在用OCCI 向Oracle中插入數(shù)據(jù)時(shí),效率不高,使用自動(dòng)提交數(shù)據(jù)的情況(默認(rèn))下一秒鐘只能插入1000條數(shù)據(jù)左右。因?yàn)椴迦霐?shù)據(jù)庫這塊是影響系統(tǒng)中效率最明顯的地方,因此很有必要提高數(shù)據(jù)插入這塊的效率。在網(wǎng)上找了一些資料后發(fā)現(xiàn),可以重用statement對象,使用批量插入的方法,先把數(shù)據(jù)保存在內(nèi)存中,積累到一定數(shù)值之后批量插入Oracle,這樣平均下來一秒鐘可以插入5000多條數(shù)據(jù),性能有所提高,不錯(cuò),這里做一個(gè)記錄。
- #include <iostream>
- #include <string.h>
- #include <time.h>
- #include <sys/time.h>
-
- #define WIN32COMMON //避免函數(shù)重定義錯(cuò)誤
- #include <occi.h>
- #include <cstdlib>
- #include <map>
- #define ArraySize 10000 //內(nèi)存中數(shù)據(jù)滿1000條批量插入到oracle中
-
- using namespace oracle::occi;
- using namespace std;
-
- /*
- *返回當(dāng)前時(shí)間,用于計(jì)算兩個(gè)操作的時(shí)間差
- */
- long getCurrentTime()
- {
- struct timeval tv;
- gettimeofday(&tv,NULL);
- return tv.tv_sec * 1000 + tv.tv_usec / 1000;
- }
-
- /*
- * @author: roger
- */
- int main(void)
- {
-
- string username = "XX";
- string pass = "XXX";
- string srvName = "XXX";
- Environment *env ;
- Connection *conn;
- Statement *stmt;
- try
- {
- env = Environment::createEnvironment(Environment::THREADED_MUTEXED);
- conn = env->createConnection(username, pass, srvName);
-
- string sql = "insert into instant_infor (motor_id, lat, lon, uploadTime, receivetime, state_id, sys_state_id) values(:fld1,:fld2,:fld3,to_timestamp(:fld4,'yyyy-mm-dd hh24:mi:ss'),to_timestamp(:fld5,'yyyy-mm-dd hh24:mi:ss'),:fld6,:fld7)";
- stmt = conn->createStatement(sql);
-
- } catch(SQLException e)
- {
- env = NULL;
- conn = NULL;
- cout<<e.what()<<endl;
- }
-
- char motorid[ArraySize][12];
- char lat[ArraySize][20];
- char lon[ArraySize][20];
- char uploadTime[ArraySize][20];
- char createTime[ArraySize][20];
- char state_id[ArraySize][50];
- char sys_state_id[ArraySize][50];
-
- ub2 motor_idLen[ArraySize] ;
- ub2 uploadTimeLen[ArraySize] ;
- ub2 createTimeLen[ArraySize];
- ub2 state_idLen[ArraySize];
- ub2 sys_state_idLen[ArraySize];
- ub2 latLen[ArraySize] ;
- ub2 lonLen[ArraySize] ;
-
- long a1 = getCurrentTime();
- for(int i=0;i<ArraySize;i++){
-
- strcpy(motorid[i],"10000100000");
- strcpy(lat[i] , "30.123");
- strcpy(lon[i] , "120.123");
- strcpy(uploadTime[i] , "2015-11-11 11:11:11");
- strcpy(createTime[i] , "2015-11-11 11:11:11");
- strcpy(state_id[i] ,"1");
- strcpy(sys_state_id[i],"1");
-
