What is the process understanding of this part of the code？

[dannyfe]

How to write after hitting a slice or creating a new slice is not well understood. Can someone help you understand it

juicefs/pkg/chunk/cached_store.go

Line 337 in f07634f

func (c *wChunk) WriteAt(p []byte, off int64) (n int, err error) {

func (c *wChunk) WriteAt(p []byte, off int64) (n int, err error) {
if int(off)+len(p) > chunkSize {
return 0, fmt.Errorf("write out of chunk boudary: %d > %d", int(off)+len(p), chunkSize)
}
if off < int64(c.uploaded) {
return 0, fmt.Errorf("Cannot overwrite uploaded block: %d < %d", off, c.uploaded)
}

// Fill previous blocks with zeros
if c.length < int(off) {
zeros := make([]byte, int(off)-c.length)
_, _ = c.WriteAt(zeros, int64(c.length)) //填充本次写入数据到上一次数据记录点得空白块
}

for n < len(p) {
//off = off - s.off 当前slice相对于本次写入偏移量得差值 13M
indx := c.index(int(off) + n)// (int(off) + n / c.store.conf.BlockSize) // 3
boff := (int(off) + n) % c.store.conf.BlockSize // 1

  var bs = pageSize
  if indx > 0 || bs > c.store.conf.BlockSize {
     bs = c.store.conf.BlockSize
  }

  bi := boff / bs //16
  bo := boff % bs //0
  var page *Page
  if bi < len(c.pages[indx]) { //16 < 0
     page = c.pages[indx][bi]
  } else {
     page = allocPage(bs)
     page.Data = page.Data[:0]
     c.pages[indx] = append(c.pages[indx], page)
  }
  left := len(p) - n
  if bo+left > bs {
     page.Data = page.Data[:bs]
  } else if len(page.Data) < bo+left {
     page.Data = page.Data[:bo+left]
  }
  n += copy(page.Data[bo:], p[n:])

}
if int(off)+n > c.length {
c.length = int(off) + n
}
return n, nil
}

[SandyXSD]

Basically this function is used to write data p to the slice (aka, wChunk). There is a field pages in wChunk that is used to cache data, so what the function does is just splitting p in an aligned way and copy bytes to pages.

As wChunk is split to several Blocks for further writing to the object storage, pages is a list of memory blocks. For better memory management, the first block in pages is split again to 64 KiB pages, in case len(p) is far smaller than BlockSize (defaults 4 MiB).

The following figure may help you better understand the process:

         64K 64K ...
wChunk: |_  _  _  _  ...  _|________|________|__...
               4 M             4M       4M      ...

     p:         |_______________________|
              offset               

[dannyfe]

I sorted out the process according to the reply, and the data writing process is much clearer. Thank you for your answer