avoid some trivial memory allocations (#2587)

* 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)

nimbus/db/aristo/aristo_blobify.nim

@@ -203,12 +203,12 @@ proc blobifyTo*(vtx: VertexRef; data: var Blob): Result[void,AristoError] =
         if vtx.ePfx.len > 0:
           vtx.ePfx.toHexPrefix(isleaf = false)
         else:
-          @[]
+          default(HexPrefixBuf)
       psLen = pSegm.len.byte
     if 33 < psLen:
       return err(BlobifyExtPathOverflow)
 
-    data &= pSegm
+    data &= pSegm.data()
     data &= lens.toBytesBE
     data &= [0x80u8 or psLen]
 
@@ -219,16 +219,16 @@ proc blobifyTo*(vtx: VertexRef; data: var Blob): Result[void,AristoError] =
     if psLen == 0 or 33 < psLen:
       return err(BlobifyLeafPathOverflow)
     vtx.lData.blobifyTo(data)
-    data &= pSegm
+    data &= pSegm.data()
     data &= [0xC0u8 or psLen]
 
   ok()
 
-proc blobify*(vtx: VertexRef): Result[Blob, AristoError] =
+proc blobify*(vtx: VertexRef): Blob =
   ## Variant of `blobify()`
-  var data: Blob
-  ? vtx.blobifyTo data
-  ok(move(data))
+  result = newSeqOfCap[byte](128)
+  if vtx.blobifyTo(result).isErr:
+    result.setLen(0) # blobify only fails on invalid verticies
 
 proc blobifyTo*(lSst: SavedState; data: var Blob): Result[void,AristoError] =
   ## Serialise a last saved state record
@@ -246,45 +246,48 @@ proc blobify*(lSst: SavedState): Result[Blob,AristoError] =
 # -------------
 proc deblobify(
     data: openArray[byte];
-    T: type LeafPayload;
-      ): Result[LeafPayload,AristoError] =
+    pyl: var LeafPayload;
+      ): Result[void,AristoError] =
   if data.len == 0:
-    return ok LeafPayload(pType: RawData)
+    pyl = LeafPayload(pType: RawData)
+    return ok()
 
   let mask = data[^1]
   if (mask and 0x10) > 0: # unstructured payload
-    return ok LeafPayload(pType: RawData, rawBlob: data[0 .. ^2])
+    pyl = LeafPayload(pType: RawData, rawBlob: data[0 .. ^2])
+    return ok()
 
   if (mask and 0x20) > 0: # Slot storage data
-    return ok LeafPayload(
+    pyl = LeafPayload(
       pType: StoData,
       stoData: ?deblobify(data.toOpenArray(0, data.len - 2), UInt256))
+    return ok()
 
+  pyl = LeafPayload(pType: AccountData)
   var
-    pAcc = LeafPayload(pType: AccountData)
     start = 0
     lens = uint16.fromBytesBE(data.toOpenArray(data.len - 3, data.len - 2))
 
   if (mask and 0x01) > 0:
     let len = lens and 0b111
-    pAcc.account.nonce = ? load64(data, start, int(len + 1))
+    pyl.account.nonce = ? load64(data, start, int(len + 1))
 
   if (mask and 0x02) > 0:
     let len = (lens shr 3) and 0b11111
-    pAcc.account.balance = ? load256(data, start, int(len + 1))
+    pyl.account.balance = ? load256(data, start, int(len + 1))
 
   if (mask and 0x04) > 0:
     let len = (lens shr 8) and 0b111
-    pAcc.stoID = (true, VertexID(? load64(data, start, int(len + 1))))
+    pyl.stoID = (true, VertexID(? load64(data, start, int(len + 1))))
 
   if (mask and 0x08) > 0:
     if data.len() < start + 32:
       return err(DeblobCodeLenUnsupported)
-    discard pAcc.account.codeHash.data.copyFrom(data.toOpenArray(start, start + 31))
+    discard pyl.account.codeHash.data.copyFrom(data.toOpenArray(start, start + 31))
   else:
-    pAcc.account.codeHash = EMPTY_CODE_HASH
+    pyl.account.codeHash = EMPTY_CODE_HASH
 
