aboutsummaryrefslogtreecommitdiff
path: root/weed/filesys/page_writer/upload_pipeline.go
blob: 0c9e136491e569cb0517b825281b8edc805481e8 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
package page_writer

import (
	"github.com/chrislusf/seaweedfs/weed/glog"
	"github.com/chrislusf/seaweedfs/weed/util"
	"github.com/chrislusf/seaweedfs/weed/util/mem"
	"sync"
	"sync/atomic"
)

type UploadPipeline struct {
	writableChunks     map[LogicChunkIndex]*MemChunk
	writableChunksLock sync.Mutex
	sealedChunks       map[LogicChunkIndex]*SealedChunk
	sealedChunksLock   sync.Mutex
	ChunkSize          int64
	writers            *util.LimitedConcurrentExecutor
	activeWriterCond   *sync.Cond
	activeWriterCount  int32
	saveToStorageFn    SaveToStorageFunc
}

type SealedChunk struct {
	chunk            *MemChunk
	referenceCounter int // track uploading or reading processes
}

func (sc *SealedChunk) FreeReference() {
	sc.referenceCounter--
	if sc.referenceCounter == 0 {
		mem.Free(sc.chunk.buf)
	}
}

func NewUploadPipeline(writers *util.LimitedConcurrentExecutor, chunkSize int64, saveToStorageFn SaveToStorageFunc) *UploadPipeline {
	return &UploadPipeline{
		ChunkSize:        chunkSize,
		writableChunks:   make(map[LogicChunkIndex]*MemChunk),
		sealedChunks:     make(map[LogicChunkIndex]*SealedChunk),
		writers:          writers,
		activeWriterCond: sync.NewCond(&sync.Mutex{}),
		saveToStorageFn:  saveToStorageFn,
	}
}

func (cw *UploadPipeline) SaveDataAt(p []byte, off int64) (n int) {
	cw.writableChunksLock.Lock()
	defer cw.writableChunksLock.Unlock()

	logicChunkIndex := LogicChunkIndex(off / cw.ChunkSize)
	offsetRemainder := off % cw.ChunkSize

	memChunk, found := cw.writableChunks[logicChunkIndex]
	if !found {
		memChunk = &MemChunk{
			buf:   mem.Allocate(int(cw.ChunkSize)),
			usage: newChunkWrittenIntervalList(),
		}
		cw.writableChunks[logicChunkIndex] = memChunk
	}
	n = copy(memChunk.buf[offsetRemainder:], p)
	memChunk.usage.MarkWritten(offsetRemainder, offsetRemainder+int64(n))
	cw.maybeMoveToSealed(memChunk, logicChunkIndex)

	return
}

func (cw *UploadPipeline) MaybeReadDataAt(p []byte, off int64) (maxStop int64) {
	logicChunkIndex := LogicChunkIndex(off / cw.ChunkSize)

	// read from sealed chunks first
	cw.sealedChunksLock.Lock()
	sealedChunk, found := cw.sealedChunks[logicChunkIndex]
	if found {
		sealedChunk.referenceCounter++
	}
	cw.sealedChunksLock.Unlock()
	if found {
		maxStop = readMemChunk(sealedChunk.chunk, p, off, logicChunkIndex, cw.ChunkSize)
		sealedChunk.FreeReference()
	}

	// read from writable chunks last
	cw.writableChunksLock.Lock()
	defer cw.writableChunksLock.Unlock()
	writableChunk, found := cw.writableChunks[logicChunkIndex]
	if !found {
		return
	}
	maxStop = max(maxStop, readMemChunk(writableChunk, p, off, logicChunkIndex, cw.ChunkSize))

	return
}

func (cw *UploadPipeline) FlushAll() {
	cw.writableChunksLock.Lock()
	defer cw.writableChunksLock.Unlock()

	for logicChunkIndex, memChunk := range cw.writableChunks {
		cw.moveToSealed(memChunk, logicChunkIndex)
	}

	cw.waitForCurrentWritersToComplete()
}

func (cw *UploadPipeline) waitForCurrentWritersToComplete() {
	cw.activeWriterCond.L.Lock()
	t := int32(100)
	for {
		t = atomic.LoadInt32(&cw.activeWriterCount)
		if t <= 0 {
			break
		}
		glog.V(4).Infof("activeWriterCond is %d", t)
		cw.activeWriterCond.Wait()
	}
	cw.activeWriterCond.L.Unlock()
}

func (cw *UploadPipeline) maybeMoveToSealed(memChunk *MemChunk, logicChunkIndex LogicChunkIndex) {
	if memChunk.usage.IsComplete(cw.ChunkSize) {
		cw.moveToSealed(memChunk, logicChunkIndex)
	}
}

func (cw *UploadPipeline) moveToSealed(memChunk *MemChunk, logicChunkIndex LogicChunkIndex) {
	atomic.AddInt32(&cw.activeWriterCount, 1)
	glog.V(4).Infof("activeWriterCount %d ++> %d", cw.activeWriterCount-1, cw.activeWriterCount)

	cw.sealedChunksLock.Lock()

	if oldMemChunk, found := cw.sealedChunks[logicChunkIndex]; found {
		oldMemChunk.FreeReference()
	}
	sealedChunk := &SealedChunk{
		chunk:            memChunk,
		referenceCounter: 1, // default 1 is for uploading process
	}
	cw.sealedChunks[logicChunkIndex] = sealedChunk
	delete(cw.writableChunks, logicChunkIndex)

	cw.sealedChunksLock.Unlock()

	cw.writers.Execute(func() {
		cw.saveOneChunk(sealedChunk.chunk, logicChunkIndex)

		// remove from sealed chunks
		sealedChunk.FreeReference()
		cw.sealedChunksLock.Lock()
		defer cw.sealedChunksLock.Unlock()
		delete(cw.sealedChunks, logicChunkIndex)

		atomic.AddInt32(&cw.activeWriterCount, -1)
		glog.V(4).Infof("activeWriterCount %d --> %d", cw.activeWriterCount+1, cw.activeWriterCount)
		// Lock and Unlock are not required,
		// but it may signal multiple times during one wakeup,
		// and the waiting goroutine may miss some of them!
		cw.activeWriterCond.L.Lock()
		cw.activeWriterCond.Broadcast()
		cw.activeWriterCond.L.Unlock()
	})
}

func (cw *UploadPipeline) saveOneChunk(memChunk *MemChunk, logicChunkIndex LogicChunkIndex) {
	for t := memChunk.usage.head.next; t != memChunk.usage.tail; t = t.next {
		reader := util.NewBytesReader(memChunk.buf[t.StartOffset:t.stopOffset])
		cw.saveToStorageFn(reader, int64(logicChunkIndex)*cw.ChunkSize+t.StartOffset, t.Size(), func() {
		})
	}
}

func readMemChunk(memChunk *MemChunk, p []byte, off int64, logicChunkIndex LogicChunkIndex, chunkSize int64) (maxStop int64) {
	memChunkBaseOffset := int64(logicChunkIndex) * chunkSize
	for t := memChunk.usage.head.next; t != memChunk.usage.tail; t = t.next {
		logicStart := max(off, int64(logicChunkIndex)*chunkSize+t.StartOffset)
		logicStop := min(off+int64(len(p)), memChunkBaseOffset+t.stopOffset)
		if logicStart < logicStop {
			copy(p[logicStart-off:logicStop-off], memChunk.buf[logicStart-memChunkBaseOffset:logicStop-memChunkBaseOffset])
			maxStop = max(maxStop, logicStop)
		}
	}
	return
}

func (p2 *UploadPipeline) Shutdown() {

}