Dave Ward 5245ec7316 Merged V4.0-BUG-FIX to HEAD
36604: ALF-13745: Merged V3.4-BUG-FIX (3.4.10) to V4.0-BUG-FIX (4.0.3)
      36602: ALF-13667 Additional OpenOffice mimetypes to be added to the mime-type maps
         - xls to pdf limit had been commented out
         - failover.transformer.PdfRenderer.PdfToImage and failover.transformer.PdfBox.PdfToImage should not have been
           been registered. Introduced an unregisteredBaseContentTransformer. In the case of these two transforms this would
           not have been a problem, as they were disable anyway due to EXPLICIT setting elsewhere.
   36608: First part of ALF-14209 SOLR - does not support query for all stores
   - dynamic SOLR cores to track a store
   - NEW, DELETE, and change properties and reload
   36635: ALF-13404 for documentlibrary-v2 APIs
   36669: Fix issue with rendering multi-valued properties in JSON from server-side JS (from DaveD)
   36676: Fix for ALF-14216 Solr Exception when you try to sort folders or files by size.
   36692: Fix ALF-12966 - Comments doesn't work on iOS Safari. Anywhere TinyMCE is present does not work correctly e.g. comments, blogs, wiki, HTML content creation.
   36693: ALF-14138: Prevent default Surf CMIS content WebScript clashing with Alfreco version (updated Surf libs r1081)
   36714: ALF-14224: WorkflowTaskDefinition form-key fetching fixed for multi-instance UserTasks
   36721: Fix for ALF-8374 - Simple view: incorrect file type icon for *.page and *.eps files
   36726: Merged BRANCHES/DEV/V3.4-BUG-FIX to BRANCHES/DEV/V4.0-BUG-FIX
      36724: Fix for ALF-14207 - cm:link not correctly handled in Share's doclist when users don't have permission
   36728: Fix for ALF-14002: "No dashlets listed when customizing Site dashboard for the "Web Site Design Project" sample Share Site "
   36732: Merged V3.4-BUG-FIX to V4.0-BUG-FIX
      36637: ALF-6162 Thumbnail is not produced for PDFs encoded with JBIG2Decode
         << Reported problem still exists, but have made improvements see https://issues.alfresco.com/jira/browse/ALF-6162?focusedCommentId=162936&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-162936 >>
         - PDFRenderer now fails with an exception in more cases, rather than just silently continuing. Specific case was not handling JBig2 images.
           previously just resulted in a black box)
         - Handle the font family CIDFontType0 which includes HiddenHorzOCR
      36640: ALF-13769: Merged BELARUS/DEV/V3.4-BUG-FIX-2012_05_04 to V3.4-BUG-FIX (3.4.10)
         36440: ALF-11956: WCM accessibility
            Assignment of TinyMCE accessibility hotkeys only to activated RTE between multiple instances is added.
            Headings and some other related labels within the context of 'Create Web Content Wizard' were shortened and clarified.
            Fixed some issues related to hidden text for accessibility support in context of IE, which doesn't calculate styles and dimensions that are applied to elements before adding to DOM
         36443: ALF-11956: WCM accessibility
            Added some changes missed in r36440
      36645: ALF-13769: Merged DEV/V3.4-BUG-FIX-2012_01_10 to V3.4-BUG-FIX (3.4.10)
         33381: ALF-10162: Web Form validation can be bypassed
            1. New method was introduced to XFormsBean which allows it to validate the last inserted values.
            2. CreateWebContentWizard finds XFormsBean in session and invokes isXformValid. If any validation
               errors exist it goes back and show the JSF error.
            3. xform.js - contains code that removes JSF error as well as the XForms error. This is necessary because we need to clear
               the JSF error on the Javascript actions
            4. container.jsp - includes a div wrapper with the id to be able unambiguously find the JSF error block.
      36700: Merged DEV to V3.4-BUG-FIX (Reviewed by Erik)
         36450: ALF-12261 : IE8 Specific: It's impossible add relationship type "Parent/Child"
            The type of the submit button should be set to 'button' in 'new-rmreference.get.html.ftl'.
      36702: Merged DEV to V3.4-BUG-FIX (Reviewed by Erik)
         36670: ALF-12825: Impossible choose the "Publication Date", the calendar isn't displayed
            The DatePicker component register a validation handler for the date entry field so that the submit button disables when an invalid date is entered. This handler register with the forms runtime instance. We should create this instance in FormUI_consructor, not is onReady method (form.js), because constructors always in order will be invoked. 
      36703: Fix for Mac Office 2011 Powerpoint save fails on CIFS. ALF-13615.
   36733: Merged V4.0 to V4.0-BUG-FIX
      36628: Added ${NOW} variable option for schema bootstrap scripts (see ALF-14174)
      36632: Fix last part of ALF-14174: The patch adding timestamps to acl_change_set breaks SOLR tracking
       - Added ${NOW} to ACL change set timestamp when upgrading: commit_time_ms = ${NOW}
      36647: ALF-14190: FSTR is not working
      - Due to someone updating dependencies without updating the build scripts!
   36734: Merged V4.0 to V4.0-BUG-FIX (RECORD ONLY)
      36605: ALF-13745: Merged V4.0-BUG-FIX (4.0.3) to V4.0 (4.0.2)
         36604: ALF-13745: Merged V3.4-BUG-FIX (3.4.10) to V4.0-BUG-FIX (4.0.3)
            36602: ALF-13667 Additional OpenOffice mimetypes to be added to the mime-type maps
      36678: Merged BRANCHES/DEV/V4.0-BUG-FIX to BRANCHES/V4.0
          36675: Fix for ALF-14216 Solr Exception when you try to sort folders or files by size.


git-svn-id: https://svn.alfresco.com/repos/alfresco-enterprise/alfresco/HEAD/root@36737 c4b6b30b-aa2e-2d43-bbcb-ca4b014f7261
2012-05-22 16:51:44 +00:00

2377 lines
89 KiB
Java

/*
* Copyright (C) 2005-2010 Alfresco Software Limited.
*
* This file is part of Alfresco
*
* Alfresco is free software: you can redistribute it and/or modify
* it under the terms of the GNU Lesser General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* Alfresco is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public License
* along with Alfresco. If not, see <http://www.gnu.org/licenses/>.
*/
package org.alfresco.repo.domain.schema;
import java.io.BufferedInputStream;
import java.io.BufferedReader;
import java.io.BufferedWriter;
import java.io.File;
import java.io.FileInputStream;
import java.io.FileNotFoundException;
import java.io.FileWriter;
import java.io.IOException;
import java.io.InputStream;
import java.io.InputStreamReader;
import java.io.PrintWriter;
import java.io.UnsupportedEncodingException;
import java.io.Writer;
import java.sql.Array;
import java.sql.Blob;
import java.sql.CallableStatement;
import java.sql.Clob;
import java.sql.Connection;
import java.sql.DatabaseMetaData;
import java.sql.DriverManager;
import java.sql.NClob;
import java.sql.PreparedStatement;
import java.sql.ResultSet;
import java.sql.SQLClientInfoException;
import java.sql.SQLException;
import java.sql.SQLWarning;
import java.sql.SQLXML;
import java.sql.Savepoint;
import java.sql.Statement;
import java.sql.Struct;
import java.sql.Types;
import java.text.MessageFormat;
import java.util.ArrayList;
import java.util.Date;
import java.util.HashMap;
import java.util.List;
import java.util.Map;
import java.util.Properties;
import javax.sql.DataSource;
import org.activiti.engine.ProcessEngine;
import org.activiti.engine.ProcessEngineConfiguration;
import org.alfresco.error.AlfrescoRuntimeException;
import org.alfresco.ibatis.SerializableTypeHandler;
import org.alfresco.repo.admin.patch.Patch;
import org.alfresco.repo.admin.patch.impl.SchemaUpgradeScriptPatch;
import org.alfresco.repo.content.filestore.FileContentWriter;
import org.alfresco.repo.domain.PropertyValue;
import org.alfresco.repo.domain.hibernate.dialect.AlfrescoOracle9Dialect;
import org.alfresco.repo.domain.hibernate.dialect.AlfrescoSQLServerDialect;
import org.alfresco.repo.domain.hibernate.dialect.AlfrescoSybaseAnywhereDialect;
import org.alfresco.service.cmr.repository.ContentWriter;
import org.alfresco.service.descriptor.DescriptorService;
import org.alfresco.util.LogUtil;
import org.alfresco.util.TempFileProvider;
import org.alfresco.util.schemacomp.ExportDb;
import org.alfresco.util.schemacomp.MultiFileDumper;
import org.alfresco.util.schemacomp.MultiFileDumper.DbToXMLFactory;
import org.alfresco.util.schemacomp.Result;
import org.alfresco.util.schemacomp.Results;
import org.alfresco.util.schemacomp.SchemaComparator;
import org.alfresco.util.schemacomp.XMLToSchema;
import org.alfresco.util.schemacomp.model.Schema;
import org.alfresco.util.schemadump.Main;
import org.apache.commons.logging.Log;
import org.apache.commons.logging.LogFactory;
import org.hibernate.HibernateException;
import org.hibernate.Session;
import org.hibernate.SessionFactory;
import org.hibernate.cfg.Configuration;
import org.hibernate.cfg.Environment;
import org.hibernate.connection.UserSuppliedConnectionProvider;
import org.hibernate.dialect.DB2Dialect;
import org.hibernate.dialect.DerbyDialect;
import org.hibernate.dialect.Dialect;
import org.hibernate.dialect.HSQLDialect;
import org.hibernate.dialect.MySQL5Dialect;
import org.hibernate.dialect.MySQLDialect;
import org.hibernate.dialect.MySQLInnoDBDialect;
import org.hibernate.dialect.Oracle10gDialect;
import org.hibernate.dialect.Oracle9Dialect;
import org.hibernate.dialect.Oracle9iDialect;
import org.hibernate.dialect.OracleDialect;
import org.hibernate.dialect.PostgreSQLDialect;
import org.hibernate.engine.ActionQueue;
import org.hibernate.tool.hbm2ddl.DatabaseMetadata;
import org.springframework.context.ApplicationContext;
import org.springframework.context.ApplicationEvent;
import org.springframework.core.io.Resource;
import org.springframework.core.io.support.PathMatchingResourcePatternResolver;
import org.springframework.core.io.support.ResourcePatternResolver;
import org.springframework.extensions.surf.util.AbstractLifecycleBean;
import org.springframework.extensions.surf.util.I18NUtil;
import org.springframework.orm.hibernate3.LocalSessionFactoryBean;
/**
* Bootstraps the schema and schema update. The schema is considered missing if the applied patch table
* is not present, and the schema is considered empty if the applied patch table is empty.
