cubrid inside - architecture, source & management components

CUBRID InsideFirst Look Inside CUBRID

Jan 18, 2009

저작권 Copyright Notice

CREATIVE COMMONS 저작자표시 -동일조건변경허락 2.0 대한민국

이용자는 아래의 조건을 따르는 경우에 한하여 자유롭게

• 이 저작물을 복제 , 배포 , 전송 , 전시 , 공연 및 방송할 수 있습니다 .

• 이차적 저작물을 작성할 수 있습니다 .

다음과 같은 조건을 따라야 합니다 :

• 저작자표시 . 귀하는 원저작자를 표시하여야 합니다 .

• What does "Attribute this work" mean?

The page you came from contained embedded licensing metadata, including how the creator wishes to be attributed for re-use. You can use the

HTML here to cite the work. Doing so will also include metadata on your page so that others can find the original work as well.

• 동일조건변경허락 . 귀하가 이 저작물을 개작 , 변형 또는 가공했을 경우에는 , 이 저작물과 동일한 이용허락조건하에서만 배포할 수 있습니다 .

• 귀하는 , 이 저작물의 재이용이나 배포의 경우 , 이 저작물에 적용된 이용허락조건을 명확하게 나타내어야 합니다 .

• 저작권자로부터 별도의 허가를 받으면 이러한 조건들은 적용되지 않습니다 .

• Nothing in this license impairs or restricts the author's moral rights.

CREATIVE COMMONS Attribution-Share Alike 2.0 Korea

You are free:

• to Share — to copy, distribute and transmit the work

• to Remix — to adapt the work

Under the following conditions:

• Attribution. You must attribute the work in the manner specified by the author or licensor (but not in any way that suggests that they endorse you or your

use of the work).

• What does "Attribute this work" mean?

The page you came from contained embedded licensing metadata, including how the creator wishes to be attributed for re-use. You can use the HTML here

to cite the work. Doing so will also include metadata on your page so that others can find the original work as well.

• Share Alike. If you alter, transform, or build upon this work, you may distribute the resulting work only under the same or similar license to this one.

• For any reuse or distribution, you must make clear to others the license terms of this work. The best way to do this is with a link to this web page.

• Any of the above conditions can be waived if you get permission from the copyright holder.

• Nothing in this license impairs or restricts the author's moral rights.

http://creativecommons.org/licenses/by-sa/2.0/kr/

Copyright 2009 Search Solution Corporation. All Rights Reserved.

CUBRID

41 First Look Inside CUBRID

Contents

1. System Architecture and Processes

2. Design & Implementation Concepts

3. Module and Source Structure

4. In The Next CUBRID Inside

4 /


System Architecture and Pro-cesses• What is CUBRID?

• System Architecture

• CUBRID Processes and IPC

• Query Processing Workflow

1

5 /


System Architecture and ProcessesWhat is CUBRID?

□ CUBRID is

Open Source DBMS owned and developed by NHN

• Hosted at nForge site

Targeted to a database platform for Internet portal service

Relational database system with object extensions – Object-Relational Model

1,000,000 line of source code scaled project

• Engine – C/C++, CUBRID Manager (GUI) and JDBC driver – Java

□ CUBRID provides

Full-featured database server engine

• Server and Broker architecture

GUI management and query tool

• CUBRID Manager

Various API (JDBC, ODBC(OLE DB), PHP, and CCI)

• Python and Ruby as contributed projects

1.1

6 /


System Architecture and ProcessesSystem Architecture

1.2

□ Server Subsystem

Multi-threaded storage engine part which runs as DB server

Consists of Transaction Manager, Object Manager, Lock Manager, Query Manager,

Log Manager, and Storage Manager

□ Client Subsystem

Client library part providing native C API

Consists of Query Parser & Optimizer, Object & Lock Cache, Transaction Manager

and Object Manager

□ Broker Subsystem

Linked with client library and implemented CCI protocol

Does job queuing, connection pooling, monitoring and logging

□ Connectors

JDBC driver

CCI library

• Many drivers are implemented with CCI library – ODBC, OLEDB, PHP, Python, and Ruby

