mongodb journal文件格式

来源:互联网 发布:罗技优联mac能用吗 编辑:程序博客网 时间:2024/05/28 15:19

如果开启journal,在dbpath选项指定的目录下会创建journal目录来存放journal文件,文件名形如j._<n>。

journal文件存储的是对数据库文件(dbname.ns、dbname.<#>系列文件)的修改日志,包括写操作和创建文件操作。对数据库文件的写操作会记录一个WriteIntent,创建数据库文件会记录一个DurOp。WriteIntent记录了写操作的指针和长度,可以定位到修改的数据文件的位置和长度。DurOp由一个操作码来确定是什么操作,不同的操作,日志的格式不一样。每个WriteIntent或者DurOp都会形成一个JEntry。

struct WriteIntent { /* copyable */private:    void *p;      // intent to write up to p    unsigned len; // up to this len};class DurOp { /* copyable */        public:                        /** serialize the op out to a builder which will then be written (presumably) to the journal */            void serialize(AlignedBuilder& ab);                        virtual void replay() = 0;        protected:            /** DurOp will have already written the opcode for you */            virtual void _serialize(AlignedBuilder& ab) = 0;        private:            const unsigned _opcode;        };

journal文件由一个头部JHeader和很多Section组成,每次group commit都会产生一个Section。Section由一个头部JSectHeader、很多个JEntry和一个JSectFooter组成,每个JEntry代表一条修改日志。另外,JEntry前面可能会有一个JDbContext,表示修改的是哪个数据文件。如果多个JEntry都是同一个数据文件的操作,则只有一个JDbContext。除掉JSectHeader 和JSectFooter ,Section中间那部分会压缩。

 

JHeader

beginning header for a journal/j._<n> file
there is nothing important int this header at this time. except perhaps version #.

char magic[2] "j\n". j means journal, then a linefeed, fwiw if you were to run "less" on the file or something... unsigned short _version 0x4149 char n1 '\n' char ts[20] ascii timestamp of file generation.  for user reading, not used by code. for example: "Jan 29 14:14:23" char n2 '\n' char dbpath[128] path/filename of this file for human reading and diagnostics.  not used by code char n3, n4 '\n', '\n' unsigned long long fileId unique identifier that will be in each JSectHeader. important as we recycle prealloced files char reserved3[8026] 8KB total for the file header char txt2[2] "\n\n" at the end

头部的大部分字段都是可读的,很直观,用文本文件打开可以看到。

JSectHeader

"Section" header. A section corresponds to a group commit.len is length of the entire section including header and footer.header and footer are not compressed, just the stuff in between.

unsigned _sectionLen unpadded length in bytes of the whole section unsigned long long seqNumber sequence number that can be used on recovery to not do too much work unsigned long long fileId matches JHeader::fileId

_sectionLen是unpadded的长度,每个Section都会填充对齐。

每个Section都会有一个seqNumber,其值是上次同步数据库文件到磁盘的服务器时间,会不断增长。LSNFile(文件名是lsn)里面也存储了一个seqNumber,含义是一样的。恢复数据的时候,如果Section的seqNumber小于LSNFile的seqNumber,则不需要恢复。

JEntry

an individual write operation within a group commit section. Either the entire section should be applied, or nothing. (We check the md5 for the whole section before doing anything on recovery.)

unsigned len  length in bytes of the data of the JEntry. does not include the JEntry header. 实际就是数据data的长度 OpCodes opcode 此字段和上面len字段共用的。如果是对数据库文件的修改,则为长度len;否则为操作类型。 unsigned ofs offset in file. 指的是被修改的数据文件的偏移量 int _fileNo
high bit is set to indicate it should be the <dbpath>/local database. dbname.#文件中的数字,即#。
如果是dbname.ns文件,则值为0x7fffffff。高位bit为1表示是local库。
char data[] 更新的数据

 OpCodes的取值

OpCode_Footer JSectFooter的sentinel字段的取值 OpCode_DbContext JDbContext的sentinel字段的取值 OpCode_FileCreated 表示创建pdfile的操作 OpCode_DropDb 表示dropDb的操作,目前没有使用

这几个取值都很大,因为跟JEntry的len是共用,必须比可能的长度值大得多
事实上JEntry、JDbContext和JSectFooter的前4字节是含义是相同的。这样比较容易解析。

当JEntry为Op时,格式根据不同的Op而变化,如OpCode_FileCreated的格式为

OpCodes opcode OpCode_FileCreated unsigned long long reserved1   unsigned long long reserved2   unsigned long long _len size of file, not length of name RelativePath _p 文件名的相对路径,以'\0'结尾的字符串

 

JDbContext

declares "the next entry(s) are for this database / file path prefix"

sentinel 哨兵值,为了解析方便,为OpCode_DbContext file path dbname.#文件的相对路径,相对于dbpath,以'\0'结尾的字符串。使用相对路径是为了节省空间

 

JSectFooter

group commit section footer. md5 is a key field.

unsigned sentinel 哨兵值,为了解析方便,为OpCode_Footer unsigned char hash[16] 除掉JSectFooter,整个Section的checksum unsigned long long reserved   char magic[4] "\n\n\n\n"

 

LSNFile

last sequence number. lsn文件。

unsigned ver 为0,用以检查是否合法 unsigned reserved2   unsigned long long lsn last sequence number。值为数据库文件上次同步到磁盘的时间。 unsigned long long checkbytes lsn的值取反,用以检查是否合法 unsigned long long reserved[8]  

 


<script type="text/javascript"><!--google_ad_client = "ca-pub-1944176156128447";/* cnblogs 首页横幅 */google_ad_slot = "5419468456";google_ad_width = 728;google_ad_height = 90;//--></script><script type="text/javascript" src="http://pagead2.googlesyndication.com/pagead/show_ads.js"></script>