2014-12-16 14:45 GMT+01:00 Bertrand Delacretaz <[email protected]>: > > Hi Marius, > > On Tue, Dec 16, 2014 at 2:24 PM, Marius Petria <[email protected]> wrote: > > ...You are trying to upgrade the target system and you want to keep a > log of > > the changes in the source system, right?.. > > Yes > > > ...We have a way to put an agent in passive mode, that means it will not > > deliver any packages, and it only queues them. When you want to sync the > > target systems you would switch the state to active.... > > ok that sounds good > > > > > ...Replaying is not currently possible, unless we clone the state of the > > agent just before switching it to active. The agent state is made of > sling > > queues + binaries for distribution packages. We can probably implement an > > agent that has a snapshot state and can be rewind to that state.... > > ok - the queue state might be harder to capture, but the snapshot idea > sounds good. >
Another idea may be to use an agent A to serialize content to be replayed from time T0 to time T1 once (this can be done today already). Then an agent B can be used to read that serialized data and distribute it as it is (eventually without removing it, so that this can be replayed multiple times). Alternatively the queuing agent can be configured to use a specific queue for each snapshot (e.g. a queue for changed resources from T0 to T1), to be specified on demand or by configuration, and consuming agents on that queue will consume the items in there in order, that would require not to remove items from the queue so that those snapshots can be reused multiple times (the queue would keep such items forever / until needed). Regards, Tommaso > > I probably won't work on this more right now but it's good to know > that it looks feasible. > > -Bertrand >