7 /


System Architecture and ProcessesSystem Architecture

1.2

AdminUtility

Server

Broker

Inter-face

ConnectionPooling

Monitoring Logging

JDBC

JobQueu-

ing

ODBC CCI

PHPOLE DB

Python

Ruby

Backup /Restore

CUBRIDManager

CUBRIDManager

GUI

CM Server

Create, Delete,

Copy, Re-name

Load /Unload

Compact /Optimize

Add Vol-ume

Check /Diag

QueryParser &

Opti-mizer

QueryManager

LockManager

LockCache

ObjectManager

Transac-tion

Manager

StorageManager

LogManager

DataVol-ume

IndexVol-ume

TempVol-ume

Ac-tiveLogAr-

chiveLog

File Based

Objects

8 /


AdminUtility

Server

Broker

Inter-face

Parser ObjectManager

SchemaManager

Transac-tion

ManagerQuery

Transform

QueryOptimizer

PlanGenera-

tion

Workspace Manager

Memory Manager

Native C API

Communication Module

ConnectionPooling

Monitoring Logging

JDBC

JobQueu-

ing

ODBC CCI

PHPOLE DB

Python

Ruby

Client Library


Transac-tion

Manager

LogManager

LockManager

QueryManager

AccessMethod

B+TreeModule

File Man-ager

SystemCatalog Module

Buffer Manager

Disk Manager

Storage Engine

DataVol-ume

IndexVol-ume

TempVol-ume

Ac-tiveLogAr-

chiveLog

Backup /Restore

CUBRIDManager

CUBRIDManager

GUI

CM Server

Create, Delete,

Copy, Re-name

Load /Unload

Compact /Optimize

Add Vol-ume

Check /Diag

File Based

Objects

System Architecture

9 /

1.2


System Architecture and ProcessesCUBRID Processes and IPC

□ Master Process ‘cub_master’

Daemon process listening on the

TCP port for clients’ connection

• csql, broker, and admin utilities

Registrar of server processes using

Unix Domain Socket

□ Server Process ‘cub_server’

Multi-threaded process serving

clients’ requests

• Has thread pool and job queue

Accesses database volume and log

files

• One server process per one data-

base

□ Broker Process ‘cub_broker’

Daemon process listening on the

TCP port for connectors’ connec-

tion

Controls cub_cas processes (fork/

kill) with connection queue

□ CAS Process ‘cub_cas’

Connects to the server process in

behalf of connectors

• According to the database and user

name

□ CSQL Program ‘csql’

Interactive SQL execution

1.3

10 /


cub_cascubridc-

s.so

cub_cascubridc-

s.socsql

cubridc-s.so

System Architecture and ProcessesCUBRID Processes and IPC

Replication

StandbyServer

Broker

APWeb Server

ActiveServer

1.3

volume filevolume file

log filelog file

cub_master

cub_servercubrid.so

cub_broker

JDBC driver

CCI library

mount(read/write)

descriptor pass

UDS

TCP

request &response

request &response

register

databases.txt

read

cubrid.conf

port listening job queuemulti-thread

connect

parse

parse

port listening

query &result

query &result

shared memory

descriptor pass

cubrid_broker.conf

parse

connect

fork

cub_admincubrid-sa.so

11 /


System Architecture and ProcessesQuery Processing Workflow

Query Pars-ing•Parse tree

Semantic Check•Attributed parse tree

View Trans-formation•Parse tree

Query Rewrite Optimization•Parse tree

Cost Based Optimization•Query graph

Plan Genera-tion•XASL

Scan Open•Heap/Index scan

B+tree Search•OID list

Object Fetch•Lock, Buffer page

Memory Rep-resentation•DB_VALUE

Predicate Eval-uation•Qualified tuple

Query Result•List(temp) file

Cursor Open•List file scan

Result Tuple Fetch•List file page

Cursor Close•Query end

1.4

12 /


System Architecture and ProcessesQuery Processing Workflow

1.4

Query ManagerTransaction

ManagerLock Manager

Query Result Cache

Query Plan Cache

Query Evalua-tor

(Interpreter)

