码迷,mamicode.com
首页 > 数据库 > 详细

配置p6spyLog输出sql完整日志

时间:2019-01-04 18:01:14      阅读:279      评论:0      收藏:0      [点我收藏+]

标签:option   memory   unix   http   格式   done   sid   sla   lin   

 
第一步:
 
配置maven
<dependency>
    <groupid>p6spy</groupid>
    <artifactid>p6spy</artifactid>
    <version>3.6.0</version>
</dependency>
 
<dependency>
    <groupid>com.alibaba</groupid>
    <artifactid>druid</artifactid>
    <version>1.1.9</version>
</dependency>
 
第二步配置数据源:
 
application.properties配置
 
#jdbc.driver=oracle.jdbc.driver.OracleDriver
#jdbc.driver=com.mysql.jdbc.Driver
jdbc.driver=com.p6spy.engine.spy.P6SpyDriver
#jdbc.url=jdbc:oracle:thin:@192.168.10.10:1521:orcl
#jdbc.url=jdbc:mysql://localhost:3306/dbname
jdbc.url=jdbc:p6spy:mysql://127.0.0.1:3306/mybatis?useUnicode=true&characterEncoding=UTF-8&useSSL=false
jdbc.username=root
jdbc.password=root
(ps:如果是oracle数据库将mysql换成oracle即可)
 
applicationContext.xml配置
 
<bean id="dataSourceTarget" class="com.mchange.v2.c3p0.ComboPooledDataSource">
  <property name="driverClass" value="${jdbc.driver}" />
  <property name="jdbcUrl" value="${jdbc.url}" />
  <property name="user" value="${jdbc.username}" />
  <property name="password" value="${jdbc.password}" />
  <property name="initialPoolSize" value="${connection_pools.initial_pool_size}" />
  <property name="minPoolSize" value="${connection_pools.min_pool_size}" />
  <property name="maxPoolSize" value="${connection_pools.max_pool_size}" />
  <property name="maxIdleTime" value="${connection_pools.max_idle_time}" />
  <property name="acquireIncrement" value="${connection_pools.acquire_increment}" />
  <property name="checkoutTimeout" value="${connection_pools.checkout_timeout}" />
</bean>
<bean id="dataSource" class="com.p6spy.engine.spy.P6DataSource">
  <constructor-arg ref="dataSourceTarget"></constructor-arg>
</bean>
 
有部分同学将数据源的驱动包换成了com.p6spy.engine.spy.P6DataSource
其实不然,p6spy只是代理,其Driver并不能作为数据源的驱动包,故我们使用源Driver即可,在其基础上代理一层即可
 
第三步:
 
sql输出工具类P6SpyLogger
 
import java.util.Properties;

import org.apache.log4j.Logger;

import com.p6spy.engine.logging.Category;
import com.p6spy.engine.spy.appender.StdoutLogger;
import com.ruiec.web.util.RuiecStringUtil;

public class P6SpyLogger extends StdoutLogger {

    protected final Logger logger = Logger.getLogger(this.getClass());
    /**
     * 重写输出方法
     * @param connectionId 连接id
     * @param now 当前时间
     * @param elapsed 执行时长,包括执行 SQL 和处理结果集的时间(可以参考来调优)
     * @param category 语句分类,statement、resultset 等
     * @param prepared 查询语句。可能是 prepared statement,表现为 select * from table1 where c1=?,问号参数形式
     * @param sql 含参数值的查询语句,如 select * from from table1 where c1=7
     */
    @Override
    public void logSQL(int connectionId, String now, long elapsed, Category category, String prepared, String sql) {
     //我做了一个开关在application.properties中,sqlLogger等于true时输出sql方便调试,不需要使用时设置为false则不输出sql Properties properties
= new Properties(); Boolean flag = false; try { properties.load(P6SpyLogger.class.getClassLoader().getResourceAsStream("application.properties")); flag = Boolean.valueOf(properties.getProperty("sqlLogger")); } catch (Exception e) { flag = false; } if (flag) { if(!Category.COMMIT.equals(category) && !prepared.startsWith("select count(")){ this.logText(this.strategy.formatMessage(connectionId, now, elapsed, category.toString(), "-prepared-", sql)); } }
     // 要是想一直打开输出,去掉if条件以外的代码即可
    }