- motor_idLen[i] = strlen( motorid[i] ) + 1;
- uploadTimeLen[i] = strlen( uploadTime[i] ) + 1;
- createTimeLen[i] = strlen( createTime[i] ) + 1;
- state_idLen[i] = strlen( state_id[i] ) + 1;
- sys_state_idLen[i] = strlen( sys_state_id[i] ) + 1;
- latLen[i] = strlen( lat[i] ) + 1;
- lonLen[i] = strlen( lon[i] ) + 1;
- }
-
- stmt->setDataBuffer(1, (dvoid*)motorid, OCCI_SQLT_STR,sizeof( motorid[0] ), motor_idLen);
- stmt->setDataBuffer(2, (dvoid*)lat, OCCI_SQLT_STR, sizeof( lat[0] ), latLen);
- stmt->setDataBuffer(3, (dvoid*)lon, OCCI_SQLT_STR, sizeof( lon[0] ), lonLen);
- stmt->setDataBuffer(4, (dvoid*)uploadTime, OCCI_SQLT_STR, sizeof( uploadTime[0] ), uploadTimeLen);
- stmt->setDataBuffer(5, (dvoid*)createTime, OCCI_SQLT_STR, sizeof( createTime[0] ), createTimeLen);
- stmt->setDataBuffer(6, (dvoid*)state_id, OCCI_SQLT_STR,sizeof( state_id[0] ), state_idLen);
- stmt->setDataBuffer(7, (dvoid*)sys_state_id, OCCI_SQLT_STR, sizeof( sys_state_id[0] ), sys_state_idLen);
-
- stmt->executeArrayUpdate(ArraySize);
- conn->terminateStatement(stmt);
- conn->commit();
-
- long a2= getCurrentTime();
-
- cout<<"插入"<<ArraySize<<"條數(shù)據(jù)完成"<<endl;
- cout<<"花費(fèi)時(shí)間: "<<(a2-a1)<<endl;
-
- }
測試中是訪問本地的
Oracle數(shù)據(jù)庫,用到的
Makefile文件如下:
- CC=g++
- OBJS=TestOracle.o
- LIB=-L/opt/oracle/oracle11g/product/11.2.0/dbhome_1/lib -L/opt/oracle/oracle11g/product/11.2.0/dbhome_1/rdbms/lib/
- INCLUDE=-I/opt/oracle/oracle11g/product/11.2.0/dbhome_1/precomp/public -I/opt/oracle/oracle11g/product/11.2.0/dbhome_1/rdbms/public
- Test: $(OBJS)
- $(CC) -o Test $(OBJS) $(LIB) -locci -lclntsh
- TestOracle.o: TestOracle.cpp
- $(CC) -c TestOracle.cpp $(INCLUDE)
- clean:
- rm -rf *.o & rm Test
程序的編寫是參考網(wǎng)上的一篇文章,寫的很好,這里作為參考:
http://rgyq.blog.163.com/blog/static/3161253820131695957501/每次創(chuàng)建statement對象時(shí),需要在客戶端和服務(wù)端分配資源,如內(nèi)存和游標(biāo)(cursor),用于存儲(chǔ)對象及數(shù)據(jù)。為了不必要的內(nèi)存重分配,應(yīng)重用statement對象。statement對象創(chuàng)建后,可以使用setSQL方法進(jìn)行重用,例如:
- Connection* conn = env->createConnection();
- Statement* stmt = conn->createStatement();
- stmt->setSQL(“INSERT INTO fruit_basket_tab VALUES(‘Apples’, 3)”);
- stmt->executeUpdate();
- stmt->setSQL(“INSERT INTO fruit_basket_tab VALUES(‘Oranges’, 4)”);
- stmt->executeUpdate();
- stmt->setSQL(“INSERT INTO fruit_basket_tab VALUES(‘Bananas’, 1)”);