Storage Man-ager

Scan Man-agement

Extended Access

Spec Lan-guage(XASL)

UNION

DIFFERENCE

INTERSECTION

OBJFETCH

SETFETCH

BUILD_LIST

BUILD_VALUE

SCAN

MERGE_LIST

READ

UPDATE

DELETE

INSERT

XASL Type

13 /


Design & Implementation Con-cepts• Client-Server and Standalone

Mode

• Object, Class and OID

• Representation, Record Descrip-

tor, and Memory Object Pointer

• Workspace – Object & Lock

Caching

• 2PL vs. MVCC

• XASL (Extended Access Spec

Language)

• Heap, B+tree, Catalog, and Temp

• Click Counter

2

14 /


Design & Implementation ConceptsClient-Server and Standalone Mode

Client-Server Mode• Database engine is

separated into server and client process

• CUBRID is a kind of Client-server DBMS architecture

• Server process (cub_server) mounts the data-base

• Client processes, which are linked with client library (libcubridcs.so), sends requests to the server process

• Broker (cub_cas) is a typical client process

• Applications (con-nectors) are com-municating with the broker.

Standalone Mode• Database engine is

combined into one process

• Standalone pro-grams, which are linked with stand-alone library (libcubridsa.so), mount the data-base without the server

• csql with –S and admin utilities are typical standalone process

• Note that the database can be mounted only one process, cub_server or standalone pro-gram

2.1

15 /


Design & Implementation ConceptsObject, Class and OID

Object

• Instances and classes: A relational database is a collection of tables, and a CUBRID data-base is a collection of classes.

• The data in a relational table is contained in rows that coincide with the instances of a class.

Class

• Objects that share the same set of attributes and methods are grouped into entities called classes.

• Class hierarchy is a single rooted and connected directed acyclic graph (DAG) of classes; superclass and subclass

• The root class, db_root, is a system defined class, which is the root of the class hierarchy

OID

• A permanent unchangeable object identifier is assigned to an object when the object is stored on a heap or when an OID is requested for an object.

• The OID uniquely identifies the object and directly indicates its physical disk address: (vol-ume identifier, page identifier, slot identifier)

• OIDs can be reclaimed only during database compaction by resolving references to the re-claimed OIDs

2.2

16 /


Design & Implementation Concepts Representation, Record Descriptor, and Memory Object Pointer

2.3

Disk Representation•Format of object(record) when it is stored on disk•A formatted and compacted se-quence of bytes optimized for query processing•Record descriptor: RECDES•Disk structure: DISK_REPR, DISK_ATTR

Memory Representation•Format of object when it is cached on memory•A hierarchy of compound structures optimized for program access•Memory Object Pointer: MOP•Memory structure: SM_CLASS, SM_ATTRIBUTE

Representation ID: REPR_ID•Version of class representation•Whenever a change is made to a class, new representation id is assigned instead of reflect-ing the change to all instances

Cache Coherency Num-ber: CHN•To synchronize in-mem-ory cached copy of the object with its corre-sponding disk object

17 /


Design & Implementation ConceptsWorkspace – Object & Lock Caching

Client memory cache of database object• B

asic concept of Client-server DBMS architecture

• Whenever the object is accessed, the object locator fetches the object from the server into workspace

• Cached objects may be decached when its state is not consistent with the one stored in the database

• Note that query results are not objects but tuples

Hash table for workspace objects with linked lists• M

aps OID to MOP via workspace hash table

• MOP is a handle to a cached object

• MOP data structure contains OID, object pointer, class pointer, cached lock mode, class link, dirty link, hash link, commit link, and state flags

Acquired lock cache• B

asic concept of Client-server DBMS architecture

• Whenever an object is fetched, necessary lock is acquired before and cached with the object (lock mode)

