Orchestration Evolution & Dataflow Concepts:

Transcription

1 Orchestration Evolution & Dataflow Concepts: Introducing Unanticipated Loops Inside a Legacy Workflow Sébastien Mosser Mireille Blay Fornarino Johan Montagnat mosser@polytech.unice.fr CNRS, I3S Laboratory, Modalis Team ( ) University of Nice Sophia Antipolis ( ) 4 th Int. Conference on Internet, Web App. & Services Thursday, May 28th 2009 Venice Mestre, Italy

2 Context ADORE Dataflow Set Evolution Conclusions Agenda 1 Context : SOA, Business Processes & Business Experts 2 ADORE : Activity model supporting orchestration Evolution 3 Introducing Dataflow concepts inside ADORE 4 The Set Evolution algorithm : v v 5 Conclusions & Perspectives 2 Sébastien Mosser Orchestration Evolution & Dataflow Concepts: CNRS I3S / University of Nice

3 Context ADORE Dataflow Set Evolution Conclusions Agenda 1 Context : SOA, Business Processes & Business Experts 2 ADORE : Activity model supporting orchestration Evolution 3 Introducing Dataflow concepts inside ADORE 4 The Set Evolution algorithm : v v 5 Conclusions & Perspectives 3 Sébastien Mosser Orchestration Evolution & Dataflow Concepts: CNRS I3S / University of Nice

4 Context ADORE Dataflow Set Evolution Conclusions Services Oriented Architecture & Workflows SOA are based on Services (it s in the title!) Specialized (i.e. elementary) code Atomic (& stateless) functionalities written by developers Implemented as Web Services [JAX-WS] SOA rely on Business Processes Value added assemblies of services calls Business driven, high reactivity,... written by business experts Implemented as Orchestrations (aka Workflows) [BPEL] 4 Sébastien Mosser Orchestration Evolution & Dataflow Concepts: CNRS I3S / University of Nice

5 jseduite, an Information Broadcasting SOA Objective : Broadcast relevant informations to school users Principles 1 Sources of information can deliver relevant informations e.g. bus schedules, alerts, restaurant menu 2 Providers retrieve and assemble these informations conforming with broadcasting policies 3 Broadcasting devices deliver information to users e.g. public plasma screens, PDA, cellphone Implementation Sources : Web Services implemented by engineer students Providers : Orchestrations implemented by SOA teachers Broadcasters : Javascript code inside kiosks browser

6 Orchestration Example : InfoProvider

7 Going back to business experts reality... Ouch!

8 Going back to business experts reality... Ouch!

9 Context ADORE Dataflow Set Evolution Conclusions Motivations : A Design By Composition approach Business Experts cannot handle complete BPEL processes... Too development oriented, too technical,... Too complex,... even for developers!... But they can focus on small parts... How to add a new source inside jseduite? How to add a cache mechanism in front of a source?... And let a framework perform the composition 1 Merging multiple processes fragments [ECSA 08] 2 Introducing loops over datasets [ICIW 09] Auto-magically handling a set of groups instead of a single one. 8 Sébastien Mosser Orchestration Evolution & Dataflow Concepts: CNRS I3S / University of Nice

10 Context ADORE Dataflow Set Evolution Conclusions Agenda 1 Context : SOA, Business Processes & Business Experts 2 ADORE : Activity model supporting orchestration Evolution 3 Introducing Dataflow concepts inside ADORE 4 The Set Evolution algorithm : v v 5 Conclusions & Perspectives 9 Sébastien Mosser Orchestration Evolution & Dataflow Concepts: CNRS I3S / University of Nice

11 Context ADORE Dataflow Set Evolution Conclusions Example : A mathematical SOA Legacy Infrastructure A Memory service Bind a mnemonic name to an integer value An Adder process Written by our mathematical business expert (small is beautiful) adder : String String x y Integer x int + y int 10 Sébastien Mosser Orchestration Evolution & Dataflow Concepts: CNRS I3S / University of Nice

12 Adder(x,y) process : Perform xint + yint

13 Adder x : x x, Adder(x,y)

14 Context ADORE Dataflow Set Evolution Conclusions Summarizing ADORE in just one slide An abstract reasoning model An orchestration O is defined as : A set of activities A A partial order between those activities An activity is defined as : An unique identifier Id, a kind K (BPEL subset) A set of inputs variables Ins, and an output one Out O (A, ) a A, a = (Id, K, Ins, Out) (a 1 a 2 ) start(a 2 ) end(a 1 ) 13 Sébastien Mosser Orchestration Evolution & Dataflow Concepts: CNRS I3S / University of Nice

15 Context ADORE Dataflow Set Evolution Conclusions Expressing the Adder process inside ADORE adder ({a 1, a 20, a 21, a 3, a 4 }, {a 1 a 20, a 1 a 21, a 20 a 3,...}) a 1 (a 1, receive, {x, y}, ) a 20 (a 20, invoke(memory, get), {x}, {x int }) a 21 (a 21, invoke(memory, get), {y}, {y int }) a 3 (a 3, assign(integeradd), {x int, y int }, {r}) a 4 (a 4, reply, {r}, ) 14 Sébastien Mosser Orchestration Evolution & Dataflow Concepts: CNRS I3S / University of Nice

16 Context ADORE Dataflow Set Evolution Conclusions ADORE evolution capabilities Behavioral evolution : Expresses orchestration & evolutions Small fragments, business driven Use a merge algorithm to compose it Datasets evolutions : Define (or reuse) a process handling a variable v More simple than directly working with sets Merge : O E O Enhance : v v Enhance it to iterate over a dataset v {v 1,..., v n } 15 Sébastien Mosser Orchestration Evolution & Dataflow Concepts: CNRS I3S / University of Nice

