OnDiskGraphDB.cpp File Reference (original) (raw)

This file implements OnDiskGraphDB, an on-disk CAS nodes database, independent of a particular hashing algorithm. More...

Macros
#define	DEBUG_TYPE "on-disk-cas"

Variables
static constexpr StringLiteral	IndexTableName = "llvm.cas.index"
static constexpr StringLiteral	DataPoolTableName = "llvm.cas.data"
static constexpr StringLiteral	IndexFilePrefix = "index."
static constexpr StringLiteral	DataPoolFilePrefix = "data."
static constexpr StringLiteral	FilePrefixObject = "obj."
static constexpr StringLiteral	FilePrefixLeaf = "leaf."
static constexpr StringLiteral	FilePrefixLeaf0 = "leaf+0."

This file implements OnDiskGraphDB, an on-disk CAS nodes database, independent of a particular hashing algorithm.

It only needs to be configured for the hash size and controls the schema of the storage.

How the data is stored inside database, either as a standalone file, or allocated inside a datapool.
How references to other objects inside the same database is stored. They are stored as internal references, instead of full hash value to save space.
How to chain databases together and import objects from upstream databases.

Here's a top-level description of the current layout:

db/index.: a file for the "index" table, named by IndexTableName and managed by TrieRawHashMap. The contents are 8B that are accessed atomically, describing the object kind and where/how it's stored (including an optional file offset). See TrieRecord for more details.
db/data.: a file for the "data" table, named by DataPoolTableName and managed by DataStore. New objects within TrieRecord::MaxEmbeddedSize are inserted here as TrieRecord::StorageKind::DataPool.
db/obj..: a file storing an object outside the main "data" table, named by its offset into the "index" table, with the format of TrieRecord::StorageKind::Standalone.
db/leaf..: a file storing a leaf node outside the main "data" table, named by its offset into the "index" table, with the format of TrieRecord::StorageKind::StandaloneLeaf.
db/leaf+0..: a file storing a null-terminated leaf object outside the main "data" table, named by its offset into the "index" table, with the format of TrieRecord::StorageKind::StandaloneLeaf0.

Definition in file OnDiskGraphDB.cpp.

◆ DEBUG_TYPE

#define DEBUG_TYPE "on-disk-cas"

size_t getPageSize ( )	static