Release Management in the UBS Data Warehouse Program

Agenda w Project Overview w Project Infrastructure w Change Management w Release Process w Summary

Project Overview UBS § Global, integrated investment services firm and leading bank in Switzerland § World’s leading provider of private banking services § Total client assets US$ 1. 41 trillion in 2001 § Acquired in 2000

Project Overview Data Warehouse Program (DWP) § Establish common infrastructure for analytical data processing § Provide a business oriented set of data warehouse and data mart services § Standardized business data model § Align the bank’s data mart portfolio § Improve flexibility, time-to-market and data quality

Project Overview Data Flow System of Records extract, condition & load (inbound) Common Data / Business Warehouse data mart sourcing (outbound) Data Marts visualization business user

Project Overview Delivery Streams § Main organizational element: team working on a subject data area / data mart § Three letter acronym as base for naming standards § Standardized infrastructure: UNIX directories, access group, meta data area, . . . § Team size typically between 1 and 5 § Delivery streams release independently from each other

Project Overview System of Records Application Structure RCL DSFDSF MDR <STR > RCL release control tools MDR meta data repository DSF DWP sourcing framework <FDS> data feed definition <STR> delivery stream <MAR> outbound stream / data mart sourcing . . . <STR <FDS > . . . <FDS > DWP Sourcing Framework <MAR> . . . <MAR> >

Project Infrastructure System Environments Developmen t • • Test IBM SP 2 cluster, AIX 4. 3 DB 2 UDB EEE Power. Center V 5. 1 DWP Sourcing Framework Cognos / Business Objects Netscape Enterprise Webserver V 3. 63 Clear. Case V 4. 2 Production

Project Infrastructure Logical Environments and Release Structure Developme nt Test Productio n emergency (SOS) releases S D A P X F V T delivery stream development framework test / delivery stream migration to new framework releases framework development mandatory optional

Project Infrastructure Release Cycles § delivery stream development: D A P § framework development: T X F [ V] after sign-off: X D A P § migration to new framework releases: {DAP} X F § emergency releases: P E P

Change Management Design Principles § Support different, clearly separated environments with different responsibilities § All environments have identical structure (products, databases, server configurations) § All program changes are done on the development system § All changes on test and production systems go through the release process and are clearly tracked § “just-in-time production”

Change Management Challenges § currently ~60 delivery streams § and ~500 feeds § between 10 and 7103 objects per release unit § >3500 single releases since May 2000 § >5500 deliver requests § ~1000 active installations

Change Management Directory Structure § Two areas: /dwp_root release area, version controlled /dwp_data dynamic data, archival on demand § Additional directory level in order to support more than one logical environment on one system § /dwp_root is organized by delivery streams, e. g. : /dwp_root/d/streams/rcl/bin

Change Management Directory Structure (continued) § /dwp_data is organized by logical processing steps, e. g. : /dwp_data/p/data/landing (landing area) /dwp_data/p/data/tgtfiles (target files) /dwp_data/p/logs/system (framework log area) § Tool support for generation of directories in source environments (delta processing) § Automatic creation of missing directories in target

Change Management Clear. Case Set-up § One single VOB /vobs/dwp § Same directory structure as below /dwp_root, directories are automatically created § In general: only linear version trees (important: synchronization with database change management) § Branch support for emergency releases

Change Management Clear. Case Set-up § In general: fully automated access layer (freeze and deliver routines) § On demand (larger teams): direct Clear. Case access via team specific views: /dwp_root/d/streams/rcl -> /view/rcl_team/vobs/dwp/streams/rcl § Allows use of DWP framework § Common labeling strategy (freeze -nocheckin)

Change Management Release Naming § <delivery stream>_<major>. <minor>. <patch> e. g. : RCL_1. 1. 0 § <major> major release number (high level “wave” planning) § <minor> minor release number (delivery stream development plan) § <patch> patch level (bug fixes) § Emergency releases: RCL_1. 1. 0_SOS_<#>

Change Management Release Attributes § Description (standard comment) § INSTALLED (list of environments where release is installed) § TRANSITIONS (describes the deployment path of a release) § RESPONSIBLE (due to centralized freeze routine) § PCRVERSION (Power. Center meta data version) § CR (change request link) § PSO (production sign off)

Change Management Versioning Example common. p m /main/1 RCL_1. 0. 0_sos _1 /main/sos/1 RCL_1. 0. 0 /main/2 /main/3 RCL_1. 0. 1 RCL_1. 0. 2 /main/4 /main/5 RCL_2. 0. 0

