head_chunk
一个series采样了120个sample，或者是大于两小时的时候，会将series所生成的内存的chunk去写入磁盘，然后通过mmap的方式访问磁盘的chunk。
120是在gorilla论文的工程实践，压缩效果比较好。
// Based on Gorilla white papers this offers near-optimal compression ratio
// so anything bigger that this has diminishing returns and increases
// the time range within which we have to decompress all samples.
const samplesPerChunk = 120
实际的话，当一个series，samples数量到达1 / 4 的时候，就会去预测这么一个结束时间next_at,如果当前的t大于next_at就会写入mmap文件中。
（这种预测的方式会尝试让样本均匀分布）
// If we reach 25% of a chunk's desired sample count, predict an end time
// for this chunk that will try to make samples equally distributed within
// the remaining chunks in the current chunk range.
// At latest it must happen at the timestamp set when the chunk was cut.
chunkdiskmapper
chunk有一个curfle，mmap了一个128m的文件。
然后我们有个buffer writer，大小为4m，包装了curfile。
写入chunkdiskmapper的chunk，先写入curfile，直到这个buffer到达4m缓存后再flush这个buffer。
其实flush buffer，并不代表数据不会丢，只是说明写入内核缓冲区了。
而curfile彻底的安全需要sync curfile。
// ChunkDiskMapper is for writing the Head block chunks to the disk
// and access chunks via mmapped file.
type ChunkDiskMapper struct {
curFileNumBytes atomic.Int64 // Bytes written in current open file.
/// Writer.
dir *os.File
writeBufferSize int
curFile *os.File // File being written to.
curFileSequence int // Index of current open file being appended to.
curFileMaxt int64 // Used for the size retention.
byteBuf [MaxHeadChunkMetaSize]byte // Buffer used to write the header of the chunk.
chkWriter *bufio.Writer // Writer for the current open file.
crc32 hash.Hash
writePathMtx sync.Mutex
/// Reader.
// The int key in the map is the file number on the disk.
mmappedChunkFiles map[int]*mmappedChunkFile // Contains the m-mapped files for each chunk file mapped with its index.
closers map[int]io.Closer // Closers for resources behind the byte slices.
readPathMtx sync.RWMutex // Mutex used to protect the above 2 maps.
pool chunkenc.Pool // This is used when fetching a chunk from the disk to allocate a chunk.
// Writer and Reader.
// We flush chunks to disk in batches. Hence, we store them in this buffer
// from which chunks are served till they are flushed and are ready for m-mapping.
chunkBuffer *chunkBuffer
// Whether the maxt field is set for all mmapped chunk files tracked within the mmappedChunkFiles map.
// This is done after iterating through all the chunks in those files using the IterateAllChunks method.
fileMaxtSet bool
closed bool
}
cut
cut将当前的curfile，sync到磁盘里头。
然后使用mmap新建一个curfile，大小128m。
writechunk
chunkdiskmapper通过writechunk把当前series加密的信息写入mmap中的文件。
返回ChunkDiskMapperRef，前32位代表磁盘上的chunk文件，后32位表示chunkenc的offset偏移。
// WriteChunk writes the chunk to the disk.
// The returned chunk ref is the reference from where the chunk encoding starts for the chunk.
func (cdm *ChunkDiskMapper) WriteChunk(seriesRef HeadSeriesRef, mint, maxt int64, chk chunkenc.Chunk) (chkRef ChunkDiskMapperRef, err error) {
在writechunk的时候，会判断curfile是否大于128m，如果大于就cut。
重新mmap一个curfile。
chunk中存储的series数据如下。
| seriedref <8 字节> | mint <8 字节，uint64> | maxt <8 字节，uint64> | 编码 <1 字节> | len | 数据 <字节> │ CRC32 <4 字节> │
chunk
chunkdiskmapper通过chunk函数从ChunkDiskMapperRef获取对应的chunkenc：
-
通过ChunkDiskMapperRef，从chunkbuffer中获取。
-
通过前32位的segmentindex从mmappedChunkFiles中获取到mmappedChunkFile（一个[]byte加maxt）。
然后通过后32位获取的chunkstart，在这个[]byte中找到对应的数据，校验crc，然后重新copy返回。
重新copy的话是因为mmapfile不受内存上的控制。（这正是我们为什么会有chunkbuffer的原因，因为chunkbuffer在系统内存里面，无需copy。）
// Make a copy of the chunk data to prevent a panic occurring because the returned // chunk data slice references an mmap-ed file which could be closed after the // function returns but while the chunk is still in use. chkDataCopy := make([]byte, len(chkData)) copy(chkDataCopy, chkData)
chunkbuffer
当前有一个chunkbuffer，表示还在buffer里头的chunk。
让用户可以通过ChunkDiskMapperRef去直接访问对应的chunk。
（通过ChunkDiskMapperRef % inBufferShards ）找到对应的map。
通过分段锁的方式降低竞争。
这个chunkbuffer，会在flushbuffer的时候被clear。
const inBufferShards = 128 // 128 is a randomly chosen number.
// chunkBuffer is a thread safe lookup table for chunks by their ref.
type chunkBuffer struct {
inBufferChunks [inBufferShards]map[ChunkDiskMapperRef]chunkenc.Chunk
inBufferChunksMtxs [inBufferShards]sync.RWMutex
}
Truncate
truncate会遍历
mmappedChunkFile，对于那些maxt小于mint的文件。使用os.remove删除
// Truncate deletes the head chunk files which are strictly below the mint.
// mint should be in milliseconds.
func (cdm *ChunkDiskMapper) Truncate(mint int64) error {
