From 9afdf134d3ff7608c69dc967f2f5724ef4c1866c Mon Sep 17 00:00:00 2001
From: Kirill Smelkov <kirr@nexedi.com>
Date: Thu, 23 Mar 2017 20:38:27 +0300
Subject: [PATCH] X Remove many allocations from inside inner loops
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
name old time/op new time/op delta
Iterate-4 17.4µs ± 0% 13.1µs ± 1% -24.65% (p=0.000 n=10+9)
name old alloc/op new alloc/op delta
Iterate-4 17.2kB ± 0% 10.6kB ± 0% -38.29% (p=0.000 n=10+10)
name old allocs/op new allocs/op delta
Iterate-4 164 ± 0% 24 ± 0% -85.37% (p=0.000 n=10+10)
---
t/neo/storage/fs1/filestorage.go | 38 +++++++++++++++------------
t/neo/storage/fs1/filestorage_test.go | 4 +++
2 files changed, 25 insertions(+), 17 deletions(-)
diff --git a/t/neo/storage/fs1/filestorage.go b/t/neo/storage/fs1/filestorage.go
index 6566d0de..44d0f511 100644
--- a/t/neo/storage/fs1/filestorage.go
+++ b/t/neo/storage/fs1/filestorage.go
@@ -75,6 +75,9 @@ type DataHeader struct {
//DataRecPos uint64 // if Data == nil -> byte position of data record containing data
// XXX include word0 ?
+
+ // underlying memory for header loading (to avoid allocations)
+ workMem [DataHeaderSize]byte
}
const (
@@ -444,7 +447,7 @@ func (dh *DataHeader) Len() int64 {
// load reads and decodes data record header
// pos: points to data header start
// no prerequisite requirements are made to previous dh state
-func (dh *DataHeader) load(r io.ReaderAt /* *os.File */, pos int64, tmpBuf *[DataHeaderSize]byte) error {
+func (dh *DataHeader) Load(r io.ReaderAt /* *os.File */, pos int64) error {
dh.Pos = pos
// XXX .Len = 0 = read error ?
@@ -452,20 +455,20 @@ func (dh *DataHeader) load(r io.ReaderAt /* *os.File */, pos int64, tmpBuf *[Dat
bug(dh, "Load() on invalid position")
}
- _, err := r.ReadAt(tmpBuf[:], pos)
+ _, err := r.ReadAt(dh.workMem[:], pos)
if err != nil {
return dh.err("read", noEOF(err))
}
// XXX also check oid.Valid() ?
- dh.Oid = zodb.Oid(binary.BigEndian.Uint64(tmpBuf[0:])) // XXX -> zodb.Oid.Decode() ?
- dh.Tid = zodb.Tid(binary.BigEndian.Uint64(tmpBuf[8:])) // XXX -> zodb.Tid.Decode() ?
+ dh.Oid = zodb.Oid(binary.BigEndian.Uint64(dh.workMem[0:])) // XXX -> zodb.Oid.Decode() ?
+ dh.Tid = zodb.Tid(binary.BigEndian.Uint64(dh.workMem[8:])) // XXX -> zodb.Tid.Decode() ?
if !dh.Tid.Valid() {
return decodeErr(dh, "invalid tid: %v", dh.Tid)
}
- dh.PrevRevPos = int64(binary.BigEndian.Uint64(tmpBuf[16:]))
- dh.TxnPos = int64(binary.BigEndian.Uint64(tmpBuf[24:]))
+ dh.PrevRevPos = int64(binary.BigEndian.Uint64(dh.workMem[16:]))
+ dh.TxnPos = int64(binary.BigEndian.Uint64(dh.workMem[24:]))
if dh.TxnPos < txnValidFrom {
return decodeErr(dh, "invalid txn position: %v", dh.TxnPos)
}
@@ -482,12 +485,12 @@ func (dh *DataHeader) load(r io.ReaderAt /* *os.File */, pos int64, tmpBuf *[Dat
}
}
- verlen := binary.BigEndian.Uint16(tmpBuf[32:])
+ verlen := binary.BigEndian.Uint16(dh.workMem[32:])
if verlen != 0 {
return decodeErr(dh, "non-zero version: #%v", verlen)
}
- dh.DataLen = int64(binary.BigEndian.Uint64(tmpBuf[34:]))
+ dh.DataLen = int64(binary.BigEndian.Uint64(dh.workMem[34:]))
if dh.DataLen < 0 {
// XXX also check DataLen < max ?
return decodeErr(dh, "invalid data len: %v", dh.DataLen)
@@ -496,12 +499,6 @@ func (dh *DataHeader) load(r io.ReaderAt /* *os.File */, pos int64, tmpBuf *[Dat
return nil
}
-// XXX do we need Load when load() is there?
-func (dh *DataHeader) Load(r io.ReaderAt, pos int64) error {
- var tmpBuf [DataHeaderSize]byte
- return dh.load(r, pos, &tmpBuf)
-}
-
// LoadPrevRev reads and decodes previous revision data record header
// prerequisite: dh .Oid .Tid .PrevRevPos are initialized:
// - TODO describe how
@@ -827,6 +824,11 @@ type dataIter struct {
Txnh *TxnHeader // header of transaction we are iterating inside
Datah DataHeader
+ // data header for data loading
+ // XXX need to use separate dh because x.LoadData() changes x state while going through backpointers.
+ // XXX here to avoid allocations
+ dhLoading DataHeader
+
sri zodb.StorageRecordInformation // ptr to this will be returned by NextData
dataBuf []byte
}
@@ -885,9 +887,11 @@ func (di *dataIter) NextData() (*zodb.StorageRecordInformation, error) {
di.sri.Oid = di.Datah.Oid
di.sri.Tid = di.Datah.Tid
- dh := di.Datah
+ // NOTE dh.LoadData() changes dh state while going through backpointers -
+ // - need to use separate dh because of this
+ di.dhLoading = di.Datah
di.sri.Data = di.dataBuf
- err = dh.LoadData(di.fsSeq, &di.sri.Data)
+ err = di.dhLoading.LoadData(di.fsSeq, &di.sri.Data)
if err != nil {
return nil, err // XXX recheck
}
@@ -897,7 +901,7 @@ func (di *dataIter) NextData() (*zodb.StorageRecordInformation, error) {
di.dataBuf = di.sri.Data
}
- di.sri.DataTid = dh.Tid
+ di.sri.DataTid = di.dhLoading.Tid
return &di.sri, nil
}
diff --git a/t/neo/storage/fs1/filestorage_test.go b/t/neo/storage/fs1/filestorage_test.go
index 0425948d..fbfc63d6 100644
--- a/t/neo/storage/fs1/filestorage_test.go
+++ b/t/neo/storage/fs1/filestorage_test.go
@@ -208,6 +208,10 @@ func testIterate(t *testing.T, fs *FileStorage, tidMin, tidMax zodb.Tid, expectv
t.Fatal("unexpected datai pointer")
}
+ // compare data headers modulo .workMem
+ // (workMem is not initialized in _1fs_dbEntryv)
+ fsi.dataIter.Datah.workMem = dh.workMem
+
if !reflect.DeepEqual(fsi.dataIter.Datah, dh) {
dataErrorf("unexpected data entry:\nhave: %q\nwant: %q", fsi.dataIter.Datah, dh)
}
--
2.30.9