Demo of Conflict-free Replicated Data Views

Both sites writing to different keys in map m:

---

When reading the map using, e.g., the Add-Wins + Multi-Value Register view, each site returns different values:

---

After the Replicate and Merge button is pressed, both sites will return the same data. In this case, the map contains two keys:

Both sites writing to the same key in map m:

---

After Replicate and Merge , the Add-Wins + Multi-Value Register view returns the conflicting values for k3:

---

To get just the last write based on physical time, we can use the Add-Wins + Last-Writer-Wins view. In this case, Site 2 made the last write:

---

Query Present to get the underlying rows:

To model nested structures, we can store a pointer to the inner structure in the outer structure. In the generic tables that we use in this demo, we can store it directly in the data column. For example, the following code can be used to create a map of sets:

---

To read the data, we just need to join using the map values:

---

Alternatively, we can use a query that outputs the result to JSON:

In the map of sets nested structure of the previous example, consider adding an element to set s1 in one site while concurrently deleting key k1 from the map (ensure Replicate and Merge is done first):

---

After Replicate and Merge , even though we have a conflict from a concurrent add and remove, querying the nested structure will always exclude k1:

---

The problem is that the writes are done on different structures, even if they are logically connected. To avoid this, we can force an add of k1 every time there is an addition to s1, in the same transaction. The built-in add_referential_integrity function creates a trigger that automatically does this. First, let's add k1 back to the map:

---

Then, after Replicate and Merge , we will make adds to s1 trigger a write of k1 on m1 (in this demo, it needs to be created in both sites):

---

Now, we can execute the concurrent add and remove again. Inspecting History of Site 1 reveals the two yet-to-be-replicated rows:

---

After Replicate and Merge , the add-wins view MapAwLww returns the updated set, while the remove-wins MapRwLww does not:

---

Likewise, we can have a similar approach for removes, e.g., mark the elements in the set as removed when we delete the key from the map. Finally, to remove the referential integrity trigger, we can use the rmv_referential_integrity function (e.g., rmv_referential_integrity('{m1, k1}', 's1')).