    @Override
    public void logText(String text) {
        StringBuilder sb = new StringBuilder();
        //匹配到最后一个|作为分隔符
        String[] arrString = text.split("\\|(?![^\\|]*\\|)");
        if (arrString.length > 1) {
            sb.append(arrString[0]);
            //去最后一段语句做替换进行格式化
            String sss=arrString[1].trim();
            if(!RuiecStringUtil.checkNull(sss) && !";".equalsIgnoreCase(sss)){
                String sql = new SQLFormatter().format(arrString[1]);
                sb.append("\r\n");
                sb.append(sql);
                sb.append("\r\n");
            }else {
                sb.append(sss);
            }
            //this.getStream().println(sb.toString());
            logger.info(sb.toString());
        } else {
            //this.getStream().println(text);
            logger.info(text);
        }
        arrString = null;
    }
}

技术分享图片

 

sql格式化工具类SQLFormatter

import java.util.HashSet;
import java.util.LinkedList;
import java.util.Set;
import java.util.StringTokenizer;

public class SQLFormatter {

    private static final Set<String> BEGIN_CLAUSES = new HashSet<String>();
    private static final Set<String> END_CLAUSES = new HashSet<String>();
    private static final Set<String> LOGICAL = new HashSet<String>();
    private static final Set<String> QUANTIFIERS = new HashSet<String>();
    private static final Set<String> DML = new HashSet<String>();
    private static final Set<String> MISC = new HashSet<String>();
    public static final String WHITESPACE = " \n\r\f\t";

    static {
        BEGIN_CLAUSES.add("left");
        BEGIN_CLAUSES.add("right");
        BEGIN_CLAUSES.add("inner");
        BEGIN_CLAUSES.add("outer");
        BEGIN_CLAUSES.add("group");
        BEGIN_CLAUSES.add("order");

        END_CLAUSES.add("where");
        END_CLAUSES.add("set");
        END_CLAUSES.add("having");
        END_CLAUSES.add("join");
        END_CLAUSES.add("from");
        END_CLAUSES.add("by");
        END_CLAUSES.add("join");
        END_CLAUSES.add("into");
        END_CLAUSES.add("union");

        LOGICAL.add("and");
        LOGICAL.add("or");
        LOGICAL.add("when");
        LOGICAL.add("else");
        LOGICAL.add("end");

        QUANTIFIERS.add("in");
        QUANTIFIERS.add("all");
        QUANTIFIERS.add("exists");
        QUANTIFIERS.add("some");
        QUANTIFIERS.add("any");

        DML.add("insert");
        DML.add("update");
        DML.add("delete");

        MISC.add("select");
        MISC.add("on");
    }

    static final String indentString = "    ";
    static final String initial = "\n    ";

    public String format(String source) {
        return new FormatProcess(source).perform();
    }

    private static class FormatProcess {
        boolean beginLine = true;
        boolean afterBeginBeforeEnd = false;
        boolean afterByOrSetOrFromOrSelect = false;
        boolean afterValues = false;
        boolean afterOn = false;
        boolean afterBetween = false;
        boolean afterInsert = false;
        int inFunction = 0;
        int parensSinceSelect = 0;
        private LinkedList<Integer> parenCounts = new LinkedList<Integer>();
        private LinkedList<Boolean> afterByOrFromOrSelects = new LinkedList<Boolean>();

        int indent = 1;

