Jacek Sieka 7b88bb3b30
Add branch cache (#2923)
Now that branches are small, we can add a branch cache that fits more
verticies in memory by only storing the branch portion (16 bytes) of the
VertexRef (136 bytes).

Where the original vertex cache hovers around a hit rate of ~60:ish,
this branch cache reaches >90% hit rate instead around block 20M which
gives a nice boost to processing.

A downside of this approach is that a new VertexRef must be allocated
for every cache hit instead of reusing an existing instance - this
causes some GC overhead that needs to be addressed.

Nice 15% improvement nonetheless, can't complain!

```
blocks: 19630784, baseline: 161h18m38s, contender: 136h23m23s
Time (total): -24h55m14s, -15.45%
```
2024-12-11 11:53:26 +01:00

234 lines
7.4 KiB
Nim

# nimbus-eth1
# Copyright (c) 2023-2024 Status Research & Development GmbH
# 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.
## Rocks DB fetch data record
## ==========================
{.push raises: [].}
import
eth/common,
rocksdb,
results,
../../[aristo_blobify, aristo_desc],
../init_common,
./rdb_desc,
std/concurrency/atomics
const extraTraceMessages = false ## Enable additional logging noise
when extraTraceMessages:
import chronicles
logScope:
topics = "aristo-rocksdb"
when defined(metrics):
import metrics
type
RdbVtxLruCounter = ref object of Counter
RdbKeyLruCounter = ref object of Counter
RdbBranchLruCounter = ref object of Counter
var
rdbVtxLruStatsMetric {.used.} = RdbVtxLruCounter.newCollector(
"aristo_rdb_vtx_lru_total",
"Vertex LRU lookup (hit/miss, world/account, branch/leaf)",
labels = ["state", "vtype", "hit"],
)
rdbKeyLruStatsMetric {.used.} = RdbKeyLruCounter.newCollector(
"aristo_rdb_key_lru_total", "HashKey LRU lookup", labels = ["state", "hit"]
)
rdbBranchLruStatsMetric {.used.} = RdbBranchLruCounter.newCollector(
"aristo_rdb_branch_lru_total", "Branch LRU lookup", labels = ["state", "hit"]
)
method collect*(collector: RdbVtxLruCounter, output: MetricHandler) =
let timestamp = collector.now()
# We don't care about synchronization between each type of metric or between
# the metrics thread and others since small differences like this don't matter
for state in RdbStateType:
for vtype in VertexType:
for hit in [false, true]:
output(
name = "aristo_rdb_vtx_lru_total",
value = float64(rdbVtxLruStats[state][vtype].get(hit)),
labels = ["state", "vtype", "hit"],
labelValues = [$state, $vtype, $ord(hit)],
timestamp = timestamp,
)
method collect*(collector: RdbKeyLruCounter, output: MetricHandler) =
let timestamp = collector.now()
for state in RdbStateType:
for hit in [false, true]:
output(
name = "aristo_rdb_key_lru_total",
value = float64(rdbKeyLruStats[state].get(hit)),
labels = ["state", "hit"],
labelValues = [$state, $ord(hit)],
timestamp = timestamp,
)
method collect*(collector: RdbBranchLruCounter, output: MetricHandler) =
let timestamp = collector.now()
for state in RdbStateType:
for hit in [false, true]:
output(
name = "aristo_rdb_branch_lru_total",
value = float64(rdbBranchLruStats[state].get(hit)),
labels = ["state", "hit"],
labelValues = [$state, $ord(hit)],
timestamp = timestamp,
)
# ------------------------------------------------------------------------------
# Public functions
# ------------------------------------------------------------------------------
proc getAdm*(rdb: RdbInst, xid: AdminTabID): Result[seq[byte], (AristoError, string)] =
var res: seq[byte]
let onData = proc(data: openArray[byte]) =
res = @data
let gotData = rdb.admCol.get(xid.toOpenArray, onData).valueOr:
const errSym = RdbBeDriverGetAdmError
when extraTraceMessages:
trace logTxt "getAdm", xid, error = errSym, info = error