-  ok(pAcc)
+  ok()
 
 proc deblobify*(
     record: openArray[byte];
@@ -336,11 +339,12 @@ proc deblobify*(
       NibblesBuf.fromHexPrefix record.toOpenArray(pLen, rLen-1)
     if not isLeaf:
       return err(DeblobLeafGotExtPrefix)
-    let pyl = ? record.toOpenArray(0, pLen - 1).deblobify(LeafPayload)
-    VertexRef(
+    let vtx = VertexRef(
       vType: Leaf,
-      lPfx:  pathSegment,
-      lData: pyl)
+      lPfx:  pathSegment)
+
+    ? record.toOpenArray(0, pLen - 1).deblobify(vtx.lData)
+    vtx
 
   else:
     return err(DeblobUnknown)

nimbus/db/aristo/aristo_compute.nim

@@ -68,7 +68,7 @@ proc computeKeyImpl(
   case vtx.vType:
   of Leaf:
     writer.startList(2)
-    writer.append(vtx.lPfx.toHexPrefix(isLeaf = true))
+    writer.append(vtx.lPfx.toHexPrefix(isLeaf = true).data())
 
     case vtx.lData.pType
     of AccountData:
@@ -111,7 +111,7 @@ proc computeKeyImpl(
       writeBranch(bwriter)
 
       writer.startList(2)
-      writer.append(vtx.ePfx.toHexPrefix(isleaf = false))
+      writer.append(vtx.ePfx.toHexPrefix(isleaf = false).data())
       writer.append(bwriter.finish().digestTo(HashKey))
     else:
       writeBranch(writer)

nimbus/db/aristo/aristo_desc/desc_nibbles.nim

@@ -8,15 +8,20 @@
 # at your option. This file may not be copied, modified, or distributed
 # except according to those terms.
 
-import stew/arrayops
+import stew/[arraybuf, arrayops]
 
-type NibblesBuf* = object
-  ## Allocation-free type for storing up to 64 4-bit nibbles, as seen in the
-  ## Ethereum MPT
-  bytes: array[32, byte]
-  ibegin, iend: int8
-    # Where valid nibbles can be found - we use indices here to avoid copies
-    # wen slicing - iend not inclusive
+export arraybuf
+
+type
+  NibblesBuf* = object
+    ## Allocation-free type for storing up to 64 4-bit nibbles, as seen in the
+    ## Ethereum MPT
+    bytes: array[32, byte]
+    ibegin, iend: int8
+      # Where valid nibbles can be found - we use indices here to avoid copies
+      # wen slicing - iend not inclusive
+
+  HexPrefixBuf* = ArrayBuf[33, byte]
 
 func high*(T: type NibblesBuf): int =
   63
@@ -61,7 +66,7 @@ func `$`*(r: NibblesBuf): string =
     const chars = "0123456789abcdef"
     result.add chars[r[i]]
 
-func slice*(r: NibblesBuf, ibegin: int, iend = -1): NibblesBuf =
+func slice*(r: NibblesBuf, ibegin: int, iend = -1): NibblesBuf {.noinit.} =
   result.bytes = r.bytes
   result.ibegin = r.ibegin + ibegin.int8
   let e =
@@ -75,7 +80,7 @@ func slice*(r: NibblesBuf, ibegin: int, iend = -1): NibblesBuf =
 template writeFirstByte(nibbleCountExpr) {.dirty.} =
   let nibbleCount = nibbleCountExpr
   var oddnessFlag = (nibbleCount and 1) != 0
-  newSeq(result, (nibbleCount div 2) + 1)
+  result.setLen((nibbleCount div 2) + 1)
   result[0] = byte((int(isLeaf) * 2 + int(oddnessFlag)) shl 4)
   var writeHead = 0
 
@@ -89,11 +94,11 @@ template writeNibbles(r) {.dirty.} =
       result[writeHead] = nextNibble shl 4
     oddnessFlag = not oddnessFlag
 
-func toHexPrefix*(r: NibblesBuf, isLeaf = false): seq[byte] =
+func toHexPrefix*(r: NibblesBuf, isLeaf = false): HexPrefixBuf =
   writeFirstByte(r.len)
   writeNibbles(r)
 
-func toHexPrefix*(r1, r2: NibblesBuf, isLeaf = false): seq[byte] =
+func toHexPrefix*(r1, r2: NibblesBuf, isLeaf = false): HexPrefixBuf =
   writeFirstByte(r1.len + r2.len)
   writeNibbles(r1)
   writeNibbles(r2)
@@ -131,7 +136,7 @@ func fromHexPrefix*(
   else:
     result.isLeaf = false
 