        StringBuilder result = new StringBuilder();
        StringTokenizer tokens;
        String lastToken;
        String token;
        String lcToken;

        public FormatProcess(String sql) {
            tokens = new StringTokenizer(
                    sql,
                    "()+*/-=<>‘`\"[]," + WHITESPACE,
                    true
            );
        }

        public String perform() {

            result.append(initial);

            while (tokens.hasMoreTokens()) {
                token = tokens.nextToken();
                lcToken = token.toLowerCase();

                if ("‘".equals(token)) {
                    String t;
                    do {
                        t = tokens.nextToken();
                        token += t;
                    }
                    while (!"‘".equals(t) && tokens.hasMoreTokens()); // cannot handle single quotes
                } else if ("\"".equals(token)) {
                    String t;
                    do {
                        t = tokens.nextToken();
                        token += t;
                    }
                    while (!"\"".equals(t));
                }

                if (afterByOrSetOrFromOrSelect && ",".equals(token)) {
                    commaAfterByOrFromOrSelect();
                } else if (afterOn && ",".equals(token)) {
                    commaAfterOn();
                } else if ("(".equals(token)) {
                    openParen();
                } else if (")".equals(token)) {
                    closeParen();
                } else if (BEGIN_CLAUSES.contains(lcToken)) {
                    beginNewClause();
                } else if (END_CLAUSES.contains(lcToken)) {
                    endNewClause();
                } else if ("select".equals(lcToken)) {
                    select();
                } else if (DML.contains(lcToken)) {
                    updateOrInsertOrDelete();
                } else if ("values".equals(lcToken)) {
                    values();
                } else if ("on".equals(lcToken)) {
                    on();
                } else if (afterBetween && lcToken.equals("and")) {
                    misc();
                    afterBetween = false;
                } else if (LOGICAL.contains(lcToken)) {
                    logical();
                } else if (isWhitespace(token)) {
                    white();
                } else {
                    misc();
                }

                if (!isWhitespace(token)) {
                    lastToken = lcToken;
                }

            }
            return result.toString();
        }

        private void commaAfterOn() {
            out();
            indent--;
            newline();
            afterOn = false;
            afterByOrSetOrFromOrSelect = true;
        }

        private void commaAfterByOrFromOrSelect() {
            out();
            newline();
        }

        private void logical() {
            if ("end".equals(lcToken)) {
                indent--;
            }
            newline();
            out();
            beginLine = false;
        }

        private void on() {
            indent++;
            afterOn = true;
            newline();
            out();
            beginLine = false;
        }

        private void misc() {
            out();
            if ("between".equals(lcToken)) {
                afterBetween = true;
            }
            if (afterInsert) {
                newline();
                afterInsert = false;
            } else {
                beginLine = false;
                if ("case".equals(lcToken)) {
                    indent++;
                }
            }
        }

        private void white() {
            if (!beginLine) {
                result.append(" ");
            }
        }

        private void updateOrInsertOrDelete() {
            out();
            indent++;
            beginLine = false;
            if ("update".equals(lcToken)) {
                newline();
            }
            if ("insert".equals(lcToken)) {
                afterInsert = true;
            }
        }

        @SuppressWarnings({"UnnecessaryBoxing"})
        private void select() {
            out();
            indent++;
            newline();
            parenCounts.addLast(Integer.valueOf(parensSinceSelect));
            afterByOrFromOrSelects.addLast(Boolean.valueOf(afterByOrSetOrFromOrSelect));
            parensSinceSelect = 0;
            afterByOrSetOrFromOrSelect = true;
        }

        private void out() {
            result.append(token);
        }

        private void endNewClause() {
            if (!afterBeginBeforeEnd) {
                indent--;
                if (afterOn) {
                    indent--;
                    afterOn = false;
                }
                newline();
            }
            out();
            if (!"union".equals(lcToken)) {
                indent++;
            }
            newline();
            afterBeginBeforeEnd = false;
            afterByOrSetOrFromOrSelect = "by".equals(lcToken)
                    || "set".equals(lcToken)
                    || "from".equals(lcToken);
        }

        private void beginNewClause() {
            if (!afterBeginBeforeEnd) {
                if (afterOn) {
                    indent--;
                    afterOn = false;
                }
                indent--;
                newline();
            }
            out();
            beginLine = false;
            afterBeginBeforeEnd = true;
        }

        private void values() {
            indent--;
            newline();
            out();
            indent++;
            newline();
            afterValues = true;
        }