return err((errSym, error))
# Correct result if needed
if not gotData:
res = EmptyBlob
ok move(res)
proc getKey*(
rdb: var RdbInst, rvid: RootedVertexID, flags: set[GetVtxFlag]
): Result[(HashKey, VertexRef), (AristoError, string)] =
block:
# Try LRU cache first
let rc =
if GetVtxFlag.PeekCache in flags:
rdb.rdKeyLru.peek(rvid.vid)
else:
rdb.rdKeyLru.get(rvid.vid)
if rc.isOk:
rdbKeyLruStats[rvid.to(RdbStateType)].inc(true)
return ok((rc.value, nil))
rdbKeyLruStats[rvid.to(RdbStateType)].inc(false)
block:
# We don't store keys for leaves, no need to hit the database
let rc = rdb.rdVtxLru.peek(rvid.vid)
if rc.isOk():
if rc.value().vType == Leaf:
return ok((VOID_HASH_KEY, rc.value()))
# Otherwise fetch from backend database
# A threadvar is used to avoid allocating an environment for onData
var res {.threadvar.}: Opt[HashKey]
var vtx {.threadvar.}: Result[VertexRef, AristoError]
let onData = proc(data: openArray[byte]) =
res = data.deblobify(HashKey)
if res.isSome():
reset(vtx)
else:
vtx = data.deblobify(VertexRef)
let gotData = rdb.vtxCol.get(rvid.blobify().data(), onData).valueOr:
const errSym = RdbBeDriverGetKeyError
when extraTraceMessages:
trace logTxt "getKey", rvid, error = errSym, info = error
return err((errSym, error))
if not gotData:
return ok((VOID_HASH_KEY, nil))
# Update cache and return - in peek mode, avoid evicting cache items
if res.isSome() and
(GetVtxFlag.PeekCache notin flags or rdb.rdKeyLru.len < rdb.rdKeyLru.capacity):
rdb.rdKeyLru.put(rvid.vid, res.value())
if vtx.isOk() and
(GetVtxFlag.PeekCache notin flags or rdb.rdVtxLru.len < rdb.rdVtxLru.capacity):
rdb.rdVtxLru.put(rvid.vid, vtx.value())
ok (res.valueOr(VOID_HASH_KEY), vtx.valueOr(nil))
proc getVtx*(
rdb: var RdbInst, rvid: RootedVertexID, flags: set[GetVtxFlag]
): Result[VertexRef, (AristoError, string)] =
# Try LRU cache first
block:
let rc =
if GetVtxFlag.PeekCache in flags:
rdb.rdBranchLru.peek(rvid.vid)
else:
rdb.rdBranchLru.get(rvid.vid)
if rc.isOk():
rdbBranchLruStats[rvid.to(RdbStateType)].inc(true)
return ok(VertexRef(vType: Branch, startVid: rc[][0], used: rc[][1]))
block:
var rc =
if GetVtxFlag.PeekCache in flags:
rdb.rdVtxLru.peek(rvid.vid)
else:
rdb.rdVtxLru.get(rvid.vid)
if rc.isOk:
rdbVtxLruStats[rvid.to(RdbStateType)][rc.value().vType].inc(true)
return ok(move(rc.value))
# Otherwise fetch from backend database
# A threadvar is used to avoid allocating an environment for onData
var res {.threadvar.}: Result[VertexRef, AristoError]
let onData = proc(data: openArray[byte]) =
res = data.deblobify(VertexRef)
let gotData = rdb.vtxCol.get(rvid.blobify().data(), onData).valueOr:
const errSym = RdbBeDriverGetVtxError
when extraTraceMessages:
trace logTxt "getVtx", vid, error = errSym, info = error
return err((errSym, error))
if not gotData:
# As a hack, we count missing data as leaf nodes
rdbVtxLruStats[rvid.to(RdbStateType)][VertexType.Leaf].inc(false)
return ok(VertexRef(nil))
if res.isErr():
return err((res.error(), "Parsing failed")) # Parsing failed
if res.value.vType == Branch and res.value.pfx.len == 0:
rdbBranchLruStats[rvid.to(RdbStateType)].inc(false)
else:
rdbVtxLruStats[rvid.to(RdbStateType)][res.value().vType].inc(false)
# Update cache and return - in peek mode, avoid evicting cache items
if GetVtxFlag.PeekCache notin flags:
if res.value.vType == Branch and res.value.pfx.len == 0:
rdb.rdBranchLru.put(rvid.vid, (res.value().startVid, res.value.used))
else:
rdb.rdVtxLru.put(rvid.vid, res.value())
ok res.value()
# ------------------------------------------------------------------------------
# End
# ------------------------------------------------------------------------------