- stmt->executeUpdate();'
- stmt->setSQL(“SELECT * FROM fruit_basket_tab WHERE quantity > 2”);
- ResultSet* rs = stmt->executeQuery();
為了進(jìn)一步控制內(nèi)存重新分配,可以通過參數(shù)化將前面3條
SQL語句變成1條,然后設(shè)置參數(shù),再執(zhí)行。注意輸入?yún)?shù)的類型變化,因?yàn)?,每次改變參?shù)類型都會(huì)觸發(fā)重綁定。參數(shù)化示例如下:
- stmt->setSQL(“INSERT INTO fruit_basket_tab VALUES(:1, :2)”);
- stmt->setString( 1, “Apples” );
- stmt->setInt( 2, 3 );
- stmt->executeUpdate();
- stmt->setString( 1, “Oranges” );
- stmt->setInt( 2, 4 );
- stmt->executeUpdate();
- stmt->setString( 1, “Bananas” );
- stmt->setInt( 2, 1 );
- stmt->executeUpdate();
對于那些經(jīng)常發(fā)生的操作,很多時(shí)間都浪費(fèi)在與服務(wù)器網(wǎng)絡(luò)通信中。OCCI提供了有效的機(jī)制用于在單次網(wǎng)絡(luò)通信中發(fā)送多行信息。
該優(yōu)化可用于INSERTs,UPDATEs和DELETEs。首先,設(shè)置最大
迭代次數(shù),然后設(shè)置可變長度參數(shù)的最大長度。在迭代過程中參數(shù)類型不可變。具體細(xì)節(jié)參考OCCI Programmers Guide,第二章。下例是上面的INSERTs的優(yōu)化:
- //prepare the batching process
- stmt->setMaxIterations( 3 );
- stmt->setMaxParamSize( 1, 8 ); //”Bananas” is longest param
- //batch the statements
- stmt->setSQL(“INSERT INTO fruit_basket_tab VALUES(:1, :2)”);
- stmt->setString( 1, “Apples” );
- stmt->setInt( 2, 3 );
- stmt->addIteration();
- stmt->setString( 1, “Oranges” );
- stmt->setInt( 2, 4 );
- stmt->addIteration();
- stmt->setString( 1, “Bananas” );
- stmt->setInt( 2, 1 );
- //execute the statements
- stmt->executeUpdate();
- Statement::setDataBuffer方法
綁定值到參數(shù)化statements的參數(shù)時(shí)需要內(nèi)存拷貝,因?yàn)闉榱吮苊庑畔⒃谥虚g執(zhí)行過程中被覆蓋,所以必須拷貝到內(nèi)部的緩沖區(qū)中??截惖拇鷥r(jià)對于大字符串尤其明顯,內(nèi)存的消耗以及拷貝所花的時(shí)間。如果應(yīng)用可以自己管理內(nèi)存,就可以通過OCCI提供的方法最小化上述開銷。
雖然許多OCI開發(fā)者使用OCCI簡明的創(chuàng)建environments和statement對象,但仍然使用許多OCI中的類型。
setDataBuffer方法允許OCI開發(fā)者執(zhí)行數(shù)組更新,最小化網(wǎng)絡(luò)通信次數(shù)。setDataBuffer方法與setXXX方法工作方法不同。一般說來,setXXX方法會(huì)將傳過來的數(shù)據(jù)拷貝到內(nèi)部緩沖區(qū)中,只要setXXX返回后參數(shù)值就可以被改變。然而,使用setDataBuffer方法可以避免將數(shù)據(jù)拷貝到內(nèi)部緩沖區(qū)中。代價(jià)是應(yīng)用程序在執(zhí)行完statement之前不可以修改緩沖區(qū)。例如:
- // insert Bananas
- char buf[BUF_SIZE] = "Bananas";
- int quantity = 1;
- ub2 buflen = strlen( buf ) + 1;
- ub2 quantlen = sizeof(int);
- stmt->setDataBuffer(1, (dvoid*)buf, OCCI_SQLT_STR, buflen, &buflen);
- stmt->setDataBuffer(2, (dvoid*)&quantity, OCCIINT, quantlen,
- &quantlen);
- stmt->executeUpdate(); // executeArrayUpdate(1) also would work.
