Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Introducing Support for Move Operations in Melda CRDT

Amos Brocco

TL;DR

This work extends Melda b20, a delta-state CRDT for JSON documents, to support move operations across hierarchical structures without introducing duplicates or significant overhead. The key idea is a minimal change to the unflattening step, removing referenced objects from the delta collection after reconstruction to ensure each object is used at most once and avoiding cycles via depth-first reconstruction. The approach preserves existing properties like non-destructive array merging and tombstone-based deletions, and is validated through a multi-client evaluation showing no duplicates or losses during concurrent relocations. The contribution enhances Melda's applicability to complex, offline-first collaborative editing tasks on nested JSON data, and establishes groundwork for further benchmarking and formal verification.

Abstract

In this paper, we present an extension to Melda (a library which implements a general purpose delta state JSON CRDT) to support move operations. This enhancement relies on minimal changes to the underlying logic of the data structure, has virtually no runtime overhead and zero storage overhead compared to the original version of the library, ensuring simplicity while addressing multiple use cases. Although concurrent reordering of the elements in a list was already supported in the original version of the library, moving objects between different containers lead to undesired outcomes, namely duplicate entries. To address this problem we revisited the original approach and introduced the necessary changes to support for relocating elements within a JSON structure. We detail those changes and provide some examples.

Introducing Support for Move Operations in Melda CRDT

TL;DR

This work extends Melda b20, a delta-state CRDT for JSON documents, to support move operations across hierarchical structures without introducing duplicates or significant overhead. The key idea is a minimal change to the unflattening step, removing referenced objects from the delta collection after reconstruction to ensure each object is used at most once and avoiding cycles via depth-first reconstruction. The approach preserves existing properties like non-destructive array merging and tombstone-based deletions, and is validated through a multi-client evaluation showing no duplicates or losses during concurrent relocations. The contribution enhances Melda's applicability to complex, offline-first collaborative editing tasks on nested JSON data, and establishes groundwork for further benchmarking and formal verification.

Abstract

In this paper, we present an extension to Melda (a library which implements a general purpose delta state JSON CRDT) to support move operations. This enhancement relies on minimal changes to the underlying logic of the data structure, has virtually no runtime overhead and zero storage overhead compared to the original version of the library, ensuring simplicity while addressing multiple use cases. Although concurrent reordering of the elements in a list was already supported in the original version of the library, moving objects between different containers lead to undesired outcomes, namely duplicate entries. To address this problem we revisited the original approach and introduced the necessary changes to support for relocating elements within a JSON structure. We detail those changes and provide some examples.

Paper Structure

This paper contains 11 sections, 14 figures, 4 algorithms.

Table of Contents

  1. Introduction
  2. Melda
  3. Non-destructive array merging
  4. Relocating objects
  5. Concurrent relocation without reparenting
  6. Concurrent relocation with reparenting
  7. Concurrent relocation (with reparenting) and deletion
  8. Introducing support for move operations
  9. Evaluation
  10. Related Work
  11. Conclusions

Figures (14)

  • Figure 1: Hierarchical structure with unordered children and corresponding JSON (version 0).
  • Figure 2: Hierarchical structure with ordered children and corresponding JSON (version 0).
  • Figure 3: Concurrent updates with creation and deletion but no reparenting.
  • Figure 4: Possible structure after merge (unordered children): either one of the versions of the root object is retained.
  • Figure 5: Structure after merge (ordered children): changes made to the root object's array are merged.
  • ...and 9 more figures