*
* @author Derek Hulley
*/
public class SchemaBootstrap extends AbstractLifecycleBean
{
/** The placeholder for the configured <code>Dialect</code> class name: <b>${db.script.dialect}</b> */
private static final String PLACEHOLDER_DIALECT = "\\$\\{db\\.script\\.dialect\\}";
/** The global property containing the default batch size used by --FOREACH */
private static final String PROPERTY_DEFAULT_BATCH_SIZE = "system.upgrade.default.batchsize";
private static final String MSG_DIALECT_USED = "schema.update.msg.dialect_used";
private static final String MSG_BYPASSING_SCHEMA_UPDATE = "schema.update.msg.bypassing";
private static final String MSG_NORMALIZED_SCHEMA = "schema.update.msg.normalized_schema";
private static final String MSG_NORMALIZED_SCHEMA_PRE = "schema.update.msg.normalized_schema_pre";
private static final String MSG_NORMALIZED_SCHEMA_POST = "schema.update.msg.normalized_schema_post";
private static final String MSG_NO_CHANGES = "schema.update.msg.no_changes";
private static final String MSG_ALL_STATEMENTS = "schema.update.msg.all_statements";
private static final String MSG_EXECUTING_GENERATED_SCRIPT = "schema.update.msg.executing_generated_script";
private static final String MSG_EXECUTING_COPIED_SCRIPT = "schema.update.msg.executing_copied_script";
private static final String MSG_EXECUTING_STATEMENT = "schema.update.msg.executing_statement";
private static final String MSG_OPTIONAL_STATEMENT_FAILED = "schema.update.msg.optional_statement_failed";
private static final String WARN_DIALECT_UNSUPPORTED = "schema.update.warn.dialect_unsupported";
private static final String WARN_DIALECT_SUBSTITUTING = "schema.update.warn.dialect_substituting";
private static final String WARN_DIALECT_HSQL = "schema.update.warn.dialect_hsql";
private static final String WARN_DIALECT_DERBY = "schema.update.warn.dialect_derby";
private static final String ERR_FORCED_STOP = "schema.update.err.forced_stop";
private static final String ERR_DIALECT_SHOULD_USE = "schema.update.err.dialect_should_use";
private static final String ERR_MULTIPLE_SCHEMAS = "schema.update.err.found_multiple";
private static final String ERR_PREVIOUS_FAILED_BOOTSTRAP = "schema.update.err.previous_failed";
private static final String ERR_STATEMENT_FAILED = "schema.update.err.statement_failed";
private static final String ERR_UPDATE_FAILED = "schema.update.err.update_failed";
private static final String ERR_VALIDATION_FAILED = "schema.update.err.validation_failed";
private static final String ERR_SCRIPT_NOT_RUN = "schema.update.err.update_script_not_run";
private static final String ERR_SCRIPT_NOT_FOUND = "schema.update.err.script_not_found";
private static final String ERR_STATEMENT_INCLUDE_BEFORE_SQL = "schema.update.err.statement_include_before_sql";
private static final String ERR_STATEMENT_VAR_ASSIGNMENT_BEFORE_SQL = "schema.update.err.statement_var_assignment_before_sql";
private static final String ERR_STATEMENT_VAR_ASSIGNMENT_FORMAT = "schema.update.err.statement_var_assignment_format";
private static final String ERR_STATEMENT_TERMINATOR = "schema.update.err.statement_terminator";
private static final String DEBUG_SCHEMA_COMP_NO_REF_FILE = "system.schema_comp.debug.no_ref_file";
private static final String INFO_SCHEMA_COMP_ALL_OK = "system.schema_comp.info.all_ok";
private static final String WARN_SCHEMA_COMP_PROBLEMS_FOUND = "system.schema_comp.warn.problems_found";
private static final String DEBUG_SCHEMA_COMP_TIME_TAKEN = "system.schema_comp.debug.time_taken";
public static final int DEFAULT_LOCK_RETRY_COUNT = 24;
public static final int DEFAULT_LOCK_RETRY_WAIT_SECONDS = 5;
public static final int DEFAULT_MAX_STRING_LENGTH = 1024;
private static volatile int maxStringLength = DEFAULT_MAX_STRING_LENGTH;
private Dialect dialect;
private ResourcePatternResolver rpr = new PathMatchingResourcePatternResolver(this.getClass().getClassLoader());
/**
* @see PropertyValue#DEFAULT_MAX_STRING_LENGTH
*/
private static final void setMaxStringLength(int length)
{
if (length < 1024)
{
throw new AlfrescoRuntimeException("The maximum string length must >= 1024 characters.");
}
SchemaBootstrap.maxStringLength = length;
}
/**
* @return Returns the maximum number of characters that a string field can be
*/
public static final int getMaxStringLength()
{
return SchemaBootstrap.maxStringLength;
}
/**
* Truncates or returns a string that will fit into the string columns in the schema. Text fields can
* either cope with arbitrarily long text fields or have the default limit, {@link #DEFAULT_MAX_STRING_LENGTH}.
*
* @param value the string to check
* @return Returns a string that is short enough for {@link SchemaBootstrap#getMaxStringLength()}
*
* @since 3.2
*/
public static final String trimStringForTextFields(String value)
{
if (value != null && value.length() > maxStringLength)
{
return value.substring(0, maxStringLength);
}
else
{
return value;
}
}
/**
* Provide a reference to the DescriptorService, used to provide information
* about the repository such as the database schema version number.
*
* @param descriptorService the descriptorService to set
*/
public void setDescriptorService(DescriptorService descriptorService)
{
this.descriptorService = descriptorService;
}
/**
* Sets the previously auto-detected Hibernate dialect.
*
* @param dialect
* the dialect
*/
public void setDialect(Dialect dialect)
{
this.dialect = dialect;
}
private static Log logger = LogFactory.getLog(SchemaBootstrap.class);
private DescriptorService descriptorService;
private DataSource dataSource;
private LocalSessionFactoryBean localSessionFactory;
private String schemaOuputFilename;
private boolean updateSchema;
private boolean stopAfterSchemaBootstrap;
private List<String> preCreateScriptUrls;
private List<String> postCreateScriptUrls;
private List<String> schemaReferenceUrls;
private List<SchemaUpgradeScriptPatch> validateUpdateScriptPatches;
private List<SchemaUpgradeScriptPatch> preUpdateScriptPatches;
private List<SchemaUpgradeScriptPatch> postUpdateScriptPatches;
private int schemaUpdateLockRetryCount = DEFAULT_LOCK_RETRY_COUNT;
private int schemaUpdateLockRetryWaitSeconds = DEFAULT_LOCK_RETRY_WAIT_SECONDS;
private int maximumStringLength;
private Properties globalProperties;
private ThreadLocal<StringBuilder> executedStatementsThreadLocal = new ThreadLocal<StringBuilder>();
private File xmlPreSchemaOutputFile34; // This must be set if there are any executed statements
public SchemaBootstrap()
{
preCreateScriptUrls = new ArrayList<String>(1);
postCreateScriptUrls = new ArrayList<String>(1);
validateUpdateScriptPatches = new ArrayList<SchemaUpgradeScriptPatch>(4);
preUpdateScriptPatches = new ArrayList<SchemaUpgradeScriptPatch>(4);
postUpdateScriptPatches = new ArrayList<SchemaUpgradeScriptPatch>(4);
maximumStringLength = -1;
globalProperties = new Properties();
}
public void setDataSource(DataSource dataSource)
{
this.dataSource = dataSource;
}
public void setLocalSessionFactory(LocalSessionFactoryBean localSessionFactory)
{
this.localSessionFactory = localSessionFactory;
}
public LocalSessionFactoryBean getLocalSessionFactory()
{
return localSessionFactory;
}
/**
* Set this to output the full database creation script
*
* @param schemaOuputFilename the name of a file to dump the schema to, or null to ignore
*/
public void setSchemaOuputFilename(String schemaOuputFilename)
{
this.schemaOuputFilename = schemaOuputFilename;
}
/**
* Set whether to modify the schema or not. Either way, the schema will be validated.
*
* @param updateSchema true to update and validate the schema, otherwise false to just
* validate the schema. Default is <b>true</b>.
*/
public void setUpdateSchema(boolean updateSchema)
{
this.updateSchema = updateSchema;
}
/**
* Set whether this component should terminate the bootstrap process after running all the
* usual checks and scripts. This has the additional effect of dumping a final schema
* structure file just before exiting.