- // insert Apples
- strcpy( buf, “Apples” );
- quantity = 3;
- buflen = strlen( buf ) + 1;
- quantlen = sizeof( int );
- stmt->setDataBuffer(1, (dvoid*)buf, OCCI_SQLT_STR, buflen, &buflen);
- stmt->setDataBuffer(2, (dvoid*)&quantity, OCCIINT, quantlen,
- &quantlen);
- stmt->executeUpdate(); // executeArrayUpdate(1) also would work.
- //commit the transaction
- conn->commit();
setDataBuffer方法可以與迭代執(zhí)行(iterative executes)和executeArrayUpdate方法結(jié)合使用。
當(dāng)進(jìn)行大量INSERTs和UPDATEs操作時(shí),可以通過executeArrayUpdate方法和setDataBuffer方法批量處理。這可以節(jié)省網(wǎng)絡(luò)通信,提高吞吐量。示例如下:
- char fruit[][BUF_SIZE] = { "Apples","Oranges","Bananas","Grapes" };
- int int_arr[]={ 3,4,1,5 };
- ub2 fruitlen[4]; // array of size of individual elements
- ub2 intsize[4];
- for(int i=0 ; i<4 ; i++)
- {
- intsize[i] = sizeof(int);
- fruitlen[i] = strlen( fruit[i] ) + 1 ; // include the null
- }
- stmt->setDataBuffer(1, (dvoid*)fruit, OCCI_SQLT_STR, BUF_SIZE,
- fruitlen);
- stmt->setDataBuffer(2, (dvoid*)int_arr, OCCIINT, sizeof(int), intsize);
- stmt->executeArrayUpdate(4);
- conn->commit();
executeArrayUpdate方法不會(huì)執(zhí)行,直到所有緩沖區(qū)均通過setDataBuffer方法設(shè)置。如果有參數(shù)需要調(diào)用setXXX方法賦值,可以調(diào)用setMaxIterations和setMaxParamSize方法,以及addIteration方法。具體如下:
- char fruits[][BUF_SIZE] = {“Apples”, “Oranges”, “Bananas”};
- ub2 fruitLen[3];
- for( int j=0; j<3; j++ )
- {
- fruitLen[j] = strlen( fruits[j] ) + 1; //include the null
- }
- stmt->setMaxIterations(3);
- //setDataBuffer only needs to be executed once
- //while all the other variables need to be set for each iteration
- stmt->setDataBuffer( 1, fruits, OCCI_SQLT_STR, sizeof(fruits[0]),
- fruitLen );
- stmt->setInt(2, 3); //Apple’s quantity
- stmt->addIteration();
- stmt->setInt(2, 4); //Orange’s quantity
- stmt->addIteration();
- stmt->setInt(2, 1); //Banana’s quantity
- //execute the iterative update
- stmt->executeUpdate(3);
對操作的列使用合適的setXXX和getXXX方法,而非統(tǒng)一作為string處理,可以省去不必要的轉(zhuǎn)換。
在NLS_LANG環(huán)境設(shè)置中使用合適的字符集,以避免獲取字符串時(shí)不必要的字符集轉(zhuǎn)換。
自動(dòng)提交模式
由于所有的SQL DML都是在事務(wù)中執(zhí)行,所以需要確認(rèn)所有的DML??梢愿鶕?jù)具體情況使用“Connection::commit”和“Connection::rollback”方法?!癝tatement::setAutoCommit”方法可以用來確認(rèn)其后的每條語句。使用該方法可節(jié)省網(wǎng)絡(luò)傳輸時(shí)間。
- //code with AutoCommit
- //transaction 1
- stmt->executeUpdate(“INSERT INTO fruit_basket_tab VALUES(“Apples”,3));
- stmt->executeUpdate(“INSERT INTO fruit_basket_tab VALUES(“Oranges”,4));
- stmt->setAutoCommit( TRUE );
- stmt->executeUpdate(“INSERT INTO fruit_basket_tab VALUES(“Bananas”,1));
