SREE repository corruption

From PTAGISWiki

Marton says that this is how we are doing the calls:

For reference here are the commands used to programmatically modify/delete/create a report template:

 RepletRegistry reg = RepletRegistry.getRegistry();
 AnalyticRepository re = (new SreeEnv()).getRepletRepository();

Deletes an old report

 // 1. removes the physical file
 File fOrgTemplate = new File(templatePath);
 fOrgTemplate.delete();
 // 2. removes the registry entry
 reg.removeReplet(fullReportName);
 reg.save();
 // 3. removes the schedule associated with this report if any
 ScheduleTask theTask = re.getScheduleTask(taskName);
 if (theTask!=null) { re.removeScheduleTask(m_oPrincipal, taskName); }

Adds a new report

 // 4. creates the actual .srt template file called saveTemplateName
 // 5. adds the sql command as a replet request
 RepletRequest theRequest = new RepletRequest(RepletRequest.INIT);
 theRequest.setParameter("sql", sReportSql);
 // 6. registers the replet
 reg.register(fullReportName, "inetsoft.sree.TemplateReplet", "", null,
 "default", 1, theRequest, null);
 reg.setTemplate(fullReportName, saveTemplateName);
 reg.save();

Charvi: The code should be wrapped in an invoked thread.

import inetsoft.report.internal.*;
import inetsoft.util.*;

threadGroup = new ThreadGroup("Report Server");
KeyRing.setKey(threadGroup,sreeHome);
Tool.invokeThread(threadGroup, new Runnable()){
  // put the code here that accesses the repository
  // put everything from getregistry to save in here
  // the whole loop
}

Should make sure that you get a global instance.

Marton: this sounds like it will work in a single machine, but how do we resolve contention when we have two machines sharing the same filesystem?

Charvi: multiple JVMs are OK as long as you use a unique threadGroup name: sreeHome

Marton: the 7.0 docs say you don't have to use clear anymore and that it could cause corruption. Is that true?

Charvi: Yes.

Marton: how can you tell which registry you have gotten, to verify.

Charvi: you could print the hashcode. But testing the content is best.