* <p>
* <b>WARNING: </b>USE FOR DEBUG AND UPGRADE TESTING ONLY
*
* @param stopAfterSchemaBootstrap <tt>true</tt> to terminate (with exception) after
* running all the usual schema updates and checks.
*/
public void setStopAfterSchemaBootstrap(boolean stopAfterSchemaBootstrap)
{
this.stopAfterSchemaBootstrap = stopAfterSchemaBootstrap;
}
/**
* Set the scripts that must be executed <b>before</b> the schema has been created.
*
* @param postCreateScriptUrls file URLs
*
* @see #PLACEHOLDER_DIALECT
*/
public void setPreCreateScriptUrls(List<String> preUpdateScriptUrls)
{
this.preCreateScriptUrls = preUpdateScriptUrls;
}
/**
* Set the scripts that must be executed <b>after</b> the schema has been created.
*
* @param postCreateScriptUrls file URLs
*
* @see #PLACEHOLDER_DIALECT
*/
public void setPostCreateScriptUrls(List<String> postUpdateScriptUrls)
{
this.postCreateScriptUrls = postUpdateScriptUrls;
}
/**
* Specifies the schema reference files that will be used to validate the repository
* schema whenever changes have been made. The database dialect placeholder will be
* resolved so that the correct reference files are loaded for the current database
* type (e.g. PostgreSQL)
*
* @param schemaReferenceUrls the schemaReferenceUrls to set
* @see #PLACEHOLDER_DIALECT
*/
public void setSchemaReferenceUrls(List<String> schemaReferenceUrls)
{
this.schemaReferenceUrls = schemaReferenceUrls;
}
/**
* Set the schema script patches that must have been applied. These will not be
* applied to the database. These can be used where the script <u>cannot</u> be
* applied automatically or where a particular upgrade path is no longer supported.
* For example, at version 3.0, the upgrade scripts for version 1.4 may be considered
* unsupported - this doesn't prevent the manual application of the scripts, though.
*
* @param scriptPatches a list of schema patches to check
*/
public void setValidateUpdateScriptPatches(List<SchemaUpgradeScriptPatch> scriptPatches)
{
this.validateUpdateScriptPatches = scriptPatches;
}
/**
* Set the schema script patches that may be applied prior to the auto-update process.
*
* @param scriptPatches a list of schema patches to check
*/
public void setPreUpdateScriptPatches(List<SchemaUpgradeScriptPatch> scriptPatches)
{
this.preUpdateScriptPatches = scriptPatches;
}
/**
* Set the schema script patches that may be applied after the auto-update process.
*
* @param postUpdateScriptPatches a list of schema patches to check
*/
public void setPostUpdateScriptPatches(List<SchemaUpgradeScriptPatch> scriptPatches)
{
this.postUpdateScriptPatches = scriptPatches;
}
/**
* Set the number times that the DB must be checked for the presence of the table
* indicating that a schema change is in progress.
*
* @param schemaUpdateLockRetryCount the number of times to retry (default 24)
*/
public void setSchemaUpdateLockRetryCount(int schemaUpdateLockRetryCount)
{
this.schemaUpdateLockRetryCount = schemaUpdateLockRetryCount;
}
/**
* Set the wait time (seconds) between checks for the schema update lock.
*
* @param schemaUpdateLockRetryWaitSeconds the number of seconds between checks (default 5 seconds)
*/
public void setSchemaUpdateLockRetryWaitSeconds(int schemaUpdateLockRetryWaitSeconds)
{
this.schemaUpdateLockRetryWaitSeconds = schemaUpdateLockRetryWaitSeconds;
}
/**
* Optionally override the system's default maximum string length. Some databases have
* limitations on how long the <b>string_value</b> columns can be while other do not.
* Some parts of the persistence have alternatives when the string values exceed this
* length while others do not. Either way, it is possible to adjust the text column sizes
* and adjust this value manually to override the default associated with the database
* being used.
* <p>
* The system - as of V2.1.2 - will attempt to adjust the maximum string length size
* automatically and therefore this method is not normally required. But it is possible
* to manually override the value if, for example, the system doesn't guess the correct
* maximum length or if the dialect is not explicitly catered for.
* <p>
* All negative or zero values are ignored and the system defaults to its best guess based
* on the dialect being used.
*
* @param maximumStringLength the maximum length of the <b>string_value</b> columns
*/
public void setMaximumStringLength(int maximumStringLength)
{
if (maximumStringLength > 0)
{
this.maximumStringLength = maximumStringLength;
}
}
/**
* Get the limit for the hibernate executions queue
*/
public int getHibernateMaxExecutions()
{
return ActionQueue.getMAX_EXECUTIONS_SIZE();
}
/**
* Set the limit for the hibernate executions queue
* Less than zero always uses event amalgamation
*/
public void setHibernateMaxExecutions(int hibernateMaxExecutions)
{
ActionQueue.setMAX_EXECUTIONS_SIZE(hibernateMaxExecutions);
}
/**
* Sets the properties map from which we look up some configuration settings.
*
* @param globalProperties
* the global properties
*/
public void setGlobalProperties(Properties globalProperties)
{
this.globalProperties = globalProperties;
}
private SessionFactory getSessionFactory()
{
return (SessionFactory) localSessionFactory.getObject();
}
private static class NoSchemaException extends Exception
{
private static final long serialVersionUID = 5574280159910824660L;
}
/**
* Used to indicate a forced stop of the bootstrap.
*
* @see SchemaBootstrap#setStopAfterSchemaBootstrap(boolean)
*
* @author Derek Hulley
* @since 3.1.1
*/
private static class BootstrapStopException extends RuntimeException
{
private static final long serialVersionUID = 4250016675538442181L;
private BootstrapStopException()
{
super(I18NUtil.getMessage(ERR_FORCED_STOP));
}
}
/**
* Count applied patches. This fails if multiple applied patch tables are found,
* which normally indicates that the schema view needs to be limited.
*
* @param cfg The Hibernate config
* @param connection a valid database connection
* @return Returns the number of applied patches
* @throws NoSchemaException if the table of applied patches can't be found
*/
private int countAppliedPatches(Configuration cfg, Connection connection) throws Exception
{
String defaultSchema = cfg.getProperty("hibernate.default_schema");
if (defaultSchema != null && defaultSchema.length() == 0)
{
defaultSchema = null;
}
String defaultCatalog = cfg.getProperty("hibernate.default_catalog");
if (defaultCatalog != null && defaultCatalog.length() == 0)
{
defaultCatalog = null;
}
DatabaseMetaData dbMetadata = connection.getMetaData();
ResultSet tableRs = dbMetadata.getTables(defaultCatalog, defaultSchema, "%", null);
boolean newPatchTable = false;
boolean oldPatchTable = false;
try
{
boolean multipleSchemas = false;
while (tableRs.next())
{
String tableName = tableRs.getString("TABLE_NAME");
if (tableName.equalsIgnoreCase("applied_patch"))
{
if (oldPatchTable || newPatchTable)
{
// Found earlier
multipleSchemas = true;
}
oldPatchTable = true;
}
else if (tableName.equalsIgnoreCase("alf_applied_patch"))
{
if (oldPatchTable || newPatchTable)
{
// Found earlier
multipleSchemas = true;
}
newPatchTable = true;
}
}
// We go through all the tables so that multiple visible schemas are detected
if (multipleSchemas)
{
throw new AlfrescoRuntimeException(ERR_MULTIPLE_SCHEMAS);
}
}
finally
{
try { tableRs.close(); } catch (Throwable e) {e.printStackTrace(); }
}
if (newPatchTable)
{
Statement stmt = connection.createStatement();
try
{
ResultSet rs = stmt.executeQuery("select count(id) from alf_applied_patch");
rs.next();
int count = rs.getInt(1);
return count;
}
catch (SQLException e)
{
// This should work at least and is probably an indication of the user viewing multiple schemas
throw new AlfrescoRuntimeException(ERR_MULTIPLE_SCHEMAS);
}
finally
{
try { stmt.close(); } catch (Throwable e) {}
}
}
else if (oldPatchTable)
{
// found the old style table name
Statement stmt = connection.createStatement();
try
{
ResultSet rs = stmt.executeQuery("select count(id) from applied_patch");
rs.next();
int count = rs.getInt(1);
return count;
}
finally
{
try { stmt.close(); } catch (Throwable e) {}
}
}
else
{
// The applied patches table is not present
throw new NoSchemaException();
}
}
/**
* @return Returns the name of the applied patch table, or <tt>null</tt> if the table doesn't exist
*/
private String getAppliedPatchTableName(Connection connection) throws Exception
{
Statement stmt = connection.createStatement();
try
{
stmt.executeQuery("select * from alf_applied_patch");
return "alf_applied_patch";
}
catch (Throwable e)
{
// we'll try another table name
}
finally
{
try { stmt.close(); } catch (Throwable e) {}
}
// for pre-1.4 databases, the table was named differently
stmt = connection.createStatement();
try
{
stmt.executeQuery("select * from applied_patch");
return "applied_patch";
}
catch (Throwable e)
{
// It is not there
return null;
}
finally
{
try { stmt.close(); } catch (Throwable e) {}
}
}
/**
* @return Returns the number of applied patches
*/
private boolean didPatchSucceed(Connection connection, String patchId, boolean alternative) throws Exception
{
String patchTableName = getAppliedPatchTableName(connection);
if (patchTableName == null)
{
// Table doesn't exist, yet
return false;
}
Statement stmt = connection.createStatement();
try
{
ResultSet rs = stmt.executeQuery("select succeeded, was_executed from " + patchTableName + " where id = '" + patchId + "'");
if (!rs.next())
{
return false;
}
boolean succeeded = rs.getBoolean(1);
boolean wasExecuted = rs.getBoolean(2);
if (alternative)
{
return succeeded && wasExecuted;
}
else
{
return succeeded;
}
}
finally
{
try { stmt.close(); } catch (Throwable e) {}
}
}
/**
* Finds the <b>version.properties</b> file and determines the installed <b>version.schema</b>.<br>
* The only way to determine the original installed schema number is by quering the for the minimum value in
* <b>alf_applied_patch.applied_to_schema</b>. This might not work if an upgrade is attempted straight from
* Alfresco v1.0!