        @SuppressWarnings({"UnnecessaryUnboxing"})
        private void closeParen() {
            parensSinceSelect--;
            if (parensSinceSelect < 0) {
                indent--;
                parensSinceSelect = parenCounts.removeLast().intValue();
                afterByOrSetOrFromOrSelect = afterByOrFromOrSelects.removeLast().booleanValue();
            }
            if (inFunction > 0) {
                inFunction--;
                out();
            } else {
                if (!afterByOrSetOrFromOrSelect) {
                    indent--;
                    newline();
                }
                out();
            }
            beginLine = false;
        }

        private void openParen() {
            if (isFunctionName(lastToken) || inFunction > 0) {
                inFunction++;
            }
            beginLine = false;
            if (inFunction > 0) {
                out();
            } else {
                out();
                if (!afterByOrSetOrFromOrSelect) {
                    indent++;
                    newline();
                    beginLine = true;
                }
            }
            parensSinceSelect++;
        }

        private static boolean isFunctionName(String tok) {
            final char begin = tok.charAt(0);
            final boolean isIdentifier = Character.isJavaIdentifierStart(begin) || ‘"‘ == begin;
            return isIdentifier &&
                    !LOGICAL.contains(tok) &&
                    !END_CLAUSES.contains(tok) &&
                    !QUANTIFIERS.contains(tok) &&
                    !DML.contains(tok) &&
                    !MISC.contains(tok);
        }

        private static boolean isWhitespace(String token) {
            return WHITESPACE.indexOf(token) >= 0;
        }

        private void newline() {
            result.append("\n");
            for (int i = 0; i < indent; i++) {
                result.append(indentString);
            }
            beginLine = true;
        }
    }

    public static void main(String[] args) {
        String sql = new SQLFormatter().format("select * from t_sss");
        System.out.println(sql);
    }


}

 

格式化工具类是为了使输出的sql更方便可读性,sql语句输出如下:
技术分享图片
 
第四步:
配置 spy.properties
#################################################################
# P6Spy Options File                                            #
# See documentation for detailed instructions                   #
# http://p6spy.github.io/p6spy/2.0/configandusage.html          #
#################################################################

#################################################################
# MODULES                                                       #
#                                                               #
# Module list adapts the modular functionality of P6Spy.        #
# Only modules listed are active.                               #
# (default is com.p6spy.engine.logging.P6LogFactory and         #
# com.p6spy.engine.spy.P6SpyFactory)                            #
# Please note that the core module (P6SpyFactory) can‘t be      #
# deactivated.                                                  #
# Unlike the other properties, activation of the changes on     #
# this one requires reload.                                     #
#################################################################
#modulelist=com.p6spy.engine.spy.P6SpyFactory,com.p6spy.engine.logging.P6LogFactory,com.p6spy.engine.outage.P6OutageFactory

################################################################
# CORE (P6SPY) PROPERTIES                                      #
################################################################

# A comma separated list of JDBC drivers to load and register.
# (default is empty)
#
# Note: This is normally only needed when using P6Spy in an
# application server environment with a JNDI data source or when
# using a JDBC driver that does not implement the JDBC 4.0 API
# (specifically automatic registration).
driverlist=com.mysql.jdbc.Driver
# for flushing per statement
# (default is false)
#autoflush = false

# sets the date format using Java‘s SimpleDateFormat routine.
# In case property is not set, miliseconds since 1.1.1970 (unix time) is used (default is empty)
dateformat=yyyy-MM-dd HH:mm:ss

