KV preserve system xattrs on target clusters during replication

History:

Part of the deliverable for XDCR / Mobile co-existence is that the system (mobile) xattribute that exists on the target document should be preserved.
XDCR’s proposal originally was to perform the XDCR composition method to preserve mobile xattribute.

Initially, MB-52947 was filed from the context of ensuring XDCR optimistic replication would be functional. However, from experimentation, we concluded that optimistic replication was no longer a XDCR requirement for initial offering.

However, we discovered a race condition that could exist if MB-52947 was not a implemented. This was documented and it mostly impacts Sync-Gateway and could take place during the SGW import process. Adam Fraser has summarized the issue on the mobile side

The team leads met a while back and discussed the concerns and asked the KV team about potentially implementing MB-52947.
As of date (7/20), MB-52947 has not been approved for trinity.

There were two concerns that were unresolved.
(1) KV does not want to know about mobile xattrs - special handling for mobile xattrs should be avoided
(2) KV wants to update CAS if it has to modify mobile xattrs (and this would cause ping-pong issue for mobile import) - changing document content without CAS change is not ideal

(Dave Rigby mentioned potentially using revID field, but I don't recall the specific conclusion on it)

Request Summary:

The ask of this MB is the same as MB-52947’s description, but asking from the perspective for Mobile team’s concern instead of XDCR’s performance concern (which is not a concern at this point). MB-52947 has been in the context of XDCR and its performance, and the discussion there are valid but not necessary for decision making in this context.

This MB is to be a place of discussion and hopefully to move towards a solution that will address the mobile team’s need, particularly number 2.

Additionally, it was discussed during our Trinity meeting that all 3 teams should come to an alignment in design. The intention is that we can reach a design agreement that addresses all stakeholders’s concerns without leaving any kicking any “unanswered questions” down the road.