*
* @return the installed schema number or <tt>-1</tt> if the installation is new.
*/
private int getInstalledSchemaNumber(Connection connection) throws Exception
{
Statement stmt = connection.createStatement();
try
{
ResultSet rs = stmt.executeQuery(
"select min(applied_to_schema) from alf_applied_patch where applied_to_schema > -1");
if (!rs.next())
{
// Nothing in the table
return -1;
}
if (rs.getObject(1) == null)
{
// Nothing in the table
return -1;
}
int installedSchema = rs.getInt(1);
return installedSchema;
}
finally
{
try { stmt.close(); } catch (Throwable e) {}
}
}
private static class LockFailedException extends Exception
{
private static final long serialVersionUID = -6676398230191205456L;
}
/**
* Records that the bootstrap process has started
*/
private synchronized void setBootstrapStarted(Connection connection) throws Exception
{
// Create the marker table
Statement stmt = connection.createStatement();
try
{
stmt.executeUpdate("create table alf_bootstrap_lock (charval CHAR(1) NOT NULL)");
// Success
return;
}
catch (Throwable e)
{
// We throw a well-known exception to be handled by retrying code if required
throw new LockFailedException();
}
finally
{
try { stmt.close(); } catch (Throwable e) {}
}
}
/**
* Records that the bootstrap process has finished
*/
private void setBootstrapCompleted(Connection connection) throws Exception
{
// Create the marker table
Statement stmt = connection.createStatement();
try
{
stmt.executeUpdate("drop table alf_bootstrap_lock");
}
catch (Throwable e)
{
// Table exists
throw AlfrescoRuntimeException.create(ERR_PREVIOUS_FAILED_BOOTSTRAP);
}
finally
{
try { stmt.close(); } catch (Throwable e) {}
}
}
/**
* Builds the schema from scratch or applies the necessary patches to the schema.
*/
private boolean updateSchema(Configuration cfg, Session session, Connection connection) throws Exception
{
boolean create = false;
try
{
countAppliedPatches(cfg, connection);
}
catch (NoSchemaException e)
{
create = true;
}
// Get the dialect
final Dialect dialect = Dialect.getDialect(cfg.getProperties());
String dialectStr = dialect.getClass().getSimpleName();
// Initialise Activiti DB, using an unclosable connection.
initialiseActivitiDBSchema(new UnclosableConnection(connection));
if (create)
{
// execute pre-create scripts (not patches)
for (String scriptUrl : this.preCreateScriptUrls)
{
executeScriptUrl(cfg, connection, scriptUrl);
}
// Build and execute changes generated by Hibernate
File tempFile = null;
Writer writer = null;
try
{
DatabaseMetadata metadata = new DatabaseMetadata(connection, dialect);
String[] sqls = cfg.generateSchemaUpdateScript(dialect, metadata);
if (sqls.length > 0)
{
tempFile = TempFileProvider.createTempFile("AlfrescoSchema-" + dialectStr + "-Update-", ".sql");
writer = new BufferedWriter(new FileWriter(tempFile));
for (String sql : sqls)
{
writer.append(sql);
writer.append(";\n");
}
try {writer.close();} catch (Throwable e) {}
executeScriptFile(cfg, connection, tempFile, null);
}
}
finally
{
if (writer != null)
{
try {writer.close();} catch (Throwable e) {}
}
}
// execute post-create scripts (not patches)
for (String scriptUrl : this.postCreateScriptUrls)
{
executeScriptUrl(cfg, connection, scriptUrl);
}
}
else
{
// Check for scripts that must have been run
checkSchemaPatchScripts(cfg, connection, validateUpdateScriptPatches, false);
// Execute any pre-auto-update scripts
checkSchemaPatchScripts(cfg, connection, preUpdateScriptPatches, true);
// Build and execute changes generated by Hibernate
File tempFile = null;
Writer writer = null;
try
{
DatabaseMetadata metadata = new DatabaseMetadata(connection, dialect);
String[] sqls = cfg.generateSchemaUpdateScript(dialect, metadata);
if (sqls.length > 0)
{
tempFile = TempFileProvider.createTempFile("AlfrescoSchema-" + dialectStr + "-Update-", ".sql");
writer = new BufferedWriter(new FileWriter(tempFile));
for (String sql : sqls)
{
writer.append(sql);
writer.append(";\n");
}
}
}
finally
{
if (writer != null)
{
try {writer.close();} catch (Throwable e) {}
}
}
// execute if there were changes raised by Hibernate
if (tempFile != null)
{
executeScriptFile(cfg, connection, tempFile, null);
}
// Execute any post-auto-update scripts
checkSchemaPatchScripts(cfg, connection, postUpdateScriptPatches, true);
}
return create;
}
/**
* Initialises the Activiti DB schema, if not present it's created, if
* present it is upgraded appropriately if necessary.
*
* @param connection Connection to use the initialise DB schema
*/
private void initialiseActivitiDBSchema(Connection connection)
{
// create instance of activiti engine to initialise schema
ProcessEngine engine = null;
ProcessEngineConfiguration engineConfig = ProcessEngineConfiguration.createStandaloneProcessEngineConfiguration();
try
{
// build the engine
engine = engineConfig.setDataSource(dataSource).
setDatabaseSchemaUpdate("none").
setProcessEngineName("activitiBootstrapEngine").
setHistory("full").
setJobExecutorActivate(false).
buildProcessEngine();
// create or upgrade the DB schema
engine.getManagementService().databaseSchemaUpgrade(connection, null, null);
}
finally
{
if (engine != null)
{
// close the process engine
engine.close();
}
}
}
/**
* Check that the necessary scripts have been executed against the database
*/
private void checkSchemaPatchScripts(
Configuration cfg,
Connection connection,
List<SchemaUpgradeScriptPatch> scriptPatches,
boolean apply) throws Exception
{
// first check if there have been any applied patches
int appliedPatchCount = countAppliedPatches(cfg, connection);
if (appliedPatchCount == 0)
{
// This is a new schema, so upgrade scripts are irrelevant
// and patches will not have been applied yet
return;
}
// Retrieve the first installed schema number
int installedSchema = getInstalledSchemaNumber(connection);
nextPatch:
for (SchemaUpgradeScriptPatch patch : scriptPatches)
{
final String patchId = patch.getId();
final String scriptUrl = patch.getScriptUrl();
// Check if any of the alternative patches were executed
List<Patch> alternatives = patch.getAlternatives();
for (Patch alternativePatch : alternatives)
{
String alternativePatchId = alternativePatch.getId();
boolean alternativeSucceeded = didPatchSucceed(connection, alternativePatchId, true);
if (alternativeSucceeded)
{
continue nextPatch;
}
}
// check if the script was successfully executed
boolean wasSuccessfullyApplied = didPatchSucceed(connection, patchId, false);
if (wasSuccessfullyApplied)
{
// Either the patch was executed before or the system was bootstrapped
// with the patch bean present.
continue;
}
else if (!patch.applies(installedSchema))
{
// Patch does not apply to the installed schema number
continue;
}
else if (!apply)
{
// the script was not run and may not be run automatically
throw AlfrescoRuntimeException.create(ERR_SCRIPT_NOT_RUN, scriptUrl);
}
// it wasn't run and it can be run now
executeScriptUrl(cfg, connection, scriptUrl);
}
}
private void executeScriptUrl(Configuration cfg, Connection connection, String scriptUrl) throws Exception
{
Dialect dialect = Dialect.getDialect(cfg.getProperties());
String dialectStr = dialect.getClass().getSimpleName();
InputStream scriptInputStream = getScriptInputStream(dialect.getClass(), scriptUrl);
// check that it exists
if (scriptInputStream == null)
{
throw AlfrescoRuntimeException.create(ERR_SCRIPT_NOT_FOUND, scriptUrl);
}
// write the script to a temp location for future and failure reference
File tempFile = null;
try
{
tempFile = TempFileProvider.createTempFile("AlfrescoSchema-" + dialectStr + "-Update-", ".sql");
ContentWriter writer = new FileContentWriter(tempFile);
writer.putContent(scriptInputStream);
}
finally
{
try { scriptInputStream.close(); } catch (Throwable e) {} // usually a duplicate close
}
// now execute it
String dialectScriptUrl = scriptUrl.replaceAll(PLACEHOLDER_DIALECT, dialect.getClass().getName());
// Replace the script placeholders
executeScriptFile(cfg, connection, tempFile, dialectScriptUrl);
}
/**
* Replaces the dialect placeholder in the resource URL and attempts to find a file for
* it. If not found, the dialect hierarchy will be walked until a compatible resource is
* found. This makes it possible to have resources that are generic to all dialects.
