Relational CDB (RCDB) Informal Workshop minutes: 8/3/2005
---------------------------------------------------------

Present: Veronique Lefebure(IT/FIO)
         Zhechka Toteva(CMS)
         Thorsten Kleinwort(IT/FIO)
         Bill Tomlin(IT/FIO), secretary
         Vlado Bahyl(IT/FIO)
         Tony Cass(IT/FIO)
         German Cancio Melia(IT/FIO)
         Tony Osborne(IT/FIO)
         Harry Renshall(IT/FIO), chair
         Eric Grancher(IT/DES)
         Nilo Segura(IT/DES)
         Miro Siket(IT/FIO)
         Maciej Stepniewski(IT/FIO)

HR: Introduced pan vs. rdbms issues
    Current design distorted to meet newer requirements
    GCM main customer and driver of requirements
    BT to act as architect of RCDB

VL: Gave presentation  <LINK>
    Mentioned that having no overall administrator or librarian was an
issue for any solution
    Provided provisional timescale estimates

HR: Why did CMS drop PAN ?

ZT: Difficult to maintain template files
    Didn't want to use obscure language specific to configuration
    Unpleasant to depend on grep/diff/sed for maintenance
    MySQL prototype ready
    Creates an xml profile per second
    Simple generic schema for 11 machines
    Business logic on client side
    Agreed that server side implementation with stored procedures better
    Said that generic schema not best solution
    Should multiply up time estimates by 1.5 or 2 to allow for testing

GCM: cdbop command line tool is 90% of updates
    Would need to provide a tool on rCDB side

TK: Need to address 2 structures: tpl structure and the global schema

GCM: Requirement for maybe > 5 levels of hierarchy

EG: In a short time, proven that correct set of RPMs can be easily
generated
    Views defined for ease of use
    Generic schemas make validation more difficult and kill performance
    Recommended more specific schema
    Strong typing possible with specific columns
    Difficulty putting data in the design in form of triggers
    Would need an overall supervisor to restrict operations
    Versioning as with pan/cvs would have to be implemented somehow in
Oracle
    Oracle can get diffs or data from specific time
    Simplify schema where possible at cost of flexibility
    Main problem is in rapid schema updates in shared environment

VB: Should Oracle Designer be used ?

EG: Not a fan

GCM: Consider how to implement ACLs
     Need architecture and E-R model to address requirements

HR: Should PAN go ?

GCM: It won't as other sites use it.
     Who will maintain it ?
     Suggested trying rCDB with PAN as failover solution
     Main issue in XML file generation for 6000 nodes.
     Current solution generates 1 profile in 0.2s

TO: Worried by lack of strong typing, rapid schema evolution & why
castor data so bad

BT: Need an administrator to avoid misuse of fields

GCM: Need to control who can update what types of information

EG: With RDBMS, very hard to evolve schema on production system

TO: Hopefully wild oscillation of schema has died down

TK: Must have ability to evolve schema

TC: What when RPMs go ?

GCM: Don't refer to RPM in schema, say 'package'

ZT: How often does the schema change ?

GCM: Core tree stable but new fields added frequently

TC: Why is it not portable to other RDBMS's ?

BT: Business logic embedded in Oracle specific code
    Need a proper prioritized Use Case model

GCM: If 3 Use Cases must be fulfilled (see
http://it-div-fio-lcg.web.cern.ch/it-div-fio-lcg/gcancio/cdb/cdb_require
ments_usesases.htm  "CDB Use Cases")
     Design model very important

HR: RDBMS advantages in interworking with other dbs and not having to
maintain PAN

TO: Gut instinct to go for RDBMS
    Next step, demonstrate

EG: Need a less flexible solution than PAN or RDBMS solution will get
very complex

GCM: That's OK. Don't want too much freedom.

NS: Many features required come for free with Oracle

GCM: Worry that all information in 300K of text files easy to see
     Data and meta-data visible
     Oracle too opaque
     Also, need ability to rollback to previous version
     How would we generate 6000 xml profiles in 10 minutes
     Architecture document should address this
     Dictionary needed to translate columns to xml paths
     Need ACLs that user a can modify x, y & z

NS: Can do ACL at Oracle level
    Kerberos 5/afs login can be used

HR: Unanimous consensus to continue rCDB to next level
    BT to start architecture document
    Early prototype for xml creation performance
    Have final similar meeting in 2 months to revue & finally decide

TC: Need a plan
    Prove feasibility of xml generation
    More detailed architecture needed