# prints a stack trace for every statement logged
#stacktrace=false
# if stacktrace=true, specifies the stack trace to print
#stacktraceclass=

# determines if property file should be reloaded
# Please note: reload means forgetting all the previously set
# settings (even those set during runtime - via JMX)
# and starting with the clean table
# (default is false)
#reloadproperties=false

# determines how often should be reloaded in seconds
# (default is 60)
#reloadpropertiesinterval=60

# specifies the appender to use for logging
# Please note: reload means forgetting all the previously set
# settings (even those set during runtime - via JMX)
# and starting with the clean table
# (only the properties read from the configuration file)
# (default is com.p6spy.engine.spy.appender.FileLogger)
#appender=com.p6spy.engine.spy.appender.Slf4JLogger
#appender=com.p6spy.engine.spy.appender.StdoutLogger
#appender=com.p6spy.engine.spy.appender.FileLogger
#自定义的SQL格式化输出,换成文件存放包名
appender=cn.wuxia.ueq.ucm.view.base.support.P6SpyLogger

# name of logfile to use, note Windows users should make sure to use forward slashes in their pathname (e:/test/spy.log)
# (used for com.p6spy.engine.spy.appender.FileLogger only)
# (default is spy.log)
#logfile = spy.log

# append to the p6spy log file. if this is set to false the
# log file is truncated every time. (file logger only)
# (default is true)
#append=true

# class to use for formatting log messages (default is: com.p6spy.engine.spy.appender.SingleLineFormat)
logMessageFormat=com.p6spy.engine.spy.appender.SingleLineFormat
#logMessageFormat=com.p6spy.engine.spy.appender.MultiLineFormat

# format that is used for logging of the date/time/... (has to be compatible with java.text.SimpleDateFormat)
# (default is dd-MMM-yy)
databaseDialectDateFormat=yyyy-MM-dd HH:mm:ss

# whether to expose options via JMX or not
# (default is true)
#jmx=true

# if exposing options via jmx (see option: jmx), what should be the prefix used?
# jmx naming pattern constructed is: com.p6spy(.<jmxPrefix>)?:name=<optionsClassName>
# please note, if there is already such a name in use it would be unregistered first (the last registered wins)
# (default is none)
#jmxPrefix=

# if set to true, the execution time will be measured in nanoseconds as opposed to milliseconds
# (default is false)
#useNanoTime=false

#################################################################
# DataSource replacement                                        #
#                                                               #
# Replace the real DataSource class in your application server  #
# configuration with the name com.p6spy.engine.spy.P6DataSource #
# (that provides also connection pooling and xa support).       #
# then add the JNDI name and class name of the real             #
# DataSource here                                               #
#                                                               #
# Values set in this item cannot be reloaded using the          #
# reloadproperties variable. Once it is loaded, it remains      #
# in memory until the application is restarted.                 #
#                                                               #
#################################################################
#realdatasource=/RealMySqlDS
#realdatasourceclass=com.mysql.jdbc.jdbc2.optional.MysqlDataSource

#################################################################
# DataSource properties                                         #
#                                                               #
# If you are using the DataSource support to intercept calls    #
# to a DataSource that requires properties for proper setup,    #
# define those properties here. Use name value pairs, separate  #
# the name and value with a semicolon, and separate the         #
# pairs with commas.                                            #
#                                                               #
# The example shown here is for mysql                           #
#                                                               #
#################################################################
#realdatasourceproperties=port;3306,serverName;myhost,databaseName;jbossdb,foo;bar