17 Context ADORE Dataflow Set Evolution Conclusions Agenda 1 Context : SOA, Business Processes & Business Experts 2 ADORE : Activity model supporting orchestration Evolution 3 Introducing Dataflow concepts inside ADORE 4 The Set Evolution algorithm : v v 5 Conclusions & Perspectives 16 Sébastien Mosser Orchestration Evolution & Dataflow Concepts: CNRS I3S / University of Nice

18 Context ADORE Dataflow Set Evolution Conclusions Dataflow expressiveness inside ADORE Defined as a set of functions Dataflow : D F Transitively retrieves activities using a given variable Firsts & Lasts F, L Retrieves activities without any predecessors / successors in Interface : Interface Identify interface activities (receive, reply or throw) Core : C Built over the previous functions Identify the body of the expected loop. 17 Sébastien Mosser Orchestration Evolution & Dataflow Concepts: CNRS I3S / University of Nice

19 Illustrating DF

20 Illustrating D F Control Flow

21 Illustrating D F x Data Flow

22 Illustrating D F D F (A, x)

23 Illustrating F & L

24 Illustrating F & L F(A, )

25 Illustrating F & L L(A, )

26 Illustrating Interface

27 Illustrating Interface Interface(A, x)

28 Illustrating C

29 Illustrating C D F (A, x)

30 Illustrating C

31 Illustrating C C(A,, x)

32 Context ADORE Dataflow Set Evolution Conclusions Agenda 1 Context : SOA, Business Processes & Business Experts 2 ADORE : Activity model supporting orchestration Evolution 3 Introducing Dataflow concepts inside ADORE 4 The Set Evolution algorithm : v v 5 Conclusions & Perspectives 22 Sébastien Mosser Orchestration Evolution & Dataflow Concepts: CNRS I3S / University of Nice

33 Context ADORE Dataflow Set Evolution Conclusions Generic algorithm : SetEnhance(o, v) Handling v {v 1,..., v n } instead of v in o 1 Identify the iteration body body C(Activities(o), Order(o), v) 2 Substitute loops input & output variables Stickel symbolic substitution σ(a b) 3 Add a loop activity Contains the core activities And 2 special activities to feed (resp. swell) the loop. 4 Reorder external activities To avoid inconsistent orders 23 Sébastien Mosser Orchestration Evolution & Dataflow Concepts: CNRS I3S / University of Nice

34 Running SetEnhance(o, x) : D F (Activities(o), x)

35 Running SetEnhance(o, x) : Loop body

36 Running SetEnhance(o, x) : σ(x x ), σ(r r )

37 Running SetEnhance(o, x) : Creating the loop

38 Running SetEnhance(o, x) : Reordering inconsistent orders

39 Running SetEnhance(o, x) : Result

40 Context ADORE Dataflow Set Evolution Conclusions Implementation : Prolog language Logical definition logical language An orchestration is defined as facts The algorithm is defined as Prolog rules It computes a set of actions to enhance a process?- [seduite,enhance ].?- setenhance(informationprovider, lecturegroup,todo). ToDo = [addactivity(l1, loop([l1_feeder,a20,a21,a4,l1_sweller]),[], []), delorder(a1,a20), addorder(a1,l1), sigma(...,...)...] 25 Sébastien Mosser Orchestration Evolution & Dataflow Concepts: CNRS I3S / University of Nice

41 Back to (jseduite) reality : Generality vs Real Life Idea : SetEnhance(O, v) as a generic algorithm The algorithm build a cartesian product This choice doesn t fit particular cases It s not a bug, it s a functionality! Need : Loop make up capability 1 Let the algorithm performs the (nasty) loop introduction task It s its job! 2 Customize the cartesian product to achieve your goal It s your job! jseduite case study : group group 1 Invoke SetEnhance(infoProvider, group) 2 Reuse a concatenate instead of cross ADORE evolution job s done.

42 Context ADORE Dataflow Set Evolution Conclusions Agenda 1 Context : SOA, Business Processes & Business Experts 2 ADORE : Activity model supporting orchestration Evolution 3 Introducing Dataflow concepts inside ADORE 4 The Set Evolution algorithm : v v 5 Conclusions & Perspectives 27 Sébastien Mosser Orchestration Evolution & Dataflow Concepts: CNRS I3S / University of Nice

43 Context ADORE Dataflow Set Evolution Conclusions Summary Done ADORE, a tool to reason on orchestrations A set of functions to handle dataflows concepts A loop introduction algorithm based on these functions To do... Enhancing the BPEL generation process Larger experiments than the jseduite one, metrics, Sébastien Mosser Orchestration Evolution & Dataflow Concepts: CNRS I3S / University of Nice

44 Context ADORE Dataflow Set Evolution Conclusions Perspectives Formal methods & process algebra (e.g. π calculus) Validating built processes Property preservation (e.g. execution order) Model checking (e.g. determinism, dead-lock, user properties) Enhancing the conflict detection mechanisms Composing loops Expressing data composition policies Composing datasets using grid computing operators 29 Sébastien Mosser Orchestration Evolution & Dataflow Concepts: CNRS I3S / University of Nice

45 Context ADORE Dataflow Set Evolution Conclusions Thanks for your attention! Any questions? Sébastien Mosser Orchestration Evolution & Dataflow Concepts: CNRS I3S / University of Nice