Change Management Change Control Board § Responsible for high level planning and impact analysis § Defines release scope and release numbers on base of delivery streams § Assigns responsibilities (delivery stream manager, data modeler, database administrator, business responsible) § Result is documented in “wave plan” document

Release Process Roles & Responsibilities Role Responsibility developer testing development, unit and integration delivery stream manager, release planning database administrator database change control release manager control, deployment, tracking, configuration administration

Release Process Overview Developmen t Test Freeze Production Receive Clear. Case Deliver Deployment Package

Release Process Release Objects § Power. Center mappings § Job dependency data § scripts / SQLs / programs and source code in general not included: § Database objects (in concept phase) § documentation (intranet database)

Release Process Release Procedure Responsible 1. Submit database change request *) delivery stream manager 2. Implement database changes *) database administrator 3. Prepare release area delivery stream manager (UNIX, Power. Center, job dependencies, Uniserv) 4. Submit release request delivery stream manager 5. Prepare release area (DDLs) *) database administrator 6. Create new release (Freeze) release manager 7. Create deployment package (Deliver) release manager 8. Apply database changes to target system *) database administrator 9. Install release in daily deployment window release manager (Receive) (IT integration / IT operation) *) in case of database changes only

Release Process Freeze Process freeze <release unit> [ -patch | -minor | -major | -sos <release> ] 1. Retrieve previous release (“element * RCL_1. 0. 0”) 2. Compare with release area (check for new, changed, deleted files) 3. Display results and ask for confirmation 4. Apply changes in Clear. Case 5. Create release label and set release attributes

Release Process Deliver Process deliver <release unit> -t <target env. > [ -r <release> ] [ -a ] 1. Retrieve specified / latest release in Clear. Case 2. Retrieve current release in target environment and create delta 3. Use -a(ll) for initialization / synchronization 4. Create deployment package (tar file + control file) 5. Update release attributes (predecessor and current release!)

Release Process Receive Process receive 1. Check hand-over area for pending releases 2. Remote copy deployment package to target system 3. Install it 4. Standardized post-installation steps: e. g. access permissions 5. Delivery stream defined post-installation steps (Post. Install. ksh file): e. g. for non-standard path names, generators, setuid bits

Release Process Delta Deployment § Smaller packages (“top 1” feed: > 7000 files) § clear change log § both ways: creation and removal of files

Release Process Release Cache Question: Which is the predecessor release in a specific target? Installation information is stored with each release label. cleartool lstype -kind lbtype cleartool desc lbtype: RCL_1. 0. 0 Problem: takes ages with 3500 release labels!

Release Process Release Cache Solution 1: Manage installation information on element version level + delta processing easy – label twice (release and installation label) – missing overview, need log files for tracking – error prone (e. g. interruption of freeze process)

Release Process Release Cache Solution 2: Release cache (including all release attributes, master information remains in Clear. Case) – additional software layer required (consistency!) + clear information structure + efficient delta processing + added value: release database

Release Process Release Database

Release Process Release Database (Details)

Release Process Release Request (1)

Release Process Release Request (2)

Release Process Release Request (3)

Release Process Release Tracker (Overview)

Release Process Release Tracker (Details)

Release Process Releasing Meta Data § Power. Center, job dependency and feed configuration meta data are stored in the database § Unload utilities to export the corresponding data and store it in the delivery stream´s release area (tar file) § Meta data is frozen and delivered together with all other file system objects § Load utility for loading objects in target environment

Release Process Database Change Management (Concept Phase) § PATROL DB-Change Manager by bmc software § Scope filter: assign database objects to delivery stream ( view in Clear. Case) § Apply database changes to development database § Create release baseline: freeze all database object versions for a delivery stream ( label in Clear. Case, name equal to release

Release Process Database Change Management (target system) § Target baseline: target database version (delivery stream plus timestamp as name) § create delta DDL depending on release baseline and latest target baseline (PATROL) § apply delta DDL § create new target baseline

Release Process Database Retrofitting Process § Rationale: certain database changes have to be applied and tested directly on the Production system (load performance optimization: indexes, summary tables, . . . ) § In order to include target system changes into next regular release, changes are promoted back to Development using the Retrofitting Process § In principle: new, database administrator driven

Summary Experiences § Over 3500 releases since May 2000 § Effort for creation and installation of new release: 5 - 60 minutes depending on amount of meta data (without database changes) § Main effort necessary for handling of meta data § No Clear. Case problem encountered so far