*
* @return The Resource, otherwise null
*/
private Resource getDialectResource(Class dialectClass, String resourceUrl)
{
// replace the dialect placeholder
String dialectResourceUrl = resolveDialectUrl(dialectClass, resourceUrl);
// get a handle on the resource
Resource resource = rpr.getResource(dialectResourceUrl);
if (!resource.exists())
{
// it wasn't found. Get the superclass of the dialect and try again
Class superClass = dialectClass.getSuperclass();
if (Dialect.class.isAssignableFrom(superClass))
{
// we still have a Dialect - try again
return getDialectResource(superClass, resourceUrl);
}
else
{
// we have exhausted all options
return null;
}
}
else
{
// we have a handle to it
return resource;
}
}
/**
* Takes resource URL containing the {@link SchemaBootstrap#PLACEHOLDER_DIALECT dialect placeholder text}
* and substitutes the placeholder with the name of the given dialect's class.
* <p/>
* For example:
* <pre>
* resolveDialectUrl(MySQLInnoDBDialect.class, "classpath:alfresco/db/${db.script.dialect}/myfile.xml")
* </pre>
* would give the following String:
* <pre>
* classpath:alfresco/db/org.hibernate.dialect.MySQLInnoDBDialect/myfile.xml
* </pre>
*
* @param dialectClass
* @param resourceUrl
* @return
*/
private String resolveDialectUrl(Class dialectClass, String resourceUrl)
{
return resourceUrl.replaceAll(PLACEHOLDER_DIALECT, dialectClass.getName());
}
/**
* Replaces the dialect placeholder in the script URL and attempts to find a file for
* it. If not found, the dialect hierarchy will be walked until a compatible script is
* found. This makes it possible to have scripts that are generic to all dialects.
*
* @return Returns an input stream onto the script, otherwise null
*/
private InputStream getScriptInputStream(Class dialectClazz, String scriptUrl) throws Exception
{
return getDialectResource(dialectClazz, scriptUrl).getInputStream();
}
/**
* @param cfg the Hibernate configuration
* @param connection the DB connection to use
* @param scriptFile the file containing the statements
* @param scriptUrl the URL of the script to report. If this is null, the script
* is assumed to have been auto-generated.
*/
private void executeScriptFile(
Configuration cfg,
Connection connection,
File scriptFile,
String scriptUrl) throws Exception
{
final Dialect dialect = Dialect.getDialect(cfg.getProperties());
StringBuilder executedStatements = executedStatementsThreadLocal.get();
if (executedStatements == null)
{
// Validate the schema, pre-upgrade
validateSchema("Alfresco-{0}-Validation-Pre-Upgrade-{1}-");
// Dump the normalized, pre-upgrade Alfresco schema. We keep the file for later reporting.
xmlPreSchemaOutputFile34 = dumpSchema34(
this.dialect,
TempFileProvider.createTempFile(
"AlfrescoSchema-" + this.dialect.getClass().getSimpleName() + "-",
"-Startup.xml").getPath(),
"Failed to dump normalized, pre-upgrade schema to file.");
dumpSchema("pre-upgrade");
// There is no lock at this stage. This process can fall out if the lock can't be applied.
setBootstrapStarted(connection);
executedStatements = new StringBuilder(8094);
executedStatementsThreadLocal.set(executedStatements);
}
if (scriptUrl == null)
{
LogUtil.info(logger, MSG_EXECUTING_GENERATED_SCRIPT, scriptFile);
}
else
{
LogUtil.info(logger, MSG_EXECUTING_COPIED_SCRIPT, scriptFile, scriptUrl);
}
InputStream scriptInputStream = new FileInputStream(scriptFile);
BufferedReader reader = new BufferedReader(new InputStreamReader(scriptInputStream, "UTF-8"));
try
{
int line = 0;
// loop through all statements
StringBuilder sb = new StringBuilder(1024);
String fetchVarName = null;
String fetchColumnName = null;
boolean doBatch = false;
int batchUpperLimit = 0;
int batchSize = 1;
Map<String, Object> varAssignments = new HashMap<String, Object>(13);
// Special variable assignments:
if (dialect instanceof PostgreSQLDialect)
{
// Needs 1/0 for true/false
varAssignments.put("true", "true");
varAssignments.put("false", "false");
varAssignments.put("TRUE", "TRUE");
varAssignments.put("FALSE", "FALSE");
}
else
{
// Needs true/false as strings
varAssignments.put("true", "1");
varAssignments.put("false", "0");
varAssignments.put("TRUE", "1");
varAssignments.put("FALSE", "0");
}
long now = System.currentTimeMillis();
varAssignments.put("now", new Long(now).toString());
varAssignments.put("NOW", new Long(now).toString());
while(true)
{
String sqlOriginal = reader.readLine();
line++;
if (sqlOriginal == null)
{
// nothing left in the file
break;
}
// trim it
String sql = sqlOriginal.trim();
// Check of includes
if (sql.startsWith("--INCLUDE:"))
{
if (sb.length() > 0)
{
// This can only be set before a new SQL statement
throw AlfrescoRuntimeException.create(ERR_STATEMENT_INCLUDE_BEFORE_SQL, (line - 1), scriptUrl);
}
String includedScriptUrl = sql.substring(10, sql.length());
// Execute the script in line
executeScriptUrl(cfg, connection, includedScriptUrl);
}
// Check for variable assignment
else if (sql.startsWith("--ASSIGN:"))
{
if (sb.length() > 0)
{
// This can only be set before a new SQL statement
throw AlfrescoRuntimeException.create(ERR_STATEMENT_VAR_ASSIGNMENT_BEFORE_SQL, (line - 1), scriptUrl);
}
String assignStr = sql.substring(9, sql.length());
String[] assigns = assignStr.split("=");
if (assigns.length != 2 || assigns[0].length() == 0 || assigns[1].length() == 0)
{
throw AlfrescoRuntimeException.create(ERR_STATEMENT_VAR_ASSIGNMENT_FORMAT, (line - 1), scriptUrl);
}
fetchVarName = assigns[0];
fetchColumnName = assigns[1];
continue;
}
// Handle looping control
else if (sql.startsWith("--FOREACH"))
{
// --FOREACH table.column batch.size.property
String[] args = sql.split("[ \\t]+");
int sepIndex;
if (args.length == 3 && (sepIndex = args[1].indexOf('.')) != -1)
{
doBatch = true;
// Select the upper bound of the table column
String stmt = "SELECT MAX(" + args[1].substring(sepIndex+1) + ") AS upper_limit FROM " + args[1].substring(0, sepIndex);
Object fetchedVal = executeStatement(connection, stmt, "upper_limit", false, line, scriptFile);
if (fetchedVal instanceof Number)
{
batchUpperLimit = ((Number)fetchedVal).intValue();
// Read the batch size from the named property
String batchSizeString = globalProperties.getProperty(args[2]);
// Fall back to the default property
if (batchSizeString == null)
{
batchSizeString = globalProperties.getProperty(PROPERTY_DEFAULT_BATCH_SIZE);
}
batchSize = batchSizeString == null ? 10000 : Integer.parseInt(batchSizeString);
}
}
continue;
}
// Allow transaction delineation
else if (sql.startsWith("--BEGIN TXN"))
{
connection.setAutoCommit(false);
continue;
}
else if (sql.startsWith("--END TXN"))
{
connection.commit();
connection.setAutoCommit(true);
continue;
}
// Check for comments
if (sql.length() == 0 ||
sql.startsWith( "--" ) ||
sql.startsWith( "//" ) ||
sql.startsWith( "/*" ) )
{
if (sb.length() > 0)
{
// we have an unterminated statement
throw AlfrescoRuntimeException.create(ERR_STATEMENT_TERMINATOR, (line - 1), scriptUrl);
}
// there has not been anything to execute - it's just a comment line
continue;
}
// have we reached the end of a statement?
boolean execute = false;
boolean optional = false;
if (sql.endsWith(";"))
{
sql = sql.substring(0, sql.length() - 1);
execute = true;
optional = false;
}
else if (sql.endsWith("(optional)") || sql.endsWith("(OPTIONAL)"))
{
// Get the end of statement
int endIndex = sql.lastIndexOf(';');
if (endIndex > -1)
{
sql = sql.substring(0, endIndex);
execute = true;
optional = true;
}
else
{
// Ends with "(optional)" but there is no semi-colon.
// Just take it at face value and probably fail.
}
}
// Add newline
if (sb.length() > 0)
{
sb.append("\n");
}
// Add leading whitespace for formatting
int whitespaceCount = sqlOriginal.indexOf(sql);
for (int i = 0; i < whitespaceCount; i++)
{
sb.append(" ");
}
// append to the statement being built up
sb.append(sql);
// execute, if required
if (execute)
{
// Now substitute and execute the statement the appropriate number of times
String unsubstituted = sb.toString();
for(int lowerBound = 0; lowerBound <= batchUpperLimit; lowerBound += batchSize)
{
sql = unsubstituted;
// Substitute in the next pair of range parameters
if (doBatch)
{
varAssignments.put("LOWERBOUND", String.valueOf(lowerBound));
varAssignments.put("UPPERBOUND", String.valueOf(lowerBound + batchSize - 1));
}
// Perform variable replacement using the ${var} format
for (Map.Entry<String, Object> entry : varAssignments.entrySet())
{
String var = entry.getKey();
Object val = entry.getValue();
sql = sql.replaceAll("\\$\\{" + var + "\\}", val.toString());
}
// Handle the 0/1 values that PostgreSQL doesn't translate to TRUE
if (this.dialect != null && this.dialect instanceof PostgreSQLDialect)
{
sql = sql.replaceAll("\\$\\{TRUE\\}", "TRUE");
}
else
{
sql = sql.replaceAll("\\$\\{TRUE\\}", "1");
}
if (this.dialect != null && this.dialect instanceof MySQLInnoDBDialect)
{
// note: enable bootstrap on MySQL 5.5 (eg. for auto-generated SQL, such as JBPM)
sql = sql.replaceAll("(?i)TYPE=InnoDB", "ENGINE=InnoDB");
}
Object fetchedVal = executeStatement(connection, sql, fetchColumnName, optional, line, scriptFile);
if (fetchVarName != null && fetchColumnName != null)
{
varAssignments.put(fetchVarName, fetchedVal);
}
}
sb.setLength(0);
fetchVarName = null;
fetchColumnName = null;
doBatch = false;
batchUpperLimit = 0;
batchSize = 1;
}
}
}
finally
{
try { reader.close(); } catch (Throwable e) {}
try { scriptInputStream.close(); } catch (Throwable e) {}
}
}
/**
* Execute the given SQL statement, absorbing exceptions that we expect during
* schema creation or upgrade.
