DDS12 — instance resource can be reclaimed in READER_DATA_LIFECYCLE QoS section

Clarification to when a instance resource can be reclaimed in READER_DATA_LIFECYCLE QoS section

Summary:

In Section 2.1.3.22 (READER_DATA_LIFECYCLE QoS) the fourth paragraph mention how "the DataReader can only reclaim all resources for instances that instance_state = NOT_ALIVE_NO_WRITERS and for which all samples have been 'taken'".
This should be corrected to state "for instances for which all samples have been 'taken' and either instance_state = NOT_ALIVE_NO_WRITERS or instance_state = NOT_ALIVE_DISPOSED and there are no 'live' writers".
In light of this the statement in the last paragraph stating that once the state becomes NOT_ALIVE_DISPOSED after the autopurge_disposed_samples_delay elapses, "the DataReader will purge all internal information regarding the instance; any untaken samples will also be lost" is not entirely true. If there are other 'live' writers, the DataReader will maintain the state on the instance of which DataWriters are writing to it.

We should change the "will purge all" to "may purge all" or even "will purge". Alternatively, we could describe in further detail when it "will purge all", i.e. when there are no 'live' writers.

The biggest thing here is to decide whether the instance lifecycle can end directly from the NOT_ALIVE_DISPOSED state (as Figure 2-11 currently states) or whether we must force it to go though the NOT_ALOVE_NO_WRITES; that is, in the case where the last writer unregisters a disposed instance do we transition to NOT_ALIVE_NO_WRITERS+NOT_ALIVE_DISPOSED or do we finish the lifecycle directly without notifying the user (as it is indicated now)

We think the current behavior is better because from the application reader point of view, the instance does not exists once it DISPOSED, the fact that we keep the instance state such that we can retain ownership is a detail inside the middleware, so it would be unnatural to get a further indication that the instance (that it no longer knows about) has now no writers.

We suggest the proposed changes should reflect this point of view.

Proposed Resolution:

Make the suggested corrections:

(1) Correct when readers can claim resources to include NOT_ALIVE_DISPOSED state when there are no live writers. So we always reclaim when there are no writers and all the samples for that instance are taken; these samples will include a sentinel mata-sample with an instance state that will be either NOT_ALIVE_NO_WRITERS or NOT_ALIVE_DISPOSED

(2) Clarify that the auto_purge_disposed samples removes only the samples, but not the instance; the instance will only removed in the above case.
Proposed Revised Text:

Section 2.1.3.22 READER_DATA_LIFECYCLE QoS

Replace the paragraph:

Under normal circumstances the DataReader can only reclaim all resources for instances that instance_state = NOT_ALIVE_NO_WRITERS and for which all samples have been 'taken.'

With

Under normal circumstances the DataReader can only reclaim all resources for instances for which there are no writers and for which all samples have been 'taken.' The last sample the DataReader will have taken for that instance will have an instance_state of either NOT_ALIVE_NO_WRITERS or NOT_ALIVE_DISPOSED depending on whether the last writer that had ownership of the instance disposed it or not. Refer to Figure 2-11 for a statechart describing the transitions possible for the instance_state.

In the Paragraph starting with "The autopurge_nowriter_samples_delay defines.."
Replace

once its view_state becomes NOT_ALIVE_NO_WRITERS

With

once its instance_state becomes NOT_ALIVE_NO_WRITERS

Replace the paragraph:

The autopurge_disposed_samples_delay defines the maximum duration for which the DataReader will maintain information regarding an instance once its view_state becomes NOT_ALIVE_DISPOSED. After this time elapses, the DataReader will purge all internal information regarding the instance; any untaken samples will also be lost

With

The autopurge_disposed_samples_delay defines the maximum duration for which the DataReader will maintain samples for an instance once its instance_state becomes NOT_ALIVE_DISPOSED. After this time elapses, the DataReader will purge all samples for the instance.

see above