Proposal: Use struct to refine PageFile meta part

In pagefile.cpp function genWriteData or MetaMergingReader::moveNext , you can see we operate the PageFile meta info by put/get.

PageUtil::put
PageUtil::get

there are three disadvantages:

• The granularity of operation is too large, you can only operate the whole PageFile.
• All operation in write/reader function and also it can't split. So it hard to change and iteration new function into it.
  • it's hard to changed , if we keep use point read/write meta info.
• Poorly readable. eq. you can found out what's info being the meta.

Suggestion :

Use the struct replace put/get function. struct can quick serialization/deserialization a memory buffer with special alignment.

here is a example for it:

// for meta head
struct meta_header {
    uint64_t meta_byte_size : 32,
    uint64_t meta_version : 32,
    uint64_t meta_sequence_id : 64,
    uint64_t meta_crc : 64
}

struct meta_wb_entry {
    uint64_t wb_type : 8,
    union wb_put_insert {
        ....
    },
    union wb_del {
        ....
    },
    union wb_ref {
        ....
    }
}

struct flied_offset {
   ...
}

and also change writer/reader to :

• meta writer/reader
  • wb writer/reader
    • filed_offset writer/reader

meta writer/reader will hold wb writer/reader , and wb writer/reader will hold filed_offset writer/reader(if type is put/upsert).

don't worry about memory problem , cause meta only need hold one write/reader with different buffer. also it can replace the meta reader hasNext , moveNext operation.

and there are two advantage for this proposal:

• It can split different in disk layer in PageFile , and make it more flexable and readable. eq. you can add more change into it.
• It can help next generation PageFile format and better compatibility with different versions of PageFile
  • If the customer needs to switch the version of PageFile(when they want to upgrade tiflash version), or just migration the data. it will be more easy to do that.
• It can help make tools more convenient, eq . if i want to make a perf tool or a test tool, it can be more convenient.

@flowbehappy @JaySon-Huang can you review this proposal?

Start by reading pagefile.cpp, especially genWriteData and MetaMergingReader::moveNext, and trace how PageUtil::put and PageUtil::get currently handle PageFile metadata. Done would require an agreed struct-based metadata design plus the proposed layered meta, write-buffer, and field-offset readers and writers, with compatibility and tooling requirements resolved.