General protocol how to test a whole new WMS release
- Test all the modes supported by the init scripts (i. e. start/stop/restart/graceful). Clear out previously created working directories and files. repeatedly try start, stop, restart etc.
- Start with a job for each supported job type:
- Single job
- Collection (two ways of specifying the JDL for compound jobs, directly including the ClassAd or pointing to another file)
- Parametric job
- DAG
- MPI
- jobs should be sent to both CREAM and LCG-CE
- test fuzzy ranking for collections and compound jobs in general
- Try a listmatch
- Try cancelling requests for single, collection and DAG jobs
- Functionality
- check that both shallow and deep resubmission work (in particular check that the token is created)
- try out the file perusal
- check that the Maradona file is remotely copied to the job sandbox
- test zipped and unzipped ISB
- test gridsite delegation protocols 1/2 in all their possible combinations
- make sure that the proxy is deregistered after a job completes
- make sure that the ISM does not get empy from time to time
- check how the brokerinfo file is generated (glite-brokerinfo getCE)
- ...
- Interoperability
- Send jobs to ARC and OSG computing elements
- General recommendations
- keep an eye to the memory occupation, if it grows too large fast it could be related to purchasing. set the purchasing rate to a very low value. If it grows on massive job submission the cause is to be searched elsewhere
- cause a job to be purged by each component (wmproxy, wm, ice etc.) and see what happens
- ...
--
MarcoCecchi - 15 Jul 2009