+// Copyright (C) The Lightning Authors. All rights reserved.
+//
+// SPDX-License-Identifier: AGPL-3.0
+
package lightning
import (
"encoding/gob"
"fmt"
"io"
+ "os"
"regexp"
"runtime"
"sort"
"sync"
"sync/atomic"
+ "github.com/klauspost/pgzip"
log "github.com/sirupsen/logrus"
"golang.org/x/crypto/blake2b"
)
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 allFiles(path string, re *regexp.Regexp) ([]string, error) {
+ var files []string
+ 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)
+ }
+ }
+ 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
+ allcgs := make([][]CompactGenome, len(files))
+ allcseqs := make([][]CompactSequence, len(files))
+ allvariantmap := map[tileLibRef]tileVariantID{}
+ errs := make(chan error, len(files))
+ 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 {
+ errs <- err
+ return
+ }
+ defer f.Close()
+ 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()
+ }
+ 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()
+ }
+ 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 {
+ err := <-errs
+ if err != nil {
+ return err
+ }
+ }
+
+ 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
+ }
+
+ 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 {
+ 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)
+ if err != nil {
+ return err
+ }
+ defer f.Close()
+ files[i] = f
+ }
+ bufws := make([]*bufio.Writer, nfiles)
+ for i := range bufws {
+ bufws[i] = bufio.NewWriterSize(files[i], 1<<26)
+ }
+ zws := make([]*pgzip.Writer, nfiles)
+ for i := range zws {
+ zws[i] = pgzip.NewWriter(bufws[i])
+ defer zws[i].Close()
+ }
+ encoders := make([]*gob.Encoder, nfiles)
+ 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)
+ for start := range files {
+ start := start
+ go func() {
+ err := encoders[start].Encode(LibraryEntry{TagSet: tilelib.taglib.Tags()})
+ 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 += 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.hashSequence(hash),
+ })
+ }
+ err := encoders[start].Encode(LibraryEntry{TileVariants: tvs})
+ if err != nil {
+ errs <- err
+ return
+ }
+ }
+ errs <- nil
+ }()
+ }
+ for range files {
+ err := <-errs
+ if err != nil {
+ return err
+ }
+ }
+ log.Info("WriteDir: flushing")
+ for i := range zws {
+ err := zws[i].Close()
+ if err != nil {
+ return err
+ }
+ err = bufws[i].Flush()
+ if err != nil {
+ return err
+ }
+ err = files[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
}
todo := make(chan jobT, 1)
scanner := bufio.NewScanner(rdr)
+ scanner.Buffer(make([]byte, 256), 1<<29) // 512 MiB, in case fasta does not have line breaks
go func() {
defer close(todo)
var fasta []byte
log.Warnf("%s %s no tags found", filelabel, job.label)
}
- skipped := 0
+ 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, job.label, droppedDup)
+ found = found[:dst]
+ }
+
+ 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, job.label, droppedOOO)
found = found[:len(keep)]
}
} else {
endpos = found[i+1].pos + taglen
}
- path[i] = tilelib.getRef(f.tagid, job.fasta[startpos:endpos])
+ path[i] = tilelib.getRef(f.tagid, job.fasta[startpos:endpos], isRef)
if countBases(job.fasta[startpos:endpos]) != endpos-startpos {
atomic.AddInt64(&lowquality, 1)
}
ret[job.label] = 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)
+ log.Infof("%s %s fasta in %d coverage in %d path len %d low-quality %d", filelabel, job.label, len(job.fasta), 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: job.label,
+ InputLength: len(job.fasta),
+ InputCoverage: basesIn,
+ PathLength: len(path),
+ DroppedOutOfOrderTags: droppedOOO,
+ DroppedRepeatedTags: droppedDup,
})
totalPathLen += len(path)
// 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