- stmt->setAutoCommit( FALSE );
- //transaction 2
- stmt->executeUpdate(“INSERT INTO fruit_basket_tab VALUES(“Apples”,5));
- stmt->executeUpdate(“INSERT INTO fruit_basket_tab VALUES(“Oranges”,6));
- stmt->setAutoCommit( TRUE );
- stmt->executeUpdate(“INSERT INTO fruit_basket_tab VALUES(“Bananas”,2));
- stmt->setAutoCommit( FALSE );
這與下面的語句是等價(jià)的,但是2次網(wǎng)絡(luò)傳輸,每個(gè)事務(wù)1次
- //code without AutoCommit
- //transaction 1
- stmt->executeUpdate(“INSERT INTO fruit_basket_tab VALUES(“Apples”,3));
- stmt->executeUpdate(“INSERT INTO fruit_basket_tab VALUES(“Oranges”,4));
- stmt->executeUpdate(“INSERT INTO fruit_basket_tab VALUES(“Bananas”,1));
- conn->commit();
- //transaction 2
- stmt->executeUpdate(“INSERT INTO fruit_basket_tab VALUES(“Apples”,5));
- stmt->executeUpdate(“INSERT INTO fruit_basket_tab VALUES(“Oranges”,6));
- stmt->executeUpdate(“INSERT INTO fruit_basket_tab VALUES(“Bananas”,2));
- conn->commit();
建議AutoCommit只在每個(gè)事務(wù)的最后一條SQL語句前面使用。
結(jié)果集作為請求的響應(yīng)返回??梢允褂胣ext和status方法處理結(jié)果集:
- ResultSet* rs = stmt->executeQuery( “SELECT * FROM fruit_basket_tab” );
- ResultSet::Status stat = rs->status(); //status is DATA_AVAILABLE
- while( rs->next() ) { //process data }
- setPrefetchRowCount and setPrefetchMemorySize
雖然rs->next()每次只能返回一行,但是可以在一次網(wǎng)絡(luò)中預(yù)取多行放到客戶端的緩存中。使用類Statement的setPrefetchRowCount方法和setPrefetchMemorySize方法,每次可以取得不止一行。上例優(yōu)化后如下:
stmt->setPrefetchRowCount( 3 );
ResultSet* rs = stmt->executeQuery( “SELECT * FROM fruit_basket_tab” );
while ( rs->next() ) { //process data }
使用上述代碼,在一次網(wǎng)絡(luò)通信中就能取得3條記錄。默認(rèn)情形中,預(yù)取功能是啟用的,每次多取一條。要想關(guān)閉預(yù)取功能,必須同時(shí)調(diào)用方法setPrefetchRowCount和setPrefetchMemorySize,參數(shù)為0。如果兩個(gè)setPrefetchXXX方法都被調(diào)用了,那么實(shí)際預(yù)取的數(shù)目是這兩個(gè)方法參數(shù)中較小的那個(gè)。
setMaxColumnSize
當(dāng)取得的結(jié)果中有些列較大時(shí),可以使用ResultSet::setMaxColumnSize方法限制從指定的列可以獲取多少數(shù)據(jù)。當(dāng)只對部分?jǐn)?shù)據(jù)感興趣或者緩存大小有限時(shí),這就有用了。
ResultSet *rs = stmt->executeQuery( “SELECT description FROM
fruit_basket_tab” );
//want only first 80 characters from the description column
rs->setMaxColumnSize( 1, 80 );
當(dāng)對結(jié)果集的處理結(jié)束后,調(diào)用Statement::closeResultSet方法手動(dòng)強(qiáng)制關(guān)閉,這樣數(shù)據(jù)庫和OCCI的使用的資源就不必等待自動(dòng)釋放。
最后,為了確保沒有內(nèi)存泄漏,并關(guān)閉相關(guān)的服務(wù)器端的游標(biāo),需要釋放所有的statement對象。
conn->terminateStatement( stmt );
參考資料
Using OCCI: Best Practices
http://www.stanford.edu/dept/itss/docs/oracle/10gR2/appdev.102/b14294/performance.htm