Hub & Spoke - Flow CRUD

carmine commented 6 years ago

The primary objective of this feature is to implement a hub and spoke version of Flow CRUD - ie a central workflow that issues commands to each component involved in setting up a Flow, maintaining a history of steps taken in the database. Additionally: deprecate sending messages to flow cache and to topology engine. Instead, save flow topology state directly in the db.

High Level Plan:

[x] Complete Arch/Design work
[x] Implement simple version of new Create
[x] Implement simple version of new Delete
[x] Implement simple version of new PATH Update
[ ] Implement Zombie APIs
[x] Implement HID / History API(s) #1195
[ ] Add a resume Zombies to a startup event.

Arch/Design Work

[x] Create initial architecture artifacts
[x] Complete initial reviews
[x] Complete architecture artifacts

Implement H&S Flow Create: The task order also represents the algorithm as well

[x] create designs
[x] create /api/v2/flow/{flow-id} POST
[x] validate flow
[x] get a new History ID to use (GUID), and use for HISTORY and DB
[x] Send to Flow History table : REQUEST RECEIVED (include request, HID)
[x] Store Path in DB, with state STORED : add to History - STORED
[x] Validate Flow Installed : Update DB State - VALIDATED : add to history
[x] Update state - DB - UP : add to history
[x] rollback if a switch goes down during rules installation/rule is invalid
[x] add protected path and flow diverse functionality
[x] add retries for flow creation / commands installation
*** If a failure happens at any stage, do a DELETE (ROLLBACK)
*** be mindful that there are two flows (forward / reverse)

Implement H&S Flow Delete: The task order also represents the algorithm as well

[x] create /api/v2/flow/{flow-id} DELETE
[x] get a new Transaction ID to use (GUID), and use for HISTORY and DB
[x] Send to Flow History table : REQUEST RECEIVED (include request, HID)
[x] validate the ID exists in the DB : update history with results - return 404 if not existing
[x] Update state - DB - DELETING_STARTED, add delete_hid : add to history
[x] (optional) pre-validate the flow (ie the network matches the db) : store results of pre-val in History (DELETING_PREVALIDATED) .. no DB update
[x] Push new state - FlowTopo : add to history
[x] Refactor TPE to enable just the "SEND DELETE INSTRUCTIONS" : Update State - DELETING_SENT : add to history
[ ] Validate Flow Deleted : Update DB State - DELETING_VALIDATED : add to history
[ ] Unpush the flow : add to History : DELETING_CACHED_REMOVED
[x] Remove the flow in DB : add to History - DELETED.

Implement H&S Flow PATH Update: The task order also represents the algorithm as well

[x] create /api/v2/flow/{flow-id} PUT
[x] similar first steps as delete - ie hid, validate flow exists ..BUT need to account for UPDATING?. Might start with just a line item in HISTORY .. then just normal CREATE, then DELETE, then ID SWAP.
[x] similar to delete - allow for pre-validate
[x] CREATE new path..add path as "flowpath_hid"
[x] CREATE send instructions
[x] VALIDATE installed
[x] SWAP flowpath_hid and flowpath
[x] DELETE send instructions
[ ] VALIDATED removed
[x] Drop old Flow Segments
[x] Add new Flow Segments.

Implement H&S Flow PATH Reroute:

[x] create /api/v2/flow/{flow-id}/reroute POST
[x] validate flow exists
[x] Find a new path, allocate flow resources (cookie, meters, etc)
[x] CREATE new path..add path as "flowpath_hid"
[x] CREATE send instructions
[x] VALIDATE installed
[x] SWAP flowpath_hid and flowpath
[x] DELETE send instructions
[ ] VALIDATED removed
[x] Drop old flowpath

Implement H&S Swap Flow PATHs

[x] API / Services / FSM / Actions

Implement H&S Swap Flow Endpoints

[x] API / Services / FSM / Actions

Implement H&S Partial Update (PATCH)

[x] API / Services / FSM / Actions
[ ] Update meter - need to clarify whether it should be a separate operation.

Implement H&S Sync Flow (Reinstall)

[ ] TBD

Implement H&S Delete multiple flows

[ ] TBD

Implement H&S Push/Unpush flow

[ ] Need to clarify whether the methods are still required.

FINALLY:

[ ] Any Single Points of Failure?
[x] Add to NB API - endpoints to get latest step in History of HID
[x] Add to NB API - endpoints to get all history of HID
[x] Add to NB API - endpoints to get all history related to flowid
[ ] Add to NB API - return all Zombies - (yet to be defined .. probably anything in History that doesn't have the last step of the history .. or any flow that isn't in a final stage .. and it isn't in memory .. or is in memory but is older than x.
[ ] When restarting the NB API server .. it should pickup incomplete tasks .. but then also validate that they made it as far as they did (ie if it says it completed step 4 of 8 ... did it really? And what about 5, because it may have started that but didn't get confirmation).

surabujin commented 6 years ago

Can we make a decision on https://github.com/telstra/open-kilda/issues/688 first? This is big chunk of new functionality. It will be extremely pity to implement it using old/existing way of communication.

carmine commented 6 years ago

Looking for initial design artifacts on July 11

sergii-iakovenko commented 5 years ago

Removing the following from the description as pre-h&s is completed:

Blocked by:

1127 Create java based DAO layer in front of Neo

561 Eliminate Cache topology

1128 Refactor out TPE for Flow stuff

telstra / open-kilda

Hub & Spoke - Flow CRUD #472

1127 Create java based DAO layer in front of Neo

561 Eliminate Cache topology

1128 Refactor out TPE for Flow stuff