• Note that lock manager is centralized in the server

Memory manager for object cache• Q

uick fit allocation

• Works with garbage collector

• Memory object contents are allocated within workspace memory

Workspace

2.4

18 /


Design & Implementation Concepts2PL vs. MVCC

Concurrency Control in DBMS

2 Phase Locking•CUBRID, SQL Server, DB2•A concurrency control locking protocol to guarantee serializability•Transaction handles its locks in two distinct, consecutive phases•Phase 1: locks are acquired and no locks are released•Phase 2: locks are released and no locks are acquired•2PL is subject to deadlocks and has deadlock resolver

Multi Version Concurrency Control•Oracle, InnoDB, PostgreSQL, AltiBase•Uses timestamps or sequence number to achieve serializability•Each version would have a write timestamp or sequence number•Read transaction does not block by maintaining several versions of an object•Drawback is the cost of maintaining multiple versions

2.5

19 /


Design & Implementation ConceptsXASL (Extended Access Spec Language)

• Run time structure that represent executable queries

• The XASL interpreter (query evaluator) uses the XASL NODEs to determine which query actions are to be performed for the query

• UNION_PROC, DIF-FERENCE_PROC, IN-TERSECTION_PROC, BUILDLIST_PROC, BUILDVALUE_PROC

• OBJFETCH_PROC, SETFETCH_PROC, SCAN_PROC, MERGE-LIST_PROC, READ_PROC

• UPDATE_PROC, DELETE_PROC, IN-SERT_PROC

• Scan info: defines the access method for the data

• Value lists: determine which values are needed in query re-sults

• Predicates: used to restrict query results

• Pointers to other XASL NODEs: sub-query, join, union, etc

• Heap scan: sequen-tial scan or full scan

• Index scan: B+tree search

• List file scan: tempo-ral and final query result

• Set scan: set data• Method scan: call-

back to the client for method execution

XASL Tree XASL Type

XASL NODESCAN Type

2.6

20 /


Design & Implementation ConceptsHeap, B+tree, Catalog, and Temp

File

Heap• Object storage

in data vol-umes

• One per class

B+tree• Index storage

in index vol-umes

• One per index

Catalog (Ex-tendible hash)• Class represen-

tation• One per system

Temp• Temporal or fi-

nal query result• In temp vol-

umes

2.7

21 /


Design & Implementation ConceptsClick Counter

□ In Terms of User

Whenever the user clicks an article, the read counter of the article is increased.

• Counts the clicks on the article

□ In Terms of Service Application Developer

The read counter is very important for bulletin board type service.

Combine getting article data and increasing read counter together

• get_article() + increase_counter() = get_article_and_increase_counter()

• (SELECT article_info) + (UPDATE read_counter+1)

□ In Terms of Database Engineer

Do SELECT and UPDATE in one SQL statement

• SELECT trigger? SELECT FOR UDPATE?

UPDATE the record with the short term lock

□ Example

SELECT doc_id, title, INCR(read_counter), post_date FROM board WHERE

doc_id=?

2.8

22 /


Design & Implementation ConceptsClick Counter

2.8

Parsing

Plan Generation

Access Spec

Scan Open

B+tree Search

Address of Record

Object Locking

Object Representation

Copy to Result Set

Return

Fetch

Parsing

Plan Generation

Access Spec

Scan Open

B+tree Search

Address of Record

Object Locking


Change Values

Logging

Return

Parsing

Plan Generation

Access Spec

Scan Open

B+tree Search

Address of Record

Object Locking


Copy to Result Set

Change Values

Logging

Return

Fetch

SELECT UPDATE Click Counter

Don’t walk the same passage twice!When you get there, do as many things as possible.

Killing two birds with one stone.

