Tensions in Practice: Automatic convergence in tension with concurrent intent¶
Distributed title edits · concurrent local writes
Two sites accept different titles, Blue and Green, before either receives the other’s change. A declared winner rule can make both sites converge on Green. That agreement does not establish that discarding Blue was acceptable. Retaining both candidates protects the opportunity to inspect the conflict, but leaves another decision before the application has one final title.
Reach a usable shared answer
Resolve concurrent writes without requiring a person at every conflict.
Preserve meaningful competing edits
Avoid silently discarding an independently accepted edit whose intent matters.
Why these aims pull against each other
A resolution rule can remove divergence by removing one update. Preserving alternatives instead transfers work from automatic agreement to explicit semantic choice.
Choose an arrangement to see what changes and what remains difficult.
Qualitative paths and conditions, not measured costs, timings or performance guarantees.
What this choice protects
What it costs
When it fits
Compare the arrangements
Pick a declared winner
Apply a deterministic precedence rule that selects Green in this illustrative conflict and propagate that outcome.
- What it protects
- Replicas can settle on one title without waiting for a person.
- What it costs
- Blue disappears even if it represented a meaningful correction or preference.
- When it fits
- Fits replaceable values where losing a concurrent edit is acceptable under an explicit, understood contract.
Illustration note: Green is the declared winner, not a claim about trustworthy physical clocks or which edit was truly later. Both local writes initially succeeded.
Retain the conflict
Preserve both candidates and surface the conflict for a person to choose a final title.
- What it protects
- The competing content remains inspectable before either intent is discarded.
- What it costs
- The application needs conflict storage, review effort and a policy for reads while no single title has been chosen.
- When it fits
- Fits edits important enough to inspect and an application able to represent a pending conflict rather than silently showing one as final.
Illustration note: The diagram stops before resolution. It does not claim human review is automatic convergence or that a reviewer can always satisfy both intents.
What this illustration does—and does not—establish
Eventual Consistency: Convergent versus Correct (scopal) supplies the convergence/correctness distinction; Multi-Leader Replication supplies the winner-versus-review choice. The two invented titles keep the lost information visible.
- Propagation and correct conflict detection are assumed; their implementation is outside the diagram.
- A conflict record can converge while the application-level question remains unsettled.
- This single-valued title is intentionally not a set where simply retaining both would satisfy the application contract.
Source entries
Eventual Consistency
Eventual Consistency: Convergent versus Correct (scopal) supplies the conflict examined here.
Convergent versus Correct (scopal)
T6 — Convergent versus Correct (scopal). Eventual consistency guarantees replicas converge to *one* agreed state, not that the agreed state is the *right* one — a last-write-wins merge converges by silently discarding a concurrent update that mattered. Convergence and correctness are different properties, and the merge rule chooses which updates survive. The failure mode is trusting a converged value as authoritative when the merge policy dropped data to achieve agreement. Diagnostic: does the merge function preserve the semantic intent of every concurrent update, or does it achieve convergence by discarding some? Convergence to a wrong-but-agreed value is still a failure.
Multi-Leader Replication with Conflict Resolution
Supplies explicit conflict strategies and a same-field concurrent-edit example.
Tuning parameters
- Resolution strategy — last-write-wins vs. field merge vs. CRDT vs. human queue: more automation means less operator effort but more risk of silent data loss.
Example
When the link heals, the two versions replicate and collide. The resolution rule decides the outcome deterministically: under last-write-wins the higher timestamp silently wins (and the other edit vanishes); under a field-merge rule the conflict is flagged for the user to pick.