1. Ivan Lanese Computer Science Department University of Bologna/INRIA Italy Decidability Results for Dynamic Installation of Compensation Handlers Joint work with Gianluigi Zavattaro

2. Map of the talk l Long-running transactions l Compensation installation l Gap in the expressive power l Conclusions

3. Map of the talk l Long-running transactions l Compensation installation l Gap in the expressive power l Conclusions

4. Handling unexpected events l Current applications run in environments such as the Internet or smartphones l Possible sources of errors –Communication partners may disconnect –Message loss –Received data may not have the expected format –Changes in the environment –... l Unexpected events should be managed so to ensure correct behavior even in unreliable environments

5. Compensation handling l In service-oriented computing the concept of a long running transaction has been proposed –Computation that either succeeds or it aborts and is compensated l The compensation needs to take back the system to a correct state –Undoing cannot always be perfect –Approximate rollback l Programming compensations is a delicate task

6. Different primitives in the literature l Long-running transactions used in practice –WS-BPEL, Jolie l A flurry of proposals in the literature –Sagas, StAC, cjoin, SOCK, dcπ, webπ, … l Are the proposed primitives equivalent? l Which are the best ones?

7. A difficult problem l Approaches to compensation handling can differ according to many features –Flat vs nested transactions –Automatic vs programmed abort of subtransactions –Static vs dynamic definition of compensations l Approaches applied to different underlying languages –Differences between the languages may hide differences between the primitives

8. Our approach l Taking the simplest possible calculus (π-calculus) l Adding different primitives to it l Comparing their expressive power l Too many possible differences l We concentrate on static vs dynamic definition of compensations l Decidability of termination (all computations terminate) allows to discriminate them –In a π-calculus without restriction

9. Map of the talk l Long-running transactions l Compensation installation l Gap in the expressive power l Conclusions

10. Static compensations l The compensation code is fixed –Java try P catch e Q –Q is the compensation for the already executed part of P –Q does not depend on when P has been interrupted l First approach that has been proposed l Still the most used in practice (WS-BPEL) l Not flexible enough

11. Dynamic compensations l The compensation can be updated during the computation –To take into account the changes in what has been done l A primitive to define a new compensation is needed –The new compensation may possibly extend the old one

12. Syntax of the calculus

13. Simple examples

14. Simple examples: compensation update

15. Classes of calculi l Dynamic compensations l Nested compensations l Parallel compensations l Replacing compensations l Static compensations

16. Classes of calculi

20. l Dynamic compensations l Nested compensations l Parallel compensations l Replacing compensations l Static compensations –Compensation updates are never used

21. A partial order Dynamic Replacing Nested Parallel Static Are the inclusions strict?

22. A partial order Dynamic Replacing Nested Parallel Static [ESOP2010] Relying on complex conditions on allowed encodings and operators [Here] Decidability of termination

23. Map of the talk l Long-running transactions l Compensation installation l Gap in the expressive power l Conclusions

24. Undecidability for nested compensations l We prove that they can code RAMs l RAMs are a Turing powerful model –Termination is undecidable l A RAM includes –A set of registers containing non negative integers –A set of indexed instructions l Two possible instructions –Inc(r j ): increment r j and go to next instruction –DecJump(r j,s): if r j is 0 go to instruction s, otherwise decrement r j and go to next instruction, l A RAM terminates if an undefined instruction is reached

25. Encoding idea

26. Decidability for parallel/replacing compensations l We exploit the theory of Well-Structured Transition Systems (WSTS) l Termination is known to be decidable for WSTS l We just have to prove that for each process P its derivatives form a WSTS

27. Well Quasi Ordering (wqo) l A reflexive and transitive relation (S,≤) is a wqo if given an infinite sequence s 1,s 2,… of elements in S, there exist i<j such that s i ≤s j

28. Well-Structured Transition System (S,→,≤) is a WSTS if –(S,→) is a finitely branching transition system –(S,≤) is a wqo –Compatibility: for every s 1 →s 2 and s 1 ≤t 1 there exists t 1 →t 2 such that s 2 ≤t 2 s1s1 t2t2 s2s2 t1t1 ≤ ≤

29. Idea of the proof l Given a process P with parallel or replacing compensations in its derivatives –No new names are generated –The set of sequential subprocesses never increases l This is not the case for nested compensations, since they allow to create infinitely many sequential processes l The order in the next slide is a wqo thanks to Higman’s lemma l Compatibility holds l Decidability follows from the theory of WSTS

30. Wqo on processes

31. Map of the talk l Long-running transactions l Compensation installation l Gap in the expressive power l Conclusions

32. Summary l We distinguished different forms of compensation installation l We showed that decidability of termination allows to highlight a gap between –Dynamic and nested compensations on one side –Static, parallel and replacing compensations on the other side l The result is robust –Different ways of managing subtransactions –The same holds for CCS with similar primitives l Absence of restriction is fundamental

33. Future work l Can we give termination preserving encodings of –Dynamic into nested compensations? –Parallel/replacing into static compensations? l The full picture of the expressive power of primitives for long running transactions is still far –Other dimensions –Which is the impact of the underlying calculus?

34. End of talk