andreacavalli/rocksdb
0
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Update comments in column_family.h (#5347)

Browse Source 
Summary:
Document relationships of data structures declared in column_family.h
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5347

Differential Revision: D15496941

Pulled By: siying

fbshipit-source-id: 47b37835abba26aa31a94fabea6b2775483e0ccb
This commit is contained in:
Siying Dong 2019-05-24 12:03:16 -07:00 committed by Facebook Github Bot
parent 767d1f3ff1
commit 596cc1547a
1 changed files with 107 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

106
db/column_family.h
View File

@ -45,6 +45,112 @@ class InstrumentedMutexLock;
struct SuperVersionContext; struct SuperVersionContext;
extern const double kIncSlowdownRatio; extern const double kIncSlowdownRatio;
// This file contains a list of data structures for managing column family
// level metadata.
//
// The basic relationships among classes declared here are illustrated as
// following:
//
// +----------------------+ +----------------------+ +--------+
// +---+ ColumnFamilyHandle 1 | +--+ ColumnFamilyHandle 2 | | DBImpl |
// | +----------------------+ | +----------------------+ +----+---+
// | +--------------------------+ |
// | | +-----------------------------+
// | | |
// | | +-----------------------------v-------------------------------+
// | | | |
// | | | ColumnFamilySet |
// | | | |
// | | +-------------+--------------------------+----------------+---+
// | | | | |
// | +-------------------------------------+ | |
// | | | | v
// | +-------------v-------------+ +-----v----v---------+
// | | | | |
// | | ColumnFamilyData 1 | | ColumnFamilyData 2 | ......
// | | | | |
// +---> | | |
// | +---------+ | |
// | | MemTable| | |
// | | List | | |
// +--------+---+--+-+----+----+ +--------------------++
// | | | |
// | | | |
// | | | +-----------------------+
// | | +-----------+ |
// v +--------+ | |
// +--------+--------+ | | |
// | | | | +----------v----------+
// +---> |SuperVersion 1.a +-----------------> |
// | +------+ | | MemTableListVersion |
// +---+-------------+ | | | | |
// | | | | +----+------------+---+
// | current | | | | |
// | +-------------+ | |mem | |
// | | | | | |
// +-v---v-------+ +---v--v---+ +-----v----+ +----v-----+
// | | | | | | | |
// | Version 1.a | | memtable | | memtable | | memtable |
// | | | 1.a | | 1.b | | 1.c |
// +-------------+ | | | | | |
// +----------+ +----------+ +----------+
//
// DBImpl keeps a ColumnFamilySet, which references to all column families by
// pointing to respective ColumnFamilyData object of each column family.
// This is how DBImpl can list and operate on all the column families.
// ColumnFamilyHandle also points to ColumnFamilyData directly, so that
// when a user executes a query, it can directly find memtables and Version
// as well as SuperVersion to the column family, without going through
// ColumnFamilySet.
//
// ColumnFamilySet points to the latest view of the LSM-tree (list of memtables
// and SST files) indirectly, while ongoing operations may hold references
// to a current or an out-of-date SuperVersion, which in turn points to a
// point-in-time view of the LSM-tree. This guarantees the memtables and SST
// files being operated on will not go away, until the SuperVersion is
// unreferenced to 0 and destoryed.
//
// The following graph illustrates a possible referencing relationships:
//
// Column +--------------+ current +-----------+
// Family +---->+ +------------------->+ |
// Data | SuperVersion +----------+ | Version A |
// | 3 | imm | | |
// Iter2 +----->+ | +-------v------+ +-----------+
// +-----+--------+ | MemtableList +----------------> Empty
// | | Version r | +-----------+
// | +--------------+ | |
// +------------------+ current| Version B |
// +--------------+ | +----->+ |
// | | | | +-----+-----+
// Compaction +>+ SuperVersion +-------------+ ^
// Job | 2 +------+ | |current
// | +----+ | | mem | +------------+
// +--------------+ | | +---------------------> |
// | +------------------------> MemTable a |
// | mem | | |
// +--------------+ | | +------------+
// | +--------------------------+
// Iter1 +-----> SuperVersion | | +------------+
// | 1 +------------------------------>+ |
// | +-+ | mem | MemTable b |
// +--------------+ | | | |
// | | +--------------+ +-----^------+
// | |imm | MemtableList | |
// | +--->+ Version s +------------+
// | +--------------+
// | +--------------+
// | | MemtableList |
// +------>+ Version t +--------> Empty
// imm +--------------+
//
// In this example, even if the current LSM-tree consists of Version A and
// memtable a, which is also referenced by SuperVersion, two older SuperVersion
// SuperVersion2 and Superversion1 still exist, and are referenced by a
// compaction job and an old iterator Iter1, respectively. SuperVersion2
// contains Version B, memtable a and memtable b; SuperVersion1 contains
// Version B and memtable b (mutable). As a result, Version B and memtable b
// are prevented from being destroyed or deleted.
// ColumnFamilyHandleImpl is the class that clients use to access different // ColumnFamilyHandleImpl is the class that clients use to access different
// column families. It has non-trivial destructor, which gets called when client // column families. It has non-trivial destructor, which gets called when client