+// Copyright (C) The Lightning Authors. All rights reserved.
+//
+// SPDX-License-Identifier: AGPL-3.0
+
package lightning
import (
"encoding/gob"
"fmt"
"io"
+ "math/rand"
"os"
"regexp"
"runtime"
retainNoCalls bool
skipOOO bool
retainTileSequences bool
+ useDups bool
taglib *tagLibrary
variant [][][blake2b.Size256]byte
refseqs map[string]map[string][]tileLibRef
compactGenomes map[string][]tileVariantID
- // count [][]int
- seq map[[blake2b.Size256]byte][]byte
- variants int64
+ seq2 map[[2]byte]map[[blake2b.Size256]byte][]byte
+ seq2lock map[[2]byte]sync.Locker
+ variants int64
// if non-nil, write out any tile variants added while tiling
encoder *gob.Encoder
+ // set Ref flag when writing new variants to encoder
+ encodeRef bool
+
+ onAddTileVariant func(libref tileLibRef, hash [blake2b.Size256]byte, seq []byte) error
+ onAddGenome func(CompactGenome) error
+ onAddRefseq func(CompactSequence) error
mtx sync.RWMutex
vlock []sync.Locker
for _, tv := range tvs {
// Assign a new variant ID (unique across all inputs)
// for each input variant.
- variantmap[tileLibRef{Tag: tv.Tag, Variant: tv.Variant}] = tilelib.getRef(tv.Tag, tv.Sequence).Variant
+ variantmap[tileLibRef{Tag: tv.Tag, Variant: tv.Variant}] = tilelib.getRef(tv.Tag, tv.Sequence, tv.Ref).Variant
}
return nil
}
-func (tilelib *tileLibrary) loadCompactGenomes(cgs []CompactGenome, variantmap map[tileLibRef]tileVariantID, onLoadGenome func(CompactGenome)) error {
+func (tilelib *tileLibrary) loadCompactGenomes(cgs []CompactGenome, variantmap map[tileLibRef]tileVariantID) error {
log.Debugf("loadCompactGenomes: %d", len(cgs))
var wg sync.WaitGroup
errs := make(chan error, 1)
}
return
}
- log.Tracef("loadCompactGenomes: cg %s tag %d variant %d => %d", cg.Name, tag, variant, newvariant)
+ // log.Tracef("loadCompactGenomes: cg %s tag %d variant %d => %d", cg.Name, tag, variant, newvariant)
cg.Variants[i] = newvariant
}
- if onLoadGenome != nil {
- onLoadGenome(cg)
+ if tilelib.onAddGenome != nil {
+ err := tilelib.onAddGenome(cg)
+ if err != nil {
+ select {
+ case errs <- err:
+ default:
+ }
+ return
+ }
}
if tilelib.encoder != nil {
err := tilelib.encoder.Encode(LibraryEntry{
}
func (tilelib *tileLibrary) loadCompactSequences(cseqs []CompactSequence, variantmap map[tileLibRef]tileVariantID) error {
- log.Debugf("loadCompactSequences: %d", len(cseqs))
+ log.Infof("loadCompactSequences: %d todo", len(cseqs))
for _, cseq := range cseqs {
+ log.Infof("loadCompactSequences: checking %s", cseq.Name)
for _, tseq := range cseq.TileSequences {
for i, libref := range tseq {
if libref.Variant == 0 {
return err
}
}
+ if tilelib.onAddRefseq != nil {
+ err := tilelib.onAddRefseq(cseq)
+ if err != nil {
+ return err
+ }
+ }
+ log.Infof("loadCompactSequences: checking %s done", cseq.Name)
}
tilelib.mtx.Lock()
defer tilelib.mtx.Unlock()
for _, cseq := range cseqs {
tilelib.refseqs[cseq.Name] = cseq.TileSequences
}
+ log.Info("loadCompactSequences: done")
return nil
}
-func (tilelib *tileLibrary) LoadDir(ctx context.Context, path string, onLoadGenome func(CompactGenome)) error {
+func allFiles(path string, re *regexp.Regexp) ([]string, error) {
var files []string
- var walk func(string) error
- walk = func(path string) error {
- f, err := open(path)
- if err != nil {
- return err
- }
- defer f.Close()
- fis, err := f.Readdir(-1)
- if err != nil {
- files = append(files, path)
- return nil
- }
- for _, fi := range fis {
- if fi.Name() == "." || fi.Name() == ".." {
- continue
- } else if fi.IsDir() {
- err = walk(path + "/" + fi.Name())
- if err != nil {
- return err
- }
- } else if strings.HasSuffix(path, ".gob") || strings.HasSuffix(path, ".gob.gz") {
- files = append(files, path)
+ f, err := open(path)
+ if err != nil {
+ return nil, err
+ }
+ defer f.Close()
+ fis, err := f.Readdir(-1)
+ if err != nil {
+ return []string{path}, nil
+ }
+ for _, fi := range fis {
+ if fi.Name() == "." || fi.Name() == ".." {
+ continue
+ } else if child := path + "/" + fi.Name(); fi.IsDir() {
+ add, err := allFiles(child, re)
+ if err != nil {
+ return nil, err
}
+ files = append(files, add...)