23 /


Module and Source Structure• Overall Structure

• Query Processing Component

• Transaction Management Com-

ponent

• Server Storage Management

Component

• Client Storage Management

Component

• Object Management Component

• Job and Thread Management

Component

• Client-Server Communication

Layer

• Broker Component

3

24 /


Module and Source StructureOverall Structure

src

api

base

broker

cci

cmserver

communication

compat

connection

executables

heaplayersjdbcjsp

object

odbc

oledb

optimizer

parser

php

query

storage

thread

transaction

25 /

3.1

Object Management

Query Processing

CUBRID Manager

Query Processing

Query Processing

Broker

Client-Server Comm

Job & Thread Management

Transaction Management

Server Storage Management

Client Storage Management

Memory Manager

Native C API

CCI Library

Job & Thread Management

Java Stored Procedure

PHP Connector

JDBC Driver

C APICommon & Support

Utilities

41 First Look Inside CUBRID26 /

3.1

Parser ObjectManager

SchemaManager

Transac-tion

ManagerQuery

Transform

QueryOptimizer

PlanGeneration

Workspace Manager

Memory Manager

Native C API


ConnectionPooling

Monitoring Logging

JDBC

JobQueuing

ODBC CCI

PHPOLE DB

Python

Ruby

Client Library


Transac-tion

Manager

LogManager

LockManager

QueryManager

AccessMethod

B+TreeModule

File Man-ager

SystemCatalog Module

Buffer Manager

Disk Manager

Storage Engine

Backup /Restore

CUBRIDManager

GUI

CM Server

Create, Delete,

Copy, Re-name

Load /Unload

Compact /Optimize

Add Volume

Check /Diag

Module and Source Structure

src/query

src/storage

src/transaction

src/compat

src/broker

src/cmserver

src/communication

src/jdbc

src/object

src/oledb

src/php

src/cci

src/odbc

src/optimizer

src/base

src/executables

src/heaplayers

src/parser

src/thread

src/connection


Module and Source StructureQuery Processing Component

□ Scanner/Parser

Parses query string and builds parse tree

Uses ANTLR v1.x as parser generator

• Will be replaced with ‘bison’ at R2.0

Source files in src/parser

• Data structure: PT_PARSER, PT_NODE, …

□ Semantic Checker

Type check and binding, name resolution,

semantic checking, view translation, and so

on


• Parse tree traversing function:

parser_walk_tree()

□ Optimizer

Rewrite optimization by heuristic rules

Select a query execution plan by CBO (Cost

Based Optimization) method

Source files in src/parser and src/optimizer

• Data structure: QO_NODE, QO_PLANNER, …

□ XASL Generator

Generate XASL tree from the query execution

plan

XASL contains scan information (heap/index/

list file/set/method scan), value list, and predi-

cates


• Data structure: XASL_NODE, REGU_VARIABLE, …

□ Query Manager (Executor)

Executes XASL tree (XASL interpreter)

Includes query evaluator, query plan cache,

query result cache, query result file, and scan

management

Source files in src/query

• Main function: qexec_execute_mainblock()

□ Cursor Manager

Extracts tuple from the list file page shipped

from the server Query Manager

Source files in src/query

3.2

27 /


Server

Broker

Client-Server Communication Layer


Module and Source StructureQuery Processing Component

3.2

Query Manager

Storage Manage-ment Component

(Server)

Semantic Checker Scanner/Parser Optimizer

XASL Generator

Object Manage-ment Component

Cursor Manager

Transaction Man-agement Compo-

nent

Query Plan/Result Cache

Query Evaluator(Interpreter)

Scan Management

28 /


Module and Source StructureTransaction Management Component

□ Transaction Manager

Controls transaction – start, commit/abort,

savepoint, and top operation

• Works tightly with lock and log manager

Supports 2PC protocol for global transaction

Initialize other related modules such as lock,

log, and recovery manager

Source files in src/transaction

• Data structure: LOG_TDES, TRANTABLE, …

□ Object Locator

Controls the flow of objects between the

workspaces and the database pages

• Translates between memory representation and

disk representation

Caches objects from the server into the

workspace with the granted locks

• Cached object and lock provides fast access to

the objects and reduces interacting with the lock

manager on the server


□ Lock Manager

Based on strict 2PL protocol

Includes support for OR feature

• class lock and instance lock

Deadlock detector with WFG

• As separate thread


• Data structure: LK_ENTRY, LK_TRAN_LOCK, …

□ Log Manager

Transaction logging and recovering

• REDO/UNDO recovery protocol

• WAL protocol

• Physical logging and logical logging

Supports group commit and asynchronous

commit feature

Source files in src/transaciton

• Data structure: LOG_GLOBAL,

LOG_HDRPAGE, LOG_PAGE,

3.3

29 /


Module and Source StructureTransaction Management Component

3.3

Server

Broker


Object Locator(Server)


