
|
2024-06-14 11:19:48 +00:00
|
|
|
proc(eradicate: bool) =
|
|
|
|
db.rdb.destroy(eradicate)
|
2023-06-20 13:26:25 +00:00
|
|
|
|
2024-06-13 18:15:11 +00:00
|
|
|
# ------------------------------------------------------------------------------
|
|
|
|
# Private functions: hosting interface changes
|
|
|
|
# ------------------------------------------------------------------------------
|
|
|
|
|
|
|
|
proc putBegHostingFn(db: RdbBackendRef): PutBegFn =
|
|
|
|
result =
|
|
|
|
proc(): Result[PutHdlRef,AristoError] =
|
|
|
|
db.rdb.begin()
|
|
|
|
if db.rdb.trgWriteEvent(db.rdb.session):
|
|
|
|
ok db.newSession()
|
|
|
|
else:
|
|
|
|
when extraTraceMessages:
|
|
|
|
trace logTxt "putBegFn: guest trigger aborted session"
|
|
|
|
db.rdb.rollback()
|
|
|
|
err(RdbGuestInstanceAborted)
|
|
|
|
|
2023-06-20 13:26:25 +00:00
|
|
|
# ------------------------------------------------------------------------------
|
|
|
|
# Public functions
|
|
|
|
# ------------------------------------------------------------------------------
|
|
|
|
|
2024-06-05 15:08:29 +00:00
|
|
|
proc rocksDbBackend*(
|
|
|
|
path: string;
|
2024-09-05 09:18:32 +00:00
|
|
|
opts: DbOptions;
|
2024-06-19 08:55:57 +00:00
|
|
|
dbOpts: DbOptionsRef;
|
|
|
|
cfOpts: ColFamilyOptionsRef;
|
|
|
|
guestCFs: openArray[ColFamilyDescriptor];
|
|
|
|
): Result[(BackendRef, seq[ColFamilyReadWrite]),AristoError] =
|
2023-09-05 13:57:20 +00:00
|
|
|
let db = RdbBackendRef(
|
2024-03-14 22:17:43 +00:00
|
|
|
beKind: BackendRocksDB)
|
2023-09-05 13:57:20 +00:00
|
|
|
|
|
|
|
# Initialise RocksDB
|
2024-06-19 08:55:57 +00:00
|
|
|
let oCfs = block:
|
2024-09-05 09:18:32 +00:00
|
|
|
let rc = db.rdb.init(path, opts, dbOpts, cfOpts, guestCFs)
|
2023-09-05 13:57:20 +00:00
|
|
|
if rc.isErr:
|
|
|
|
when extraTraceMessages:
|
|
|
|
trace logTxt "constructor failed",
|
|
|
|
error=rc.error[0], info=rc.error[1]
|
eth: bump (#2308)
* eth: bump
Speed up basic operations like hashing and creating RLP:s - up to 25%
improvement in certain block ranges!
```
876729c.csv /data/nimbus_stats/stats-20240605_2204-ed4f6221.csv
stats-20240605_2000-c876729c.csv vs stats-20240605_2204-ed4f6221.csv
bps_x bps_y tps_x tps_y bpsd tpsd timed
block_number
(500001, 888889] 1,017.72 996.07 1,784.96 1742.438676 -2.72% -2.72% 3.31%
(888889, 1277778] 528.00 536.30 2,159.79 2198.781046 1.69% 1.69% -1.44%
(1277778, 1666667] 324.29 317.78 2,064.48 2008.106377 -2.82% -2.82% 3.33%
(1666667, 2055556] 253.87 258.74 1,840.94 1872.935273 1.67% 1.67% -1.39%
(2055556, 2444445] 175.79 178.66 1,340.61 1363.248939 0.93% 0.93% -0.74%
(2444445, 2833334] 137.27 159.74 958.75 1113.323757 14.24% 14.24% -10.69%
(2833334, 3222223] 170.48 228.63 1,272.70 1704.047195 34.41% 34.41% -25.17%
(3222223, 3611112] 127.49 125.48 1,572.39 1548.835791 -1.19% -1.19% 1.47%
(3611112, 4000001] 37.25 40.42 1,100.65 1184.740493 9.58% 9.58% -7.04%
blocks: 3501696, baseline: 11h59m40s, contender: 11h21m38s
bpsd (mean): 6.18%
tpsd (mean): 6.18%
Time (sum): -38m1s, -4.26%
bpsd = blocks per sec diff (+), tpsd = txs per sec diff, timed = time to process diff (-)
+ = more is better, - = less is better
```
* ignore gitignore
2024-06-06 23:39:09 +00:00
|
|
|
return err(rc.error[0])
|
2024-06-19 08:55:57 +00:00
|
|
|
rc.value()
|
2023-06-20 13:26:25 +00:00
|
|
|
|
|
|
|
db.getVtxFn = getVtxFn db
|
|
|
|
db.getKeyFn = getKeyFn db
|
2024-06-04 15:05:13 +00:00
|
|
|
db.getTuvFn = getTuvFn db
|
2024-05-31 17:32:22 +00:00
|
|
|
db.getLstFn = getLstFn db
|
2023-06-20 13:26:25 +00:00
|
|
|
|
|
|
|
db.putBegFn = putBegFn db
|
|
|
|
db.putVtxFn = putVtxFn db
|
2024-06-04 15:05:13 +00:00
|
|
|
db.putTuvFn = putTuvFn db
|
2024-05-31 17:32:22 +00:00
|
|
|
db.putLstFn = putLstFn db
|
2023-06-20 13:26:25 +00:00
|
|
|
db.putEndFn = putEndFn db
|
|
|
|
|
|
|
|
db.closeFn = closeFn db
|
2024-06-19 08:55:57 +00:00
|
|
|
ok((db, oCfs))
|
2024-06-13 18:15:11 +00:00
|
|
|
|
|
|
|
|
|
|
|
proc rocksDbSetEventTrigger*(
|
|
|
|
be: BackendRef;
|
|
|
|
hdl: RdbWriteEventCb;
