Signed-off-by: Fabian Reinartz <freinartz@google.com>
5.6 KiB
WAL Disk Format
The write ahead log operates in segments that that are numbered and sequential,
e.g. 000000
, 000001
, 000002
, etc., and are limited to 128MB by default.
A segment is written to in pages of 32KB. Only the last page of the most recent segment
may be partial. A WAL record is an opaque byte slice that gets split up into sub-records
should it exceed the remaining space of the current page. Records are never split across
segment boundaries.
The encoding of pages is largely borrowed from [LevelDB's/RocksDB's wirte ahead log.][1]
Notable deviations are that the record fragment is encoded as:
┌───────────┬──────────┬────────────┬──────────────┐ │ type <1b> │ len <2b> │ CRC32 <4b> │ data │ └───────────┴──────────┴────────────┴──────────────┘
Record encoding
The records written to the write ahead log are encoded as follows:
Series records
Series records encode the labels that identifier a series and its unique ID.
┌────────────────────────────────────────────┐ │ type = 1 <1b> │ ├────────────────────────────────────────────┤ │ ┌─────────┬──────────────────────────────┐ │ │ │ id <8b> │ n = len(labels) │ │ │ ├─────────┴────────────┬─────────────────┤ │ │ │ len(str_1) │ str_1 │ │ │ ├──────────────────────┴─────────────────┤ │ │ │ ... │ │ │ ├───────────────────────┬────────────────┤ │ │ │ len(str_2n) │ str_2n │ │ │ └───────────────────────┴────────────────┘ │ │ . . . │ └────────────────────────────────────────────┘
Sample records
Sample records encode samples as a list of triples (series_id, timestamp, value)
.
Series reference and timestamp are encoded as deltas w.r.t the first sample.
┌──────────────────────────────────────────────────────────────────┐ │ type = 2 <1b> │ ├──────────────────────────────────────────────────────────────────┤ │ ┌────────────────────┬───────────────────────────┬─────────────┐ │ │ │ id <8b> │ timestamp <8b> │ value <8b> │ │ │ └────────────────────┴───────────────────────────┴─────────────┘ │ │ ┌────────────────────┬───────────────────────────┬─────────────┐ │ │ │ id_delta │ timestamp_delta │ value <8b> │ │ │ └────────────────────┴───────────────────────────┴─────────────┘ │ │ . . . │ └──────────────────────────────────────────────────────────────────┘
Tombstone records
Tombstone records encode tombstones as a list of triples (series_id, min_time, max_time)
and specify an interval for which samples of a series got deleted.
┌─────────────────────────────────────────────────────┐ │ type = 3 <1b> │ ├─────────────────────────────────────────────────────┤ │ ┌─────────┬───────────────────┬───────────────────┐ │ │ │ id <8b> │ min_time │ max_time │ │ │ └─────────┴───────────────────┴───────────────────┘ │ │ . . . │ └─────────────────────────────────────────────────────┘
[1][https://github.com/facebook/rocksdb/wiki/Write-Ahead-Log-File-Format]