隨筆 - 41  文章 - 29  trackbacks - 0
          <2009年3月>
          22232425262728
          1234567
          891011121314
          15161718192021
          22232425262728
          2930311234

          常用鏈接

          留言簿(5)

          隨筆分類(28)

          隨筆檔案(23)

          收藏夾(6)

          Inside JVM

          Java

          java performance

          Solr

          搜索

          •  

          最新評論

          閱讀排行榜

          評論排行榜

          Database performance is the critical component of a web application. Recently, we spend lots of time to monitor and analyze SQL server performance based on SQL profiling log. However, SQL profiling log is row-based result set and is difficult to do trouble shooting directly. This message is trying to introduce some very good SQL monitoring and profiling tools to get much better SQL profiling result.

          Another Chinese Article on this topic

          There is a very good article about this topic. However, it is Chinese version. “http://www.ibm.com/developerworks/cn/java/j-lo-p6spy/index.html#resources”. 


          Background

          So, we are going to introduce three tools: P6Spy, IronTrackSQL and SQL Profiler (NOT SQL Server Profiler). Using these tools are very easy to log all SQL statements and show the detailed statistics results.

          P6Spy

          P6Spy is an open source framework for applications that intercept and optionally modify database statements. The P6Spy distribution includes the following modules:

          • P6Log. P6Log intercepts and logs the database statements of any application that uses JDBC. P6Spy is designed to be installed in minutes and requires no code changes.
          • P6Outage. P6Outage detects long-running statements that may be indicative of a database outage proble and will log any statement that surpasses the configurable time boundary during its execution.

          The deployment is very easy

          • Copy p6spy.jar to CLASSPATH. if web application, copy it to YourWebApp/WEB-INF/lib/
          • Copy spy.properties to CLASSPATH. if web application, copy it to YourWebApp/WEB-INF/classess/. NOTE: it is not lib.
          • Find your database connection setting, for example, Database resource setting within context.xml, and use “com.p6spy.engine.spy.P6SpyDriver” to replace the original driver class
          • Update the property “realdriver” as your real database driver as following:
            realdriver=net.sourceforge.jtds.jdbc.Driver
          • Run you application, and you find “spy.log” in your bin folder. And this log file path can be updated in spy.properties

          The Log Format

          current time|execution time|category|statement SQL String|effective SQL string


          1237024804522|93|5|statement|SELECT vp.purchaseID, vp.externalVendorID, vp.uniqueID, vp.purchaseWord, vp.purchaseValue, vp.purchaseType, vs.chains AS brands, vs.starRating, vs.amenities, vs.types, vsd.communityRatingTotal, vsd.communityRatingCount FROM VS_VendorPurchase AS vp LEFT JOIN VS_VendorSearch AS vs ON vp.externalVendorID = vs.externalVendorID LEFT JOIN VS_VendorSearchData AS vsd ON vp.externalVendorID = vsd.externalVendorID |SELECT vp.purchaseID, vp.externalVendorID, vp.uniqueID, vp.purchaseWord, vp.purchaseValue, vp.purchaseType, vs.chains AS brands, vs.starRating, vs.amenities, vs.types, vsd.communityRatingTotal, vsd.communityRatingCount FROM VS_VendorPurchase AS vp LEFT JOIN VS_VendorSearch AS vs ON vp.externalVendorID = vs.externalVendorID LEFT JOIN VS_VendorSearchData AS vsd ON vp.externalVendorID = vsd.externalVendorID

          Other Setting:
          P6Spy have lot of customized setting, such as time format, included table names or excluded table names, the interval values used to identify the long-running SQL, etc. Please see http://www.p6spy.com/documentation/other.htm.

          IronTrackSQL

          IronTrack SQL provides a GUI interface with reports and graphs and a detailed analysis of any SQL statement, comparing it with the system average.

          The deployment is also very easy

          • First of all, all jar files and property file need be copied from IronTrackSQL.zip
          • Copy irontracksql.jar, p6spy.jar and log4j-1.2.8.jar to CLASSPATH. if web application, copy it to YourWebApp/WEB-INF/lib/.
          • Copy spy.properties to CLASSPATH. if web application, copy it to YourWebApp/WEB-INF/classess/. NOTE: it is not lib.
          • Find your database connection setting, for example, Database resource setting within context.xml, and use “com.p6spy.engine.spy.P6SpyDriver” to replace the original driver class
          • Update the property “realdriver” as your real database driver as following:
            realdriver=net.sourceforge.jtds.jdbc.Driver
          • Run you application, and you find “spy.log” in your bin folder. And this log file path can be updated in spy.properties
          • Double click irontracksql.jar to start GUI, and click “File=>Connect” to connect to your application (or web server). From the status bar, you will see the connection information if successfully connecting to application

          GUI example

          • please see attached file
          • You can see count, average time and max time for each SQL statement
          • And you can filter logs based on some filter conditions, such as avg time from 10ms to 15ms

          Analyzing SQL Statement Performance with the SQL Monitor Section

          • High preparation time
            Queries that require high preparation time may perform unnecessary activities in the time between when a statement is initialized and when it is executed. Consider moving any code related to building the SQL query before creating the Statement object.
          • Frequent queries
            • Consider queries that are called frequently as candidates for data caching.
            • The Java PreparedStatement object gives the database an opportunity to pre-parse the query, which can save time when a query is called repeatedly.
            • Consider consolidating multiple queries of the same table into a single statement.
          • High execution time — Time spent in the executeQuery() or executeUpdate() methods is time that the database is spending to process and execute the query.
          • High retrieval time — Retrieval time includes all of the time spent returning records from the database to the application. The primary drivers of this time are the number of rows returned, and the amount of data in each row. Examine queries for opportunities to limit the number of columns, rows, or both returned.

          SQL Profiler – Not SQL Server Profiling


          This Swing-based SQL profiler tries to offer a tool to monitor which tables and columns are accessed the most in SQL queries to recommend index creation.

          The tool has the very similar capabilities of IronTrackSQL. However, i think IronTrackSQL has a better user interface. I recommend you to use IronTrackSQL.

          However, this tool can generate recommended indexes based on the SQL log. This is a very interesting feature.

          Overall

          Overall, these 3 tools facilitate developers on SQL statement logging and profiling without any code change, and is independent to any web container and database management system.

          Any issue, please let me know.

          posted on 2009-03-14 21:06 Justin Chen 閱讀(3551) 評論(0)  編輯  收藏 所屬分類: Tools
          主站蜘蛛池模板: 大安市| 罗定市| 铁力市| 蓬安县| 兴隆县| 临桂县| 璧山县| 铜川市| 广元市| 常熟市| 林甸县| 甘德县| 广州市| 凤台县| 交口县| 溆浦县| 遂川县| 富平县| 永新县| 南雄市| 页游| 亳州市| 延安市| 太仆寺旗| 鄂州市| 安仁县| 上林县| 镇康县| 久治县| 石河子市| 麦盖提县| 洛隆县| 宣汉县| 民县| 玉山县| 桐庐县| 竹山县| 汤原县| 廊坊市| 五指山市| 苏州市|