(Server)

Lock Manager



(Client)

Object Management Component

Transaction Man-ager

(Client)

Transaction Man-ager

(Server)

Log Manager

Object Locator(Client)

30 /


Module and Source StructureServer Storage Management Component

□ File I/O Manager

Provides disk I/O access functions and implements file

structure

Coordinates the allocation and deallocation of un-

structured files and pages

• File>Sector>Page

• File manager requests sectors from the disk manager

Source files in src/storage

• Identifier: vdes, PAGEID

□ Disk Manager

Volume file management – implements volume struc-

ture in a OS file

• A volume corresponds to a singe file

• A set of contiguous pages of fixed size


• Identifier: volid

□ Page Buffer Manager

Data buffer pool with LRU replacement algorithm

Uses buffer hash table and fix/unfix protocol (latch)


• Identifier: VPID

• Data structure: PGBUF_BUFFER_POOL

□ Slotted Page Manager

Manages insertions, deletions, and modifications of

records on pages

• Each record has an associated slot identifier


• OID = (pageid, slotid, volid)

□ Overflow File Manager

Larger object than the size of a page is placed on

overflow pages


□ Object Heap Manager

Inserts/deletes/updates objects within a file

• Permanent unchangeable OID is assigned to an object

when it is stored on a heap

• Provides heap scan method (sequential scan) and direct

fetch method

An object heap holds the instances of one class


• Identifier: HFID

□ B+-tree Manager

Implements prefix B+-tree index with KVL

• Provides index scan method – unique find and range

search

3.4

31 /


Module and Source StructureServer Storage Management Component

3.4

Server

Broker



I/O Manager

Page Buffer Man-ager

File Manager

Catalog Manager

Object Heap Manager

B+-tree ManagerLarge Object

ManagerExtendible Hash

Manager

Query Processing Compo-nent

(Server)

Transaction Management Component

(Server)

Disk Manager

Slotted Page Manager

Overflow File Manager

32 /


Module and Source StructureClient Storage Management Component

□ Workspace Manager

Maintains a cache of database objects

within the virtual memory address

space of client process

• Maintains hash table that maps a OID

into MOP

• MOP has OID field and a pointer to

memory object

Source files in src/object

□ Quick Fit Allocator

Memory manager for workspace

Source files in src/object, src/base

□ Garbage Collector

To collect and free the

unused(dangled) memory objects

Normally GC is disabled

□ Object and Lock Caching

When the contents of object(s) is(are) re-

quired, the object locator is called to fetch

the object(s) from the database via the

server-side object locator

The object locator translate the disk repre-

sentation of the objet into the memory rep-

resentation and allocates MOP to points it

While accessing the object in the database,

the requested lock is held by the lock man-

ager in the server and the memory object

has the cached lock mode

The cached objects can be decached at any

time when its state is not consistent with the

disk objects, or will be decached at the time

of transaction abort

• At the time of transaction commit, only the

cached locks are released; objects remains

with null lock

3.5

33 /


Module and Source StructureClient Storage Management Component

3.5

Server

Broker



Garbage Collector Quick File AllocatorWorkspace Manager

Object Management Component

Transaction Man-agement Compo-

nent Object Locator

(Client)

