Fix the way of send the rpc requests to OperationEngine service
Bug #1572865 reported by
zengchen
This bug affects 1 person
| Affects | Status | Importance | Assigned to | Milestone | |
|---|---|---|---|---|---|
| Karbor |
Triaged
|
Critical
|
Unassigned | ||
Bug Description
Suppose there are two instances of OperationEngine as A and B on two machines.
For triggers, we hope that A and B have all triggers, so we use cast interface of RPC
to send requests to A and B.
For operations, we hope A and B have different operations. so we use call interface
of RPC to send requests to A or B to create operations; and we hope we can delete
operations completely, we use cast interface of RPC to send requests to A and B.
| Changed in smaug: | |
| importance: | Undecided → Medium |
| status: | New → Triaged |
Revision history for this message
| Saggi Mizrahi (ficoos) wrote | #1 |
| Changed in karbor: | |
| importance: | Medium → Critical |
| Changed in karbor: | |
| milestone: | none → ocata |
Revision history for this message
| heyiheng (hyh01) wrote | #2 |
To post a comment you must log in.
Every update needs to be done with a call then push information to the DB and then the node that did the operation needs to cast that there was a change. Other nodes need to check the DB for changes periodically. Ownership can be done in the DB. There are currently existing mechanisms for knowing if an instance of a service is up or not. Operation Engine instances need to make sure to take ownership of disowned operation (or a scheduler needs to do this).