+ } else if re == nil || re.MatchString(child) {
+ files = append(files, child)
}
- return nil
}
- err := walk(path)
+ sort.Strings(files)
+ return files, nil
+}
+
+var matchGobFile = regexp.MustCompile(`\.gob(\.gz)?$`)
+
+func (tilelib *tileLibrary) LoadDir(ctx context.Context, path string) error {
+ log.Infof("LoadDir: walk dir %s", path)
+ files, err := allFiles(path, matchGobFile)
if err != nil {
return err
}
ctx, cancel := context.WithCancel(ctx)
defer cancel()
var mtx sync.Mutex
- cgs := []CompactGenome{}
- cseqs := []CompactSequence{}
- variantmap := map[tileLibRef]tileVariantID{}
+ allcgs := make([][]CompactGenome, len(files))
+ allcseqs := make([][]CompactSequence, len(files))
+ allvariantmap := map[tileLibRef]tileVariantID{}
errs := make(chan error, len(files))
- for _, path := range files {
- path := path
+ log.Infof("LoadDir: read %d files", len(files))
+ for fileno, path := range files {
+ fileno, path := fileno, path
go func() {
f, err := open(path)
if err != nil {
return
}
defer f.Close()
- errs <- DecodeLibrary(f, strings.HasSuffix(path, ".gz"), func(ent *LibraryEntry) error {
+ defer log.Infof("LoadDir: finished reading %s", path)
+
+ var variantmap = map[tileLibRef]tileVariantID{}
+ var cgs []CompactGenome
+ var cseqs []CompactSequence
+ err = DecodeLibrary(f, strings.HasSuffix(path, ".gz"), func(ent *LibraryEntry) error {
if ctx.Err() != nil {
return ctx.Err()
}
- mtx.Lock()
- defer mtx.Unlock()
- if err := tilelib.loadTagSet(ent.TagSet); err != nil {
- return err
+ if len(ent.TagSet) > 0 {
+ mtx.Lock()
+ if tilelib.taglib == nil || tilelib.taglib.Len() != len(ent.TagSet) {
+ // load first set of tags, or
+ // report mismatch if 2 sets
+ // have different #tags.
+ if err := tilelib.loadTagSet(ent.TagSet); err != nil {
+ mtx.Unlock()
+ return err
+ }
+ }
+ mtx.Unlock()
}
- if err := tilelib.loadTileVariants(ent.TileVariants, variantmap); err != nil {
- return err
+ for _, tv := range ent.TileVariants {
+ variantmap[tileLibRef{Tag: tv.Tag, Variant: tv.Variant}] = tilelib.getRef(tv.Tag, tv.Sequence, tv.Ref).Variant
}
cgs = append(cgs, ent.CompactGenomes...)
cseqs = append(cseqs, ent.CompactSequences...)
return nil
})
+ allcgs[fileno] = cgs
+ allcseqs[fileno] = cseqs
+ mtx.Lock()
+ defer mtx.Unlock()
+ for k, v := range variantmap {
+ allvariantmap[k] = v
+ }
+ errs <- err
}()
}
for range files {
return err
}
}
- err = tilelib.loadCompactGenomes(cgs, variantmap, onLoadGenome)
+
+ log.Info("LoadDir: loadCompactGenomes")
+ var flatcgs []CompactGenome
+ for _, cgs := range allcgs {
+ flatcgs = append(flatcgs, cgs...)