34 /


Module and Source StructureObject Management Component

□ Object Accessor

Interface for object creation, deletion, inspec-

tion, and modification


□ Object Representation

Transforms (or translates) between the disk

representation and the memory representation

of the object

• Resolves byte ordering also

Source files in src/object, src/base

• Functions: or_xxx(), tf_mem_to_disk(),

tf_disk_to_mem(), …

□ Data Type and Domain

Internal data structure for data type and do-

main


• Data structure: TP_DOMAIN

□ Sets


• Data structure: SETOBJ, COL

□ Schema Manager

Implements OR model regarding database schema

• Definitions of the class object – attributes, method,

class inheritance, and conflict resolution

• Memory representation of class object

Source files in src/objet

• Data structure: SM_CLASS, SM_ATTRIBUTE, …

□ Authentication and Authorization

User authentication and database privilege

• User, GRANT, REVOKE, …


□ Trigger Manager

Implements trigger feature

Source files src/object

□ Dynamic Loader

Support dynamic loading of shared object as

‘method’

• Depreciated feature

□ Large Object

Implements glo, elo, fbo classes and supporting

methods


3.6

35 /


Module and Source StructureObject Management Component

3.6

Server

Broker



Set

Schema ManagerAuthentication

& AuthorizationTrigger Manager

Dynamic Loader

Object Representa-tion


(Client)

Data Type and Do-main

Object Accessor

36 /


Module and Source StructureJob and Thread Management Component

□ Connection Manager

Connection establishment and man-

agement and request handler

Source files in src/communication, src/

connection

• Data structure: CSS_QUEUE_ENTRY,

CSS_CONN_ENTRY, …

□ Job Queue

Master thread monitors socket fds (se-

lect()), receives clients’ request packet,

en-queues it, and then wake up a worker

thread in the thread pool

Source files in src/connection

• Data structure: CSS_JOB_ENTRY, …

□ Critical Section

Implements semaphore

Source files in src/thread

□ System Threads

Master thread

Deadlock detection thread

• Run deadlock detection algorithm periodically

• Resolves deadlock by aborting unilaterally vic-

tim transaction(s)

Checkpoint thread

• Executes periodic checkpoint action

OOB handling thread

• Catch OOB signal

Page flush thread

• Flushes dirty data buffer pages to the volume

files in the background

• To reduce burden of page replacement

Log flush thread

• Flushes dirty log buffer pages to the log file in

the background

• Implements group commit and asynchronous

commit features

Source files in src/thread

3.7

37 /


Module and Source StructureClient-Server Communication Layer

□ Socket Transceiver

Socket utility functions

• TCP/IP, Unix domain socket, and

named pipe


• Linux and Windows port

□ Packet Handler

Sends requests and receives re-

sponse between client and server

• Packet types: request packet, data

packet, close packet, out-of-band

packet, error packet, abort packet

• Packet format: packet header + pay-

load

Queuing request and data packets


□ Communication Interface

Layer

RPC stub style functions

• Implements object serialization

Source files in src/communication

3.8

38 /


Module and Source StructureBroker Component

□ Broker Process

Accepts connection requests from

the Connectors (JDBC, CCI, …)

Allocates and/or assigns CAS

process to the request

Controls and monitors CAS pro-

cesses

Has four threads

• main: manages cas process

• receiver_thread: does accept() and

enqueues new job

• dispatch_thread: dequeue job and

assigns cas process

• cas_monitor_thread: monitors cas

process to restart

Source files in src/broker

□ CAS Process

Linked with CUBRID client library

• So, run as a single thread

Receives and processes requests from

the application(Connectors)

• Connect to database

• Prepare and execute query

• Fetch the result

• Commit/rollback transaction

• Get/set parameters

Has four states

• IDLE: not connected with AP

• BUSY: processing request

• CLIENT_WAIT: connected and waiting for

request; within a transaction

• CLOSE_WAIT: connected and waiting for

request; out of a transaction

Source files in src/broker

3.9

39 /


In The Next CUBRID Inside4

40 /


In The Next CUBRID Inside

Academic theory?

Implemen-tation de-

tails?

New fea-tures?

OrFuture of

the project?

Processing flow?

Practical is-sues?

Details of Transaction Manage-ment

Communication between Broker and Connectors

Replication and/or HA feature

4

What topic do you want to see covered in the next CUBRID Inside?

Suggesting Topics

41 /

Question?

cubrid inside - architecture, source & management components

Technology