#################################################################
# JNDI DataSource lookup                                        #
#                                                               #
# If you are using the DataSource support outside of an app     #
# server, you will probably need to define the JNDI Context     #
# environment.                                                  #
#                                                               #
# If the P6Spy code will be executing inside an app server then #
# do not use these properties, and the DataSource lookup will   #
# use the naming context defined by the app server.             #
#                                                               #
# The two standard elements of the naming environment are       #
# jndicontextfactory and jndicontextproviderurl. If you need    #
# additional elements, use the jndicontextcustom property.      #
# You can define multiple properties in jndicontextcustom,      #
# in name value pairs. Separate the name and value with a       #
# semicolon, and separate the pairs with commas.                #
#                                                               #
# The example shown here is for a standalone program running on #
# a machine that is also running JBoss, so the JDNI context     #
# is configured for JBoss (3.0.4).                              #
#                                                               #
# (by default all these are empty)                              #
#################################################################
#jndicontextfactory=org.jnp.interfaces.NamingContextFactory
#jndicontextproviderurl=localhost:1099
#jndicontextcustom=java.naming.factory.url.pkgs;org.jboss.nameing:org.jnp.interfaces

#jndicontextfactory=com.ibm.websphere.naming.WsnInitialContextFactory
#jndicontextproviderurl=iiop://localhost:900

################################################################
# P6 LOGGING SPECIFIC PROPERTIES                               #
################################################################

# filter what is logged
# please note this is a precondition for usage of: include/exclude/sqlexpression
# (default is false)
#filter=false

# comma separated list of strings to include
# please note that special characters escaping (used in java) has to be done for the provided regular expression
# (default is empty)
#include =
# comma separated list of strings to exclude
# (default is empty)
#exclude =

# sql expression to evaluate if using regex
# please note that special characters escaping (used in java) has to be done for the provided regular expression
# (default is empty)
#sqlexpression =

#list of categories to exclude: error, info, batch, debug, statement,
#commit, rollback and result are valid values
# (default is info,debug,result,resultset,batch)
#excludecategories=info,debug,result,resultset,batch

# Execution threshold applies to the standard logging of P6Spy.
# While the standard logging logs out every statement
# regardless of its execution time, this feature puts a time
# condition on that logging. Only statements that have taken
# longer than the time specified (in milliseconds) will be
# logged. This way it is possible to see only statements that
# have exceeded some high water mark.
# This time is reloadable.
#
# executionThreshold=integer time (milliseconds)
# (default is 0)
#executionThreshold=

################################################################
# P6 OUTAGE SPECIFIC PROPERTIES                                #
################################################################
# Outage Detection
#
# This feature detects long-running statements that may be indicative of
# a database outage problem. If this feature is turned on, it will log any
# statement that surpasses the configurable time boundary during its execution.
# When this feature is enabled, no other statements are logged except the long
# running statements. The interval property is the boundary time set in seconds.
# For example, if this is set to 2, then any statement requiring at least 2
# seconds will be logged. Note that the same statement will continue to be logged
# for as long as it executes. So if the interval is set to 2, and the query takes
# 11 seconds, it will be logged 5 times (at the 2, 4, 6, 8, 10 second intervals).
#
# outagedetection=true|false
# outagedetectioninterval=integer time (seconds)
#
# (default is false)
#outagedetection=false
# (default is 60)
#outagedetectioninterval=30

 

总结:
1.出于安全需要,我改为使用 logger 在 debug 模式下输出。
2.关于使用Slf4j还是Log4j个人推荐使用Slf4j,虽然只是一个简单日志门面,log4j比它强大,但是部署项目时可以方便的变更日志系统
3.很多人可能在配置数据源的时候卡住,p6spy只是代理,其Driver并不能作为数据源的驱动包,故我们使用源Driver即可,在其基础上代理一层即可
 

配置p6spyLog输出sql完整日志

标签:option   memory   unix   http   格式   done   sid   sla   lin   

原文地址:https://www.cnblogs.com/MrSong97/p/10221243.html

(0)
(0)
   
举报
评论 一句话评论(0
登录后才能评论!
© 2014 mamicode.com 版权所有  联系我们:gaon5@hotmail.com
迷上了代码!