+ }
+ err = tilelib.loadCompactGenomes(flatcgs, allvariantmap)
if err != nil {
return err
}
- err = tilelib.loadCompactSequences(cseqs, variantmap)
+
+ log.Info("LoadDir: loadCompactSequences")
+ var flatcseqs []CompactSequence
+ for _, cseqs := range allcseqs {
+ flatcseqs = append(flatcseqs, cseqs...)
+ }
+ err = tilelib.loadCompactSequences(flatcseqs, allvariantmap)
if err != nil {
return err
}
+
+ log.Info("LoadDir done")
return nil
}
func (tilelib *tileLibrary) WriteDir(dir string) error {
- nfiles := 128
+ ntilefiles := 128
+ nfiles := ntilefiles + len(tilelib.refseqs)
files := make([]*os.File, nfiles)
for i := range files {
f, err := os.OpenFile(fmt.Sprintf("%s/library.%04d.gob.gz", dir, i), os.O_CREATE|os.O_WRONLY, 0666)
for i := range encoders {
encoders[i] = gob.NewEncoder(zws[i])
}
+
+ cgnames := make([]string, 0, len(tilelib.compactGenomes))
+ for name := range tilelib.compactGenomes {
+ cgnames = append(cgnames, name)
+ }
+ sort.Strings(cgnames)
+
+ refnames := make([]string, 0, len(tilelib.refseqs))
+ for name := range tilelib.refseqs {
+ refnames = append(refnames, name)
+ }
+ sort.Strings(refnames)
+
+ log.Infof("WriteDir: writing %d files", nfiles)
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
errs := make(chan error, nfiles)
errs <- err
return
}
- if start == 0 {
- // For now, just write all the genomes and refs
- // to the first file
- for name, cg := range tilelib.compactGenomes {
- err := encoders[start].Encode(LibraryEntry{CompactGenomes: []CompactGenome{{
- Name: name,
- Variants: cg,
- }}})
- if err != nil {
- errs <- err
- return
- }
- }
- for name, tseqs := range tilelib.refseqs {
- err := encoders[start].Encode(LibraryEntry{CompactSequences: []CompactSequence{{
- Name: name,
- TileSequences: tseqs,
- }}})
- if err != nil {
- errs <- err
- return
- }
+ if refidx := start - ntilefiles; refidx >= 0 {
+ // write each ref to its own file
+ // (they seem to load very slowly)
+ name := refnames[refidx]
+ errs <- encoders[start].Encode(LibraryEntry{CompactSequences: []CompactSequence{{
+ Name: name,
+ TileSequences: tilelib.refseqs[name],
+ }}})
+ return
+ }
+ for i := start; i < len(cgnames); i += ntilefiles {
+ err := encoders[start].Encode(LibraryEntry{CompactGenomes: []CompactGenome{{
+ Name: cgnames[i],
+ Variants: tilelib.compactGenomes[cgnames[i]],
+ }}})
+ if err != nil {
+ errs <- err
+ return
}
}
tvs := []TileVariant{}
- for tag := start; tag < len(tilelib.variant) && ctx.Err() == nil; tag += nfiles {
+ for tag := start; tag < len(tilelib.variant) && ctx.Err() == nil; tag += ntilefiles {
tvs = tvs[:0]
for idx, hash := range tilelib.variant[tag] {
tvs = append(tvs, TileVariant{
Tag: tagID(tag),
Variant: tileVariantID(idx + 1),
Blake2b: hash,
- Sequence: tilelib.seq[hash],
+ Sequence: tilelib.hashSequence(hash),
})
}
err := encoders[start].Encode(LibraryEntry{TileVariants: tvs})
return err
}
}
+ log.Info("WriteDir: flushing")
for i := range zws {
err := zws[i].Close()
if err != nil {
return err
}
}
+ log.Info("WriteDir: done")
return nil
}
// Load library data from rdr. Tile variants might be renumbered in
// the process; in that case, genomes variants will be renumbered to
// match.