-func `&`*(a, b: NibblesBuf): NibblesBuf =
+func `&`*(a, b: NibblesBuf): NibblesBuf {.noinit.} =
   for i in 0 ..< a.len:
     result[i] = a[i]
 

nimbus/db/aristo/aristo_init/memory_db.nim

@@ -133,14 +133,7 @@ proc putVtxFn(db: MemBackendRef): PutVtxFn =
       let hdl = hdl.getSession db
       if hdl.error.isNil:
         if vtx.isValid:
-          let rc = vtx.blobify()
-          if rc.isErr:
-            hdl.error = TypedPutHdlErrRef(
-              pfx:  VtxPfx,
-              vid:  rvid.vid,
-              code: rc.error)
-            return
-          hdl.sTab[rvid] = rc.value
+          hdl.sTab[rvid] = vtx.blobify()
         else:
           hdl.sTab[rvid] = EmptyBlob
 

nimbus/db/aristo/aristo_init/rocks_db.nim

@@ -312,10 +312,8 @@ iterator walkVtx*(
     be: RdbBackendRef;
       ): tuple[evid: RootedVertexID, vtx: VertexRef] =
   ## Variant of `walk()` iteration over the vertex sub-table.
-  for (rvid, data) in be.rdb.walkVtx:
-    let rc = data.deblobify VertexRef
-    if rc.isOk:
-      yield (rvid, rc.value)
+  for (rvid, vtx) in be.rdb.walkVtx:
+    yield (rvid, vtx)
 
 iterator walkKey*(
     be: RdbBackendRef;

nimbus/db/aristo/aristo_init/rocks_db/rdb_get.nim

@@ -63,10 +63,10 @@ proc getKey*(
     return ok(move(rc.value))
 
   # Otherwise fetch from backend database
-  var res: Result[HashKey,(AristoError,string)]
+  # A threadvar is used to avoid allocating an environment for onData
+  var res{.threadvar.}: Opt[HashKey]
   let onData = proc(data: openArray[byte]) =
-    res = HashKey.fromBytes(data).mapErr(proc(): auto =
-      (RdbHashKeyExpected,""))
+    res = HashKey.fromBytes(data)
 
   let gotData = rdb.keyCol.get(rvid.blobify().data(), onData).valueOr:
      const errSym = RdbBeDriverGetKeyError
@@ -76,9 +76,9 @@ proc getKey*(
 
   # Correct result if needed
   if not gotData:
-    res = ok(VOID_HASH_KEY)
+    res.ok(VOID_HASH_KEY)
   elif res.isErr():
-    return res # Parsing failed
+    return err((RdbHashKeyExpected,"")) # Parsing failed
 
   # Update cache and return
   ok rdb.rdKeyLru.lruAppend(rvid.vid, res.value(), RdKeyLruMaxSize)
@@ -93,10 +93,10 @@ proc getVtx*(
     return ok(move(rc.value))
 
   # Otherwise fetch from backend database
-  var res: Result[VertexRef,(AristoError,string)]
+  # 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).mapErr(proc(error: AristoError): auto =
-      (error,""))
+    res = data.deblobify(VertexRef)
 
   let gotData = rdb.vtxCol.get(rvid.blobify().data(), onData).valueOr:
     const errSym = RdbBeDriverGetVtxError
@@ -105,9 +105,9 @@ proc getVtx*(
     return err((errSym,error))
 
   if not gotData:
-    res = ok(VertexRef(nil))
+    res.ok(VertexRef(nil))
   elif res.isErr():
-    return res # Parsing failed
+    return err((res.error(), "Parsing failed")) # Parsing failed
 
   # Update cache and return
   ok rdb.rdVtxLru.lruAppend(rvid.vid, res.value(), RdVtxLruMaxSize)

nimbus/db/aristo/aristo_init/rocks_db/rdb_put.nim

@@ -122,12 +122,7 @@ proc putVtx*(
       ): Result[void,(VertexID,AristoError,string)] =
   let dsc = rdb.session
   if vtx.isValid:
-    let rc = vtx.blobify()
-    if rc.isErr:
-      # Caller must `rollback()` which will flush the `rdVtxLru` cache
-      return err((rvid.vid,rc.error,""))
-
-    dsc.put(rvid.blobify().data(), rc.value, rdb.vtxCol.handle()).isOkOr:
+    dsc.put(rvid.blobify().data(), vtx.blobify(), rdb.vtxCol.handle()).isOkOr:
       # Caller must `rollback()` which will flush the `rdVtxLru` cache
       const errSym = RdbBeDriverPutVtxError
       when extraTraceMessages:

nimbus/db/aristo/aristo_init/rocks_db/rdb_walk.nim

@@ -72,8 +72,8 @@ iterator walkKey*(rdb: RdbInst): tuple[rvid: RootedVertexID, data: Blob] =
         yield (rvid, val)
 