*
* @param fetchColumnName the name of the column value to return
*/
private Object executeStatement(
Connection connection,
String sql,
String fetchColumnName,
boolean optional,
int line,
File file) throws Exception
{
StringBuilder executedStatements = executedStatementsThreadLocal.get();
if (executedStatements == null)
{
throw new IllegalArgumentException("The executedStatementsThreadLocal must be populated");
}
Statement stmt = connection.createStatement();
Object ret = null;
try
{
if (logger.isDebugEnabled())
{
LogUtil.debug(logger, MSG_EXECUTING_STATEMENT, sql);
}
boolean haveResults = stmt.execute(sql);
// Record the statement
executedStatements.append(sql).append(";\n\n");
if (haveResults && fetchColumnName != null)
{
ResultSet rs = stmt.getResultSet();
if (rs.next())
{
// Get the result value
ret = rs.getObject(fetchColumnName);
}
}
}
catch (SQLException e)
{
if (optional)
{
// it was marked as optional, so we just ignore it
LogUtil.debug(logger, MSG_OPTIONAL_STATEMENT_FAILED, sql, e.getMessage(), file.getAbsolutePath(), line);
}
else
{
LogUtil.error(logger, ERR_STATEMENT_FAILED, sql, e.getMessage(), file.getAbsolutePath(), line);
throw e;
}
}
finally
{
try { stmt.close(); } catch (Throwable e) {}
}
return ret;
}
/**
* Performs dialect-specific checking. This includes checking for InnoDB, dumping the dialect being used
* as well as setting any runtime, dialect-specific properties.
*/
private void checkDialect(Dialect dialect)
{
Class dialectClazz = dialect.getClass();
LogUtil.info(logger, MSG_DIALECT_USED, dialectClazz.getName());
if (dialectClazz.equals(MySQLDialect.class) || dialectClazz.equals(MySQL5Dialect.class))
{
LogUtil.error(logger, ERR_DIALECT_SHOULD_USE, dialectClazz.getName(), MySQLInnoDBDialect.class.getName());
throw AlfrescoRuntimeException.create(WARN_DIALECT_UNSUPPORTED, dialectClazz.getName());
}
else if (dialectClazz.equals(HSQLDialect.class))
{
LogUtil.info(logger, WARN_DIALECT_HSQL);
}
else if (dialectClazz.equals(DerbyDialect.class))
{
LogUtil.info(logger, WARN_DIALECT_DERBY);
}
else if (dialectClazz.equals(Oracle9iDialect.class) || dialectClazz.equals(Oracle10gDialect.class))
{
LogUtil.error(logger, ERR_DIALECT_SHOULD_USE, dialectClazz.getName(), AlfrescoOracle9Dialect.class.getName());
throw AlfrescoRuntimeException.create(WARN_DIALECT_UNSUPPORTED, dialectClazz.getName());
}
else if (dialectClazz.equals(OracleDialect.class) || dialectClazz.equals(Oracle9Dialect.class))
{
LogUtil.error(logger, ERR_DIALECT_SHOULD_USE, dialectClazz.getName(), AlfrescoOracle9Dialect.class.getName());
throw AlfrescoRuntimeException.create(WARN_DIALECT_UNSUPPORTED, dialectClazz.getName());
}
int maxStringLength = SchemaBootstrap.DEFAULT_MAX_STRING_LENGTH;
int serializableType = SerializableTypeHandler.getSerializableType();
// Adjust the maximum allowable String length according to the dialect
if (dialect instanceof AlfrescoSQLServerDialect)
{
// string_value nvarchar(1024) null,
// serializable_value image null,
maxStringLength = SchemaBootstrap.DEFAULT_MAX_STRING_LENGTH;
}
else if (dialect instanceof AlfrescoSybaseAnywhereDialect)
{
// string_value text null,
// serializable_value varbinary(8192) null,
maxStringLength = Integer.MAX_VALUE;
}
else if (dialect instanceof DB2Dialect)
{
// string_value varchar(1024),
// serializable_value varchar(8192) for bit data,
maxStringLength = SchemaBootstrap.DEFAULT_MAX_STRING_LENGTH;
serializableType = Types.BLOB;
}
else if (dialect instanceof HSQLDialect)
{
// string_value varchar(1024),
// serializable_value varbinary(8192),
maxStringLength = SchemaBootstrap.DEFAULT_MAX_STRING_LENGTH;
}
else if (dialect instanceof MySQLInnoDBDialect)
{
// string_value text,
// serializable_value blob,
maxStringLength = Integer.MAX_VALUE;
}
else if (dialect instanceof AlfrescoOracle9Dialect)
{
// string_value varchar2(1024 char),
// serializable_value blob,
maxStringLength = SchemaBootstrap.DEFAULT_MAX_STRING_LENGTH;
}
else if (dialect instanceof PostgreSQLDialect)
{
// string_value varchar(1024),
// serializable_value bytea,
maxStringLength = SchemaBootstrap.DEFAULT_MAX_STRING_LENGTH;
}
SchemaBootstrap.setMaxStringLength(maxStringLength);
SerializableTypeHandler.setSerializableType(serializableType);
// Now override the maximum string length if it was set directly
if (maximumStringLength > 0)
{
SchemaBootstrap.setMaxStringLength(maximumStringLength);
}
}
@Override
protected synchronized void onBootstrap(ApplicationEvent event)
{
// Use the application context to load resources
rpr = (ApplicationContext)event.getSource();
// do everything in a transaction
Session session = getSessionFactory().openSession();
Connection connection = null;
try
{
// make sure that we AUTO-COMMIT
connection = dataSource.getConnection();
connection.setAutoCommit(true);
Configuration cfg = localSessionFactory.getConfiguration();
// Check and dump the dialect being used
checkDialect(this.dialect);
// Ensure that our static connection provider is used
String defaultConnectionProviderFactoryClass = cfg.getProperty(Environment.CONNECTION_PROVIDER);
cfg.setProperty(Environment.CONNECTION_PROVIDER, SchemaBootstrapConnectionProvider.class.getName());
SchemaBootstrapConnectionProvider.setBootstrapConnection(connection);
// Update the schema, if required.
if (updateSchema)
{
// Retries are required here as the DB lock will be applied lazily upon first statement execution.
// So if the schema is up to date (no statements executed) then the LockFailException cannot be
// thrown. If it is thrown, the the update needs to be rerun as it will probably generate no SQL
// statements the second time around.
boolean updatedSchema = false;
boolean createdSchema = false;
for (int i = 0; i < schemaUpdateLockRetryCount; i++)
{
try
{
createdSchema = updateSchema(cfg, session, connection);
updatedSchema = true;
break;
}
catch (LockFailedException e)
{
try { this.wait(schemaUpdateLockRetryWaitSeconds * 1000L); } catch (InterruptedException ee) {}
}
}
if (!updatedSchema)
{
// The retries were exceeded
throw new AlfrescoRuntimeException(ERR_PREVIOUS_FAILED_BOOTSTRAP);
}
// Copy the executed statements to the output file
File schemaOutputFile = null;
if (schemaOuputFilename != null)
{
schemaOutputFile = new File(schemaOuputFilename);
}
else
{
schemaOutputFile = TempFileProvider.createTempFile(
"AlfrescoSchema-" + this.dialect.getClass().getSimpleName() + "-All_Statements-",
".sql");
}
StringBuilder executedStatements = executedStatementsThreadLocal.get();
if (executedStatements == null)
{
LogUtil.info(logger, MSG_NO_CHANGES);
}
else
{
FileContentWriter writer = new FileContentWriter(schemaOutputFile);
writer.setEncoding("UTF-8");
String executedStatementsStr = executedStatements.toString();
writer.putContent(executedStatementsStr);
LogUtil.info(logger, MSG_ALL_STATEMENTS, schemaOutputFile.getPath());
}
if (! createdSchema)
{
// verify that all patches have been applied correctly
checkSchemaPatchScripts(cfg, connection, validateUpdateScriptPatches, false); // check scripts
checkSchemaPatchScripts(cfg, connection, preUpdateScriptPatches, false); // check scripts
checkSchemaPatchScripts(cfg, connection, postUpdateScriptPatches, false); // check scripts
}
if (executedStatements != null)
{
// Remove the flag indicating a running bootstrap
setBootstrapCompleted(connection);
}
// Report normalized dumps
if (executedStatements != null)
{
// Validate the schema, post-upgrade
validateSchema("Alfresco-{0}-Validation-Post-Upgrade-{1}-");
// Dump the normalized, post-upgrade Alfresco schema.