-//
-// If onLoadGenome is non-nil, call it on each CompactGenome entry.
-func (tilelib *tileLibrary) LoadGob(ctx context.Context, rdr io.Reader, gz bool, onLoadGenome func(CompactGenome)) error {
+func (tilelib *tileLibrary) LoadGob(ctx context.Context, rdr io.Reader, gz bool) error {
cgs := []CompactGenome{}
cseqs := []CompactSequence{}
variantmap := map[tileLibRef]tileVariantID{}
if ctx.Err() != nil {
return ctx.Err()
}
- err = tilelib.loadCompactGenomes(cgs, variantmap, onLoadGenome)
+ err = tilelib.loadCompactGenomes(cgs, variantmap)
if err != nil {
return err
}
}
type importStats struct {
- InputFile string
- InputLabel string
- InputLength int
- InputCoverage int
- PathLength int
- DroppedOutOfOrderTiles int
+ InputFile string
+ InputLabel string
+ InputLength int
+ InputCoverage int
+ PathLength int
+ DroppedRepeatedTags int
+ DroppedOutOfOrderTags int
}
-func (tilelib *tileLibrary) TileFasta(filelabel string, rdr io.Reader, matchChromosome *regexp.Regexp) (tileSeq, []importStats, error) {
+func (tilelib *tileLibrary) TileFasta(filelabel string, rdr io.Reader, matchChromosome *regexp.Regexp, isRef bool) (tileSeq, []importStats, error) {
ret := tileSeq{}
- type jobT struct {
- label string
- fasta []byte
- }
- todo := make(chan jobT, 1)
- scanner := bufio.NewScanner(rdr)
- go func() {
- defer close(todo)
- var fasta []byte
- var seqlabel string
- for scanner.Scan() {
- buf := scanner.Bytes()
- if len(buf) > 0 && buf[0] == '>' {
- todo <- jobT{seqlabel, append([]byte(nil), fasta...)}
- seqlabel, fasta = strings.SplitN(string(buf[1:]), " ", 2)[0], fasta[:0]
- log.Debugf("%s %s reading fasta", filelabel, seqlabel)
- } else {
- fasta = append(fasta, bytes.ToLower(buf)...)
- }
- }
- todo <- jobT{seqlabel, fasta}
- }()
type foundtag struct {
pos int
tagid tagID
skippedSequences := 0
taglen := tilelib.taglib.TagLen()
var stats []importStats
- for job := range todo {
- if len(job.fasta) == 0 {
- continue
- } else if !matchChromosome.MatchString(job.label) {
+
+ in := bufio.NewReader(rdr)
+readall:
+ for {
+ var seqlabel string
+ // Advance to next '>', then
+ // read seqlabel up to \r?\n
+ readseqlabel:
+ for seqlabelStarted := false; ; {
+ rune, _, err := in.ReadRune()
+ if err == io.EOF {
+ break readall
+ } else if err != nil {
+ return nil, nil, err
+ }
+ switch {
+ case rune == '\r':
+ case seqlabelStarted && rune == '\n':
+ break readseqlabel
+ case seqlabelStarted:
+ seqlabel += string(rune)
+ case rune == '>':
+ seqlabelStarted = true
+ default:
+ }
+ }
+ log.Debugf("%s %s reading fasta", filelabel, seqlabel)
+ if !matchChromosome.MatchString(seqlabel) {
skippedSequences++
continue
}
- log.Debugf("%s %s tiling", filelabel, job.label)
+ log.Debugf("%s %s tiling", filelabel, seqlabel)
+ fasta := bytes.NewBuffer(nil)
found = found[:0]
- tilelib.taglib.FindAll(job.fasta, func(tagid tagID, pos, taglen int) {
+ err := tilelib.taglib.FindAll(in, fasta, func(tagid tagID, pos, taglen int) {
found = append(found, foundtag{pos: pos, tagid: tagid})
})
+ if err != nil {
+ return nil, nil, err
+ }
totalFoundTags += len(found)
if len(found) == 0 {
- log.Warnf("%s %s no tags found", filelabel, job.label)
+ log.Warnf("%s %s no tags found", filelabel, seqlabel)
+ }
+
+ droppedDup := 0
+ if !tilelib.useDups {
+ // Remove any tags that appeared more than once
+ dup := map[tagID]bool{}
+ for _, ft := range found {
+ _, dup[ft.tagid] = dup[ft.tagid]
+ }
+ dst := 0
+ for _, ft := range found {
+ if !dup[ft.tagid] {
+ found[dst] = ft
+ dst++
+ }
+ }