 
-iterator walkVtx*(rdb: RdbInst): tuple[rvid: RootedVertexID, data: Blob] =
-  ## Walk over key-value pairs of the hash key column of the database.
+iterator walkVtx*(rdb: RdbInst): tuple[rvid: RootedVertexID, data: VertexRef] =
+  ## Walk over key-value pairs of the vertex column of the database.
   ##
   ## Non-decodable entries are are ignored.
   ##
@@ -84,12 +84,32 @@ iterator walkVtx*(rdb: RdbInst): tuple[rvid: RootedVertexID, data: Blob] =
       break walkBody
     defer: rit.close()
 
-    for (key,val) in rit.pairs:
-      if val.len != 0:
-        let rvid = key.deblobify(RootedVertexID).valueOr:
-          continue
-
-        yield (rvid, val)
+    rit.seekToFirst()
+    var key: RootedVertexID
+    var value: VertexRef
+    while rit.isValid():
+      var valid = true
+      rit.key(
+        proc(data: openArray[byte]) =
+          key = deblobify(data, RootedVertexID).valueOr:
+            valid = false
+            default(RootedVertexID)
+      )
+      if not valid:
+        continue
+
+      rit.value(
+        proc(data: openArray[byte]) =
+          value = deblobify(data, VertexRef).valueOr:
+            valid = false
+            default(VertexRef)
+      )
+      if not valid:
+        continue
+
+      rit.next()
+      yield (key, value)
+    rit.close()
 
 # ------------------------------------------------------------------------------
 # End

nimbus/db/aristo/aristo_serialise.nim

@@ -86,7 +86,7 @@ proc to*(node: NodeRef; T: type seq[Blob]): T =
 
       var wrx = initRlpWriter()
       wrx.startList(2)
-      wrx.append node.ePfx.toHexPrefix(isleaf = false)
+      wrx.append node.ePfx.toHexPrefix(isleaf = false).data()
       wrx.append brHash
 
       result.add wrx.finish()
@@ -104,7 +104,7 @@ proc to*(node: NodeRef; T: type seq[Blob]): T =
 
     var wr = initRlpWriter()
     wr.startList(2)
-    wr.append node.lPfx.toHexPrefix(isleaf = true)
+    wr.append node.lPfx.toHexPrefix(isleaf = true).data()
     wr.append node.lData.serialise(getKey0).value
 
     result.add (wr.finish())
@@ -127,7 +127,7 @@ proc digestTo*(node: NodeRef; T: type HashKey): T =
       let brHash = wr.finish().digestTo(HashKey)
       wr = initRlpWriter()
       wr.startList(2)
-      wr.append node.ePfx.toHexPrefix(isleaf = false)
+      wr.append node.ePfx.toHexPrefix(isleaf = false).data()
       wr.append brHash
 
   of Leaf:
@@ -138,7 +138,7 @@ proc digestTo*(node: NodeRef; T: type HashKey): T =
       ok(node.key[0]) # always succeeds
 
     wr.startList(2)
-    wr.append node.lPfx.toHexPrefix(isleaf = true)
+    wr.append node.lPfx.toHexPrefix(isleaf = true).data()
     wr.append node.lData.serialise(getKey0).value
 
   wr.finish().digestTo(HashKey)

tests/test_aristo/test_blobify.nim

@@ -65,8 +65,8 @@ suite "Aristo blobify":
       )
 
     check:
-      deblobify(blobify(leafRawData)[], VertexRef)[] == leafRawData
-      deblobify(blobify(leafAccount)[], VertexRef)[] == leafAccount
-      deblobify(blobify(leafStoData)[], VertexRef)[] == leafStoData
-      deblobify(blobify(branch)[], VertexRef)[] == branch
-      deblobify(blobify(extension)[], VertexRef)[] == extension
+      deblobify(blobify(leafRawData), VertexRef)[] == leafRawData
+      deblobify(blobify(leafAccount), VertexRef)[] == leafAccount
+      deblobify(blobify(leafStoData), VertexRef)[] == leafStoData
+      deblobify(blobify(branch), VertexRef)[] == branch
+      deblobify(blobify(extension), VertexRef)[] == extension