1leveldb Log format 2================== 3The log file contents are a sequence of 32KB blocks. The only exception is that 4the tail of the file may contain a partial block. 5 6Each block consists of a sequence of records: 7 8 block := record* trailer? 9 record := 10 checksum: uint32 // crc32c of type and data[] ; little-endian 11 length: uint16 // little-endian 12 type: uint8 // One of FULL, FIRST, MIDDLE, LAST 13 data: uint8[length] 14 15A record never starts within the last six bytes of a block (since it won't fit). 16Any leftover bytes here form the trailer, which must consist entirely of zero 17bytes and must be skipped by readers. 18 19Aside: if exactly seven bytes are left in the current block, and a new non-zero 20length record is added, the writer must emit a FIRST record (which contains zero 21bytes of user data) to fill up the trailing seven bytes of the block and then 22emit all of the user data in subsequent blocks. 23 24More types may be added in the future. Some Readers may skip record types they 25do not understand, others may report that some data was skipped. 26 27 FULL == 1 28 FIRST == 2 29 MIDDLE == 3 30 LAST == 4 31 32The FULL record contains the contents of an entire user record. 33 34FIRST, MIDDLE, LAST are types used for user records that have been split into 35multiple fragments (typically because of block boundaries). FIRST is the type 36of the first fragment of a user record, LAST is the type of the last fragment of 37a user record, and MIDDLE is the type of all interior fragments of a user 38record. 39 40Example: consider a sequence of user records: 41 42 A: length 1000 43 B: length 97270 44 C: length 8000 45 46**A** will be stored as a FULL record in the first block. 47 48**B** will be split into three fragments: first fragment occupies the rest of 49the first block, second fragment occupies the entirety of the second block, and 50the third fragment occupies a prefix of the third block. This will leave six 51bytes free in the third block, which will be left empty as the trailer. 52 53**C** will be stored as a FULL record in the fourth block. 54 55---- 56 57## Some benefits over the recordio format: 58 591. We do not need any heuristics for resyncing - just go to next block boundary 60 and scan. If there is a corruption, skip to the next block. As a 61 side-benefit, we do not get confused when part of the contents of one log 62 file are embedded as a record inside another log file. 63 642. Splitting at approximate boundaries (e.g., for mapreduce) is simple: find the 65 next block boundary and skip records until we hit a FULL or FIRST record. 66 673. We do not need extra buffering for large records. 68 69## Some downsides compared to recordio format: 70 711. No packing of tiny records. This could be fixed by adding a new record type, 72 so it is a shortcoming of the current implementation, not necessarily the 73 format. 74 752. No compression. Again, this could be fixed by adding new record types. 76