+ droppedDup = len(found) - dst
+ log.Infof("%s %s dropping %d non-unique tags", filelabel, seqlabel, droppedDup)
+ found = found[:dst]
}
- skipped := 0
+ droppedOOO := 0
if tilelib.skipOOO {
- log.Infof("%s %s keeping longest increasing subsequence", filelabel, job.label)
keep := longestIncreasingSubsequence(len(found), func(i int) int { return int(found[i].tagid) })
for i, x := range keep {
found[i] = found[x]
}
- skipped = len(found) - len(keep)
+ droppedOOO = len(found) - len(keep)
+ log.Infof("%s %s dropping %d out-of-order tags", filelabel, seqlabel, droppedOOO)
found = found[:len(keep)]
}
- log.Infof("%s %s getting %d librefs", filelabel, job.label, len(found))
- throttle := &throttle{Max: runtime.NumCPU()}
+ log.Infof("%s %s getting %d librefs", filelabel, seqlabel, len(found))
path = path[:len(found)]
var lowquality int64
- for i, f := range found {
- i, f := i, f
- throttle.Acquire()
- go func() {
- defer throttle.Release()
- var startpos, endpos int
- if i == 0 {
- startpos = 0
- } else {
- startpos = f.pos
- }
- if i == len(found)-1 {
- endpos = len(job.fasta)
- } else {
- endpos = found[i+1].pos + taglen
- }
- path[i] = tilelib.getRef(f.tagid, job.fasta[startpos:endpos])
- if countBases(job.fasta[startpos:endpos]) != endpos-startpos {
- atomic.AddInt64(&lowquality, 1)
- }
- }()
+ // Visit each element of found, but start at a random
+ // index, to reduce the likelihood of lock contention
+ // when importing many samples concurrently.
+ startpoint := rand.Int() % len(found)
+ for offset := range found {
+ i := startpoint + offset
+ if i >= len(found) {
+ i -= len(found)
+ }
+ f := found[i]
+ var startpos, endpos int
+ if i == 0 {
+ startpos = 0
+ } else {
+ startpos = f.pos
+ }
+ if i == len(found)-1 {
+ endpos = fasta.Len()
+ } else {
+ endpos = found[i+1].pos + taglen
+ }
+ path[i] = tilelib.getRef(f.tagid, fasta.Bytes()[startpos:endpos], isRef)
+ if countBases(fasta.Bytes()[startpos:endpos]) != endpos-startpos {
+ lowquality++
+ }
}
- throttle.Wait()
- log.Infof("%s %s copying path", filelabel, job.label)
+ log.Infof("%s %s copying path", filelabel, seqlabel)
pathcopy := make([]tileLibRef, len(path))
copy(pathcopy, path)
- ret[job.label] = pathcopy
+ ret[seqlabel] = pathcopy
- basesIn := countBases(job.fasta)
- log.Infof("%s %s fasta in %d coverage in %d path len %d low-quality %d skipped-out-of-order %d", filelabel, job.label, len(job.fasta), basesIn, len(path), lowquality, skipped)
+ basesIn := countBases(fasta.Bytes())
+ log.Infof("%s %s fasta in %d coverage in %d path len %d low-quality %d", filelabel, seqlabel, fasta.Len(), basesIn, len(path), lowquality)
stats = append(stats, importStats{
- InputFile: filelabel,
- InputLabel: job.label,
- InputLength: len(job.fasta),
- InputCoverage: basesIn,
- PathLength: len(path),
- DroppedOutOfOrderTiles: skipped,
+ InputFile: filelabel,
+ InputLabel: seqlabel,
+ InputLength: fasta.Len(),
+ InputCoverage: basesIn,
+ PathLength: len(path),
+ DroppedOutOfOrderTags: droppedOOO,
+ DroppedRepeatedTags: droppedDup,
})
totalPathLen += len(path)
}
log.Printf("%s tiled with total path len %d in %d sequences (skipped %d sequences that did not match chromosome regexp, skipped %d out-of-order tags)", filelabel, totalPathLen, len(ret), skippedSequences, totalFoundTags-totalPathLen)
- return ret, stats, scanner.Err()
+ return ret, stats, nil
}
func (tilelib *tileLibrary) Len() int64 {
// Return a tileLibRef for a tile with the given tag and sequence,
// adding the sequence to the library if needed.
