#opendaylight-group-policy Meeting

Meeting summary

1. what has changed between 0.95 and 0.96? (regXboi, 17:08:45)
   1. minor changes (regXboi, 17:08:51)
   2. https://docs.google.com/file/d/0B0Pf6vxIzl4lX3J6cDBQZ3ZJZms/edit (regXboi, 17:09:51)
   3. the above is the 0.96 model (regXboi, 17:10:01)
   4. labels have now been subdivided into roles, capabilities, conditions, and qualities (regXboi, 17:12:33)
   5. clauses have also been added (regXboi, 17:12:47)
   6. groups select contracts via selectors (or name releationships). A selector is formula on the contract's qualities (regXboi, 17:14:08)
   7. question from Jan: what does target? (regXboi, 17:15:21)
   8. mike says that the target is how the contract presents itself (regXboi, 17:15:48)
   9. is this same as label we had? (dconde, 17:17:03)
   10. see above .... labels have been subdivided statement... (regXboi, 17:17:22)
   11. understdood (dconde, 17:18:58)
   12. regXboi asks for the use case for a contract presenting multiple targets (regXboi, 17:21:40)
   13. dvokinista and alagalah say the use case is allowing different users to have different selectors for the same contract (regXboi, 17:22:03)
   14. regXboi says he'll think about it and scream if it doesn't make sense (regXboi, 17:22:19)
   15. alagalah taking over scribing (alagalah, 17:22:55)
   16. jmedved continues to ask why I would need separate targets (regXboi, 17:23:15)
   17. dvorkinista says think of presenting targets for development versus test? (regXboi, 17:23:56)
   18. jmedved was concerned the selector with multiple targets modifying the contract. dvorkinista pointed out this doesn't change anything in the contract, its a way of selecting the contract (alagalah, 17:25:58)
   19. regXboi made an analogy of UNIX file systems. The contract is the file, the selector and targets are logical links, a way to traverse to the file (alagalah, 17:26:43)
   20. dvorkinista pointed out its a lot of like labels in gmail (alagalah, 17:27:01)
   21. Earlier on, dvorkinista pointed out that labels have subsets, Roles, Capabilities, Conditions and Qualities (missed that earlier) (alagalah, 17:28:38)
   22. s3wong asks if this makes the prior label scheme less flexible (regXboi, 17:28:56)
   23. dvorkinista said that 0.95 -> 0.96 is some renaming of things to make them clearer (ie the sub-groups above of labels) (alagalah, 17:29:09)
   24. dvorkinista says they are equivalent (regXboi, 17:29:10)
   25. readams asked a question that I couldn't pick up (alagalah, 17:29:57)
   26. Why would you want to selectively match against targets that are being *provided* as opposed to consumed (readams, 17:33:21)
   27. mickey_spiegel says we have to remember what provides and consumes means (regXboi, 17:34:09)
   28. so that we can add new contracts without disturbing existing contracts (regXboi, 17:34:32)
   29. readams wonders why we want this particular semantic feature (multiple contracts, multiple targets) (alagalah, 17:35:52)
   30. Example: Existing contract for http and https, now add another protocol that requires going through another appliance (mickey_spiegel, 17:36:09)
   31. dvorkinista says it allows for provision of services and combination of services qithout wondering how contracts are structured (alagalah, 17:36:26)
   32. Now add another contract with additional qualifier, without disrupting others (mickey_spiegel, 17:36:28)
   33. Service does not even need to know you are going through another appliance (mickey_spiegel, 17:36:32)
   34. jmedved wants use case/concrete examples (alagalah, 17:37:57)
   35. dvorkinista using DB cluster as an example. Shows a target for anything that wants to consume the database (alagalah, 17:38:23)
   36. alagalah is going to find a way to record video of whiteboard sessions. Cos scribing this is nigh on impossible (alagalah, 17:38:47)
   37. Discussion is around backup providers (alagalah, 17:41:09)
   38. Back-up providers is a usecase to demonstrate the need for the selector concept (ChrisPriceAB, 17:42:36)
   39. dvorkinista says basic contract is to provide a contract based on a query. (alagalah, 17:43:38)
   40. dvorkinista says basic concept is to provide a contract based on a query. (alagalah, 17:43:52)
   41. missed readams response (alagalah, 17:44:14)
   42. regXboi asks how two contracts with same set of labels get resolved ? (alagalah, 17:45:43)
   43. dvorkinista points out that target is only used for selection. It's a way of selecting multiple contracts (alagalah, 17:46:19)
   44. regXboi since targets are only used for selection, then there needs to be a discussion on contract conflict resolution (alagalah, 17:46:56)
   45. dvorkinista agreed (alagalah, 17:47:28)
   46. ChrisPriceAB wanted to know if you should have a more specific query mechanism (alagalah, 17:48:03)
   47. ChrisPriceAB in a way to avoid handling conflict resolution (alagalah, 17:48:18)
   48. dvorkinista agrees conflict resolution is important. (alagalah, 17:48:32)
   49. jmedved buids on regXboi's question about resolution at the contract level or something else (regXboi, 17:50:19)
   50. dvorkinista says its at the rules level and not the contract level (regXboi, 17:50:35)
   51. mickey_spiegel says at the end of the day, we are sending packets and rules will be selected based on the information in the packet (regXboi, 17:51:28)
   52. note: that can include prior state (regXboi, 17:51:39)
   53. er... that last is my editorial note (regXboi, 17:51:50)
   54. in data plane, you determine contract based on source and destination addresses (mickey_spiegel, 17:51:56)
   55. Anywhere where you do enforcement in the data plane, have to come up with the same answer for conflict resolution, for the same pair of source and destination addresses (mickey_spiegel, 17:52:15)

Action items

1. (none)

People present (lines said)

1. alagalah (46)
2. regXboi (40)
3. s3wong (6)
4. mickey_spiegel (6)
5. ChrisPriceAB (4)
6. odl_meetbot (3)
7. dconde (2)
8. readams (2)