|
|
|
|
): Result[void,AristoError] =
|
|
|
|
## Store event trigger. This also changes the backend type.
|
|
|
|
if hdl.isNil:
|
|
|
|
err(RdbBeWrTriggerNilFn)
|
|
|
|
else:
|
|
|
|
let db = RdbBackendRef(be)
|
|
|
|
db.rdb.trgWriteEvent = hdl
|
|
|
|
db.beKind = BackendRdbHosting
|
|
|
|
db.putBegFn = putBegHostingFn db
|
|
|
|
ok()
|
|
|
|
|
|
|
|
|
2024-03-14 22:17:43 +00:00
|
|
|
proc dup*(db: RdbBackendRef): RdbBackendRef =
|
|
|
|
## Duplicate descriptor shell as needed for API debugging
|
|
|
|
new result
|
|
|
|
init_common.init(result[], db[])
|
|
|
|
result.rdb = db.rdb
|
|
|
|
|
2023-06-20 13:26:25 +00:00
|
|
|
# ------------------------------------------------------------------------------
|
|
|
|
# Public iterators (needs direct backend access)
|
|
|
|
# ------------------------------------------------------------------------------
|
|
|
|
|
|
|
|
iterator walkVtx*(
|
|
|
|
be: RdbBackendRef;
|
Speed up initial MPT root computation after import (#2788)
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.
2024-10-27 11:08:37 +00:00
|
|
|
kinds = {Branch, Leaf};
|
2024-07-04 13:46:52 +00:00
|
|
|
): tuple[evid: RootedVertexID, vtx: VertexRef] =
|
2023-06-20 13:26:25 +00:00
|
|
|
## Variant of `walk()` iteration over the vertex sub-table.
|
Speed up initial MPT root computation after import (#2788)
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.
2024-10-27 11:08:37 +00:00
|
|
|
for (rvid, vtx) in be.rdb.walkVtx(kinds):
|
2024-09-02 14:03:10 +00:00
|
|
|
yield (rvid, vtx)
|
2023-06-20 13:26:25 +00:00
|
|
|
|
2023-08-22 18:44:54 +00:00
|
|
|
iterator walkKey*(
|
2023-06-20 13:26:25 +00:00
|
|
|
be: RdbBackendRef;
|
2024-07-04 13:46:52 +00:00
|
|
|
): tuple[rvid: RootedVertexID, key: HashKey] =
|
2023-06-20 13:26:25 +00:00
|
|
|
## Variant of `walk()` iteration over the Markle hash sub-table.
|
2024-07-04 13:46:52 +00:00
|
|
|
for (rvid, data) in be.rdb.walkKey:
|
Store keys together with node data (#2849)
Currently, computed hash keys are stored in a separate column family
with respect to the MPT data they're generated from - this has several
disadvantages:
* A lot of space is wasted because the lookup key (`RootedVertexID`) is
repeated in both tables - this is 30% of the `AriKey` content!
* rocksdb must maintain in-memory bloom filters and LRU caches for said
keys, doubling its "minimal efficient cache size"
* An extra disk traversal must be made to check for existence of cached
hash key
* Doubles the amount of files on disk due to each column family being
its own set of files
Here, the two CFs are joined such that both key and data is stored in
`AriVtx`. This means:
* we save ~30% disk space on repeated lookup keys
* we save ~2gb of memory overhead that can be used to cache data instead
of indices
* we can skip storing hash keys for MPT leaf nodes - these are trivial
to compute and waste a lot of space - previously they had to present in
the `AriKey` CF to avoid having to look in two tables on the happy path.
* There is a small increase in write amplification because when a hash
value is updated for a branch node, we must write both key and branch
data - previously we would write only the key
* There's a small shift in CPU usage - instead of performing lookups in
the database, hashes for leaf nodes are (re)-computed on the fly
* We can return to slightly smaller on-disk SST files since there's
fewer of them, which should reduce disk traffic a bit
Internally, there are also other advantages:
* when clearing keys, we no longer have to store a zero hash in memory -
instead, we deduce staleness of the cached key from the presence of an
updated VertexRef - this saves ~1gb of mem overhead during import
* hash key cache becomes dedicated to branch keys since leaf keys are no
longer stored in memory, reducing churn
* key computation is a lot faster thanks to the skipped second disk
traversal - a key computation for mainnet can be completed in 11 hours
instead of ~2 days (!) thanks to better cache usage and less read
amplification - with additional improvements to the on-disk format, we
can probably get rid of the initial full traversal method of seeding the
key cache on first start after import
All in all, this PR reduces the size of a mainnet database from 160gb to
110gb and the peak memory footprint during import by ~1-2gb.
2024-11-20 08:56:27 +00:00
|
|
|
yield (rvid, data)
|
2023-06-20 13:26:25 +00:00
|
|
|
|
|
|
|
# ------------------------------------------------------------------------------
|
|
|
|
# End
|
|
|
|
# ------------------------------------------------------------------------------
|