-func (tilelib *tileLibrary) getRef(tag tagID, seq []byte) tileLibRef {
+func (tilelib *tileLibrary) getRef(tag tagID, seq []byte, usedByRef bool) tileLibRef {
dropSeq := false
if !tilelib.retainNoCalls {
for _, b := range seq {
vlock.Unlock()
if tilelib.retainTileSequences && !dropSeq {
- tilelib.mtx.Lock()
- if tilelib.seq == nil {
- tilelib.seq = map[[blake2b.Size256]byte][]byte{}
+ seqCopy := append([]byte(nil), seq...)
+ if tilelib.seq2 == nil {
+ tilelib.mtx.Lock()
+ if tilelib.seq2 == nil {
+ tilelib.seq2lock = map[[2]byte]sync.Locker{}
+ m := map[[2]byte]map[[blake2b.Size256]byte][]byte{}
+ var k [2]byte
+ for i := 0; i < 256; i++ {
+ k[0] = byte(i)
+ for j := 0; j < 256; j++ {
+ k[1] = byte(j)
+ m[k] = map[[blake2b.Size256]byte][]byte{}
+ tilelib.seq2lock[k] = &sync.Mutex{}
+ }
+ }
+ tilelib.seq2 = m
+ }
+ tilelib.mtx.Unlock()
}
- tilelib.seq[seqhash] = append([]byte(nil), seq...)
- tilelib.mtx.Unlock()
+ var k [2]byte
+ copy(k[:], seqhash[:])
+ locker := tilelib.seq2lock[k]
+ locker.Lock()
+ tilelib.seq2[k][seqhash] = seqCopy
+ locker.Unlock()
}
+ saveSeq := seq
+ if dropSeq {
+ // Save the hash, but not the sequence
+ saveSeq = nil
+ }
if tilelib.encoder != nil {
- saveSeq := seq
- if dropSeq {
- // Save the hash, but not the sequence
- saveSeq = nil
- }
tilelib.encoder.Encode(LibraryEntry{
TileVariants: []TileVariant{{
Tag: tag,
+ Ref: usedByRef,
Variant: variant,
Blake2b: seqhash,
Sequence: saveSeq,
}},
})
}
+ if tilelib.onAddTileVariant != nil {
+ tilelib.onAddTileVariant(tileLibRef{tag, variant}, seqhash, saveSeq)
+ }
return tileLibRef{Tag: tag, Variant: variant}
}
+func (tilelib *tileLibrary) hashSequence(hash [blake2b.Size256]byte) []byte {
+ var partition [2]byte
+ copy(partition[:], hash[:])
+ return tilelib.seq2[partition][hash]
+}
+
func (tilelib *tileLibrary) TileVariantSequence(libref tileLibRef) []byte {
if libref.Variant == 0 || len(tilelib.variant) <= int(libref.Tag) || len(tilelib.variant[libref.Tag]) < int(libref.Variant) {
return nil
}
- return tilelib.seq[tilelib.variant[libref.Tag][libref.Variant-1]]
+ return tilelib.hashSequence(tilelib.variant[libref.Tag][libref.Variant-1])
}
// Tidy deletes unreferenced tile variants and renumbers variants so
projects / lightning.git / blobdiff