File xmlPostSchemaOutputFile34 = dumpSchema34(
this.dialect,
TempFileProvider.createTempFile(
"AlfrescoSchema-" + this.dialect.getClass().getSimpleName() + "-",
".xml").getPath(),
"Failed to dump normalized, post-upgrade schema to file.");
// 4.0+ schema dump
dumpSchema("post-upgrade");
if (createdSchema)
{
// This is a new schema
if (xmlPostSchemaOutputFile34 != null)
{
LogUtil.info(logger, MSG_NORMALIZED_SCHEMA, xmlPostSchemaOutputFile34.getPath());
}
}
else
{
// We upgraded, so have to report pre- and post- schema dumps
if (xmlPreSchemaOutputFile34 != null)
{
LogUtil.info(logger, MSG_NORMALIZED_SCHEMA_PRE, xmlPreSchemaOutputFile34.getPath());
}
if (xmlPostSchemaOutputFile34 != null)
{
LogUtil.info(logger, MSG_NORMALIZED_SCHEMA_POST, xmlPostSchemaOutputFile34.getPath());
}
}
}
}
else
{
LogUtil.info(logger, MSG_BYPASSING_SCHEMA_UPDATE);
}
if (stopAfterSchemaBootstrap)
{
// We have been forced to stop, so we do one last dump of the schema and throw an exception to
// escape further startup procedures
File xmlStopSchemaOutputFile = dumpSchema34(
this.dialect,
TempFileProvider.createTempFile(
"AlfrescoSchema-" + this.dialect.getClass().getSimpleName() + "-",
"-ForcedExit.xml").getPath(),
"Failed to dump normalized, post-upgrade, forced-exit schema to file.");
// 4.0+ schema dump
dumpSchema("forced-exit");
if (xmlStopSchemaOutputFile != null)
{
LogUtil.info(logger, MSG_NORMALIZED_SCHEMA, xmlStopSchemaOutputFile);
}
LogUtil.error(logger, ERR_FORCED_STOP);
throw new BootstrapStopException();
}
// Reset the configuration
cfg.setProperty(Environment.CONNECTION_PROVIDER, defaultConnectionProviderFactoryClass);
// all done successfully
((ApplicationContext) event.getSource()).publishEvent(new SchemaAvailableEvent(this));
}
catch (BootstrapStopException e)
{
// We just let this out
throw e;
}
catch (Throwable e)
{
LogUtil.error(logger, e, ERR_UPDATE_FAILED);
if (updateSchema)
{
throw new AlfrescoRuntimeException(ERR_UPDATE_FAILED, e);
}
else
{
throw new AlfrescoRuntimeException(ERR_VALIDATION_FAILED, e);
}
}
finally
{
try
{
if (connection != null)
{
connection.close();
}
}
catch (Throwable e)
{
logger.warn("Error closing DB connection: " + e.getMessage());
}
// Remove the connection reference from the threadlocal boostrap
SchemaBootstrapConnectionProvider.setBootstrapConnection(null);
}
}
/**
* Collate differences and validation problems with the schema with respect to an appropriate
* reference schema.
*
* @param outputFileNameTemplate
* @return the number of potential problems found.
*/
public synchronized int validateSchema(String outputFileNameTemplate)
{
int totalProblems = 0;
// Discover available reference files (e.g. for prefixes alf_, avm_ etc.)
// and process each in turn.
for (String schemaReferenceUrl : schemaReferenceUrls)
{
Resource referenceResource = getDialectResource(dialect.getClass(), schemaReferenceUrl);
if (referenceResource == null || !referenceResource.exists())
{
String resourceUrl = resolveDialectUrl(dialect.getClass(), schemaReferenceUrl);
LogUtil.debug(logger, DEBUG_SCHEMA_COMP_NO_REF_FILE, resourceUrl);
}
else
{
// Validate schema against each reference file
int problems = validateSchema(referenceResource, outputFileNameTemplate);
totalProblems += problems;
}
}
// Return number of problems found across all reference files.
return totalProblems;
}
private int validateSchema(Resource referenceResource, String outputFileNameTemplate)
{
try
{
return attemptValidateSchema(referenceResource, outputFileNameTemplate);
}
catch (Throwable e)
{
if (logger.isErrorEnabled())
{
logger.error("Unable to validate database schema.", e);
}
return 0;
}
}
private int attemptValidateSchema(Resource referenceResource, String outputFileNameTemplate)
{
Date startTime = new Date();
InputStream is = null;
try
{
is = new BufferedInputStream(referenceResource.getInputStream());
}
catch (IOException e)
{
throw new RuntimeException("Unable to open schema reference file: " + referenceResource);
}
XMLToSchema xmlToSchema = new XMLToSchema(is);
xmlToSchema.parse();
Schema reference = xmlToSchema.getSchema();
ExportDb exporter = new ExportDb(dataSource, dialect, descriptorService);
// Ensure that the database objects we're validating are filtered
// by the same prefix as the reference file.
exporter.setNamePrefix(reference.getDbPrefix());
exporter.execute();
Schema target = exporter.getSchema();
SchemaComparator schemaComparator = new SchemaComparator(reference, target, dialect);
schemaComparator.validateAndCompare();
Results results = schemaComparator.getComparisonResults();
Object[] outputFileNameParams = new Object[]
{
dialect.getClass().getSimpleName(),
reference.getDbPrefix()
};
String outputFileName = MessageFormat.format(outputFileNameTemplate, outputFileNameParams);
File outputFile = TempFileProvider.createTempFile(outputFileName, ".txt");
PrintWriter pw = null;
try
{
pw = new PrintWriter(outputFile, SchemaComparator.CHAR_SET);
}
catch (FileNotFoundException error)
{
throw new RuntimeException("Unable to open file for writing: " + outputFile);
}
catch (UnsupportedEncodingException error)
{
throw new RuntimeException("Unsupported char set: " + SchemaComparator.CHAR_SET, error);
}
// Populate the file with details of the comparison's results.
for (Result result : results)
{
pw.print(result.describe());
pw.print(SchemaComparator.LINE_SEPARATOR);
}
pw.close();
if (results.size() == 0)
{
LogUtil.info(logger, INFO_SCHEMA_COMP_ALL_OK, referenceResource);
}
else
{
int numProblems = results.size();
LogUtil.warn(logger, WARN_SCHEMA_COMP_PROBLEMS_FOUND, numProblems, outputFile);
}
Date endTime = new Date();
long durationMillis = endTime.getTime() - startTime.getTime();
LogUtil.debug(logger, DEBUG_SCHEMA_COMP_TIME_TAKEN, durationMillis);
return results.size();
}
/**
* Older schema dump tool. Left here for the moment but is essentially duplicated
* by the newer XML dumper which is used in automatic diff/validation of the schema
* against a known good reference.
*
* @return Returns the file that was written to or <tt>null</tt> if it failed
*/
private File dumpSchema34(Dialect dialect, String fileName, String err)
{
File xmlSchemaOutputFile = new File(fileName);
try
{
Main xmlSchemaOutputMain = new Main(dataSource, dialect);
xmlSchemaOutputMain.execute(xmlSchemaOutputFile);
}
catch (Throwable e)
{
xmlSchemaOutputFile = null;
// Don't fail the upgrade on this
if (logger.isDebugEnabled())
{
logger.debug(err, e);
}
else
{
logger.error(err + " Error: " + e.getMessage());
}
}
return xmlSchemaOutputFile;
}
/**
* Produces schema dump in XML format: this is performed pre- and post-upgrade (i.e. if
* changes are made to the schema) and can made upon demand via JMX.
*
* @return List of output files.
*/
public List<File> dumpSchema()
{
return dumpSchema("", null);
}
/**
* Produces schema dump in XML format: this is performed pre- and post-upgrade (i.e. if
* changes are made to the schema) and can made upon demand via JMX.
*
* @param dbPrefixes Array of database object prefixes to produce the dump for, e.g. "alf_".
* @return List of output files.
*/
public List<File> dumpSchema(String[] dbPrefixes)
{
return dumpSchema("", dbPrefixes);
}
/**
* Dumps the DB schema to temporary file(s), named similarly to:
* <pre>
* Alfresco-schema-DialectName-whenDumped-dbPrefix-23498732.xml
* </pre>
* Where the digits serve to create a unique temp file name. If whenDumped is empty or null,
* then the output is similar to:
* <pre>
* Alfresco-schema-DialectName-dbPrefix-23498732.xml
* </pre>
* If dbPrefixes is null, then the default list is used (see {@link MultiFileDumper#DEFAULT_PREFIXES})
* The dump files' paths are logged at info level.
*
* @param whenDumped
* @param dbPrefixes Array of database object prefixes to filter by, e.g. "alf_"
* @return List of output files.
*/
private List<File> dumpSchema(String whenDumped, String[] dbPrefixes)
{
// Build a string to use as the file name template,
// e.g. "Alfresco-schema-MySQLDialect-pre-upgrade-{0}-"
StringBuilder sb = new StringBuilder(64);
sb.append("Alfresco-schema-").
append(dialect.getClass().getSimpleName());
if (whenDumped != null && whenDumped.length() > 0)
{
sb.append("-");
sb.append(whenDumped);
}
sb.append("-{0}-");
File outputDir = TempFileProvider.getTempDir();
String fileNameTemplate = sb.toString();
return dumpSchema(outputDir, fileNameTemplate, dbPrefixes);
}
/**
* Same as for {@link #dumpSchema(String, String[])} - except that the default list
* of database object prefixes is used for filtering.
*
* @see #dumpSchema(String, String[])
* @param whenDumped
* @return
*/
private List<File> dumpSchema(String whenDumped)
{
return dumpSchema(whenDumped, null);
}
private List<File> dumpSchema(File outputDir, String fileNameTemplate, String[] dbPrefixes)
{
try
{
return attemptDumpSchema(outputDir, fileNameTemplate, dbPrefixes);
}
catch (Throwable e)
{
if (logger.isErrorEnabled())
{
logger.error("Unable to dump schema to directory " + outputDir, e);
}
return null;
}
}
private List<File> attemptDumpSchema(File outputDir, String fileNameTemplate, String[] dbPrefixes)
{
DbToXMLFactory dbToXMLFactory = new MultiFileDumper.DbToXMLFactoryImpl(getApplicationContext());
MultiFileDumper dumper;
if (dbPrefixes == null)
{
dumper = new MultiFileDumper(outputDir, fileNameTemplate, dbToXMLFactory);
}
else
{
dumper = new MultiFileDumper(dbPrefixes, outputDir, fileNameTemplate, dbToXMLFactory);
}
List<File> files = dumper.dumpFiles();
for (File file : files)
{
if (logger.isInfoEnabled())
{
LogUtil.info(logger, MSG_NORMALIZED_SCHEMA, file.getAbsolutePath());
}
}
return files;
}
@Override
protected void onShutdown(ApplicationEvent event)
{
// Shut down DB, if required
Class<?> dialectClazz = this.dialect.getClass();
if (dialectClazz.equals(DerbyDialect.class))
{
try
{
DriverManager.getConnection("jdbc:derby:;shutdown=true");
}
// Derby shutdown always triggers an exception, even when clean
catch (Throwable e)
{
}
}
}
/**
* This is a workaround for the odd Spring-Hibernate interaction during configuration.
* The Hibernate code assumes that schema scripts will be generated during context
* initialization. We want to do it afterwards and have a little more control. Hence this class.
* <p>
* The connection that is used will not be closed or manipulated in any way. This class
* merely serves to give the connection to Hibernate.
*
* @author Derek Hulley
*/
public static class SchemaBootstrapConnectionProvider extends UserSuppliedConnectionProvider
{
private static ThreadLocal<Connection> threadLocalConnection = new ThreadLocal<Connection>();
public SchemaBootstrapConnectionProvider()
{
}
/**
* Set the connection for Hibernate to use for schema generation.
*/
public static void setBootstrapConnection(Connection connection)
{
threadLocalConnection.set(connection);
}
/**
* Unsets the connection.
*/
@Override
public void close()
{
// Leave the connection well alone, just remove it
threadLocalConnection.set(null);
}
/**
* Does nothing. The connection was given by a 3rd party and they can close it.
*/
@Override
public void closeConnection(Connection conn)
{
}
/**
* Does nothing.
*/
@Override
public void configure(Properties props) throws HibernateException
{
}
/**
* @see #setBootstrapConnection(Connection)
*/
@Override
public Connection getConnection()
{
return threadLocalConnection.get();
}
@Override
public boolean supportsAggressiveRelease()
{
return false;
}
}
/**
* A {@link Connection} wrapper that delegates all calls to the wrapped class
* except for the close methods, which are ignored.
*
* @author Frederik Heremans
*/
public class UnclosableConnection implements Connection
{
private Connection wrapped;
public UnclosableConnection(Connection wrappedConnection)
{
this.wrapped = wrappedConnection;
}
@Override
public boolean isWrapperFor(Class<?> iface) throws SQLException
{
return wrapped.isWrapperFor(iface);
}
@Override
public <T> T unwrap(Class<T> iface) throws SQLException
{
return wrapped.unwrap(iface);
}
@Override
public void clearWarnings() throws SQLException
{
wrapped.clearWarnings();
}
@Override
public void close() throws SQLException
{
// Ignore this call
}
@Override
public void commit() throws SQLException
{
wrapped.commit();
}
@Override
public Array createArrayOf(String typeName, Object[] elements) throws SQLException
{
return wrapped.createArrayOf(typeName, elements);
}
@Override
public Blob createBlob() throws SQLException
{
return wrapped.createBlob();
}
@Override
public Clob createClob() throws SQLException
{
return wrapped.createClob();
}
@Override
public NClob createNClob() throws SQLException
{
return wrapped.createNClob();
}
@Override
public SQLXML createSQLXML() throws SQLException
{
return wrapped.createSQLXML();
}
@Override
public Statement createStatement() throws SQLException
{
return wrapped.createStatement();
}
@Override
public Statement createStatement(int resultSetType, int resultSetConcurrency)
throws SQLException
{
return wrapped.createStatement(resultSetType, resultSetConcurrency);
}
@Override
public Statement createStatement(int resultSetType, int resultSetConcurrency,
int resultSetHoldability) throws SQLException
{
return wrapped.createStatement(resultSetType, resultSetConcurrency, resultSetHoldability);
}
@Override
public Struct createStruct(String typeName, Object[] attributes) throws SQLException
{
return wrapped.createStruct(typeName, attributes);
}
@Override
public boolean getAutoCommit() throws SQLException
{
return wrapped.getAutoCommit();
}
@Override
public String getCatalog() throws SQLException
{
return wrapped.getCatalog();
}
@Override
public Properties getClientInfo() throws SQLException
{
return wrapped.getClientInfo();
}
@Override
public String getClientInfo(String name) throws SQLException
{
return wrapped.getClientInfo(name);
}
@Override
public int getHoldability() throws SQLException
{
return wrapped.getHoldability();
}
@Override
public DatabaseMetaData getMetaData() throws SQLException
{
return wrapped.getMetaData();
}
@Override
public int getTransactionIsolation() throws SQLException
{
return wrapped.getTransactionIsolation();
}
@Override
public Map<String, Class<?>> getTypeMap() throws SQLException
{
return wrapped.getTypeMap();
}
@Override
public SQLWarning getWarnings() throws SQLException
{
return wrapped.getWarnings();
}
@Override
public boolean isClosed() throws SQLException
{
return wrapped.isClosed();
}
@Override
public boolean isReadOnly() throws SQLException
{
return wrapped.isReadOnly();
}
@Override
public boolean isValid(int timeout) throws SQLException
{
return wrapped.isValid(timeout);
}
@Override
public String nativeSQL(String sql) throws SQLException
{
return wrapped.nativeSQL(sql);
}
@Override
public CallableStatement prepareCall(String sql) throws SQLException
{
return wrapped.prepareCall(sql);
}
@Override
public CallableStatement prepareCall(String sql, int resultSetType, int resultSetConcurrency)
throws SQLException
{
return wrapped.prepareCall(sql, resultSetType, resultSetConcurrency);
}
@Override
public CallableStatement prepareCall(String sql, int resultSetType,
int resultSetConcurrency, int resultSetHoldability) throws SQLException
{
return wrapped.prepareCall(sql, resultSetType, resultSetConcurrency, resultSetHoldability);
}
@Override
public PreparedStatement prepareStatement(String sql) throws SQLException
{
return wrapped.prepareStatement(sql);
}
@Override
public PreparedStatement prepareStatement(String sql, int autoGeneratedKeys)
throws SQLException
{
return wrapped.prepareStatement(sql, autoGeneratedKeys);
}
@Override
public PreparedStatement prepareStatement(String sql, int[] columnIndexes)
throws SQLException
{
return wrapped.prepareStatement(sql, columnIndexes);
}
@Override
public PreparedStatement prepareStatement(String sql, String[] columnNames)
throws SQLException
{
return wrapped.prepareStatement(sql, columnNames);
}
@Override
public PreparedStatement prepareStatement(String sql, int resultSetType,
int resultSetConcurrency) throws SQLException
{
return wrapped.prepareStatement(sql, resultSetType, resultSetConcurrency);
}
@Override
public PreparedStatement prepareStatement(String sql, int resultSetType,
int resultSetConcurrency, int resultSetHoldability) throws SQLException
{
return wrapped.prepareStatement(sql, resultSetType, resultSetConcurrency, resultSetHoldability);
}
@Override
public void releaseSavepoint(Savepoint savepoint) throws SQLException
{
wrapped.releaseSavepoint(savepoint);
}
@Override
public void rollback() throws SQLException
{
wrapped.rollback();
}
@Override
public void rollback(Savepoint savepoint) throws SQLException
{
wrapped.rollback(savepoint);
}
@Override
public void setAutoCommit(boolean autoCommit) throws SQLException
{
wrapped.setAutoCommit(autoCommit);
}
@Override
public void setCatalog(String catalog) throws SQLException
{
wrapped.setCatalog(catalog);
}
@Override
public void setClientInfo(Properties properties) throws SQLClientInfoException
{
wrapped.setClientInfo(properties);
}
@Override
public void setClientInfo(String name, String value) throws SQLClientInfoException
{
wrapped.setClientInfo(name, value);
}
@Override
public void setHoldability(int holdability) throws SQLException
{
wrapped.setHoldability(holdability);
}
@Override
public void setReadOnly(boolean readOnly) throws SQLException
{
wrapped.setReadOnly(readOnly);
}
@Override
public Savepoint setSavepoint() throws SQLException
{
return wrapped.setSavepoint();
}
@Override
public Savepoint setSavepoint(String name) throws SQLException
{
return wrapped.setSavepoint(name);
}
@Override
public void setTransactionIsolation(int level) throws SQLException
{
wrapped.setTransactionIsolation(level);
}
@Override
public void setTypeMap(Map<String, Class<?>> map) throws SQLException
{
wrapped.setTypeMap(map);
}
}
}