Skip to content

Vector Clock — Senior

At senior level, focus on this question:

Given two vector clocks, how do you determine whether one happened strictly before the other, or whether they're truly concurrent?

Prerequisite: middle.md.

The comparison rules

Given two vectors V1 and V2:

  • V1 happened-before V2 if every element of V1 is ≤ the corresponding element of V2, and at least one element is strictly <.
  • V1 and V2 are concurrent if neither happened-before the other — meaning some elements of V1 are greater and some are less than the corresponding elements of V2.
flowchart TD V1["V1 = {A:2, B:1}"] V2a["V2 = {A:3, B:1}\n(every element >= V1's,\nat least one strictly >)"] V2b["V2 = {A:1, B:2}\n(A is LESS, B is MORE\nthan V1 - MIXED)"] V1 -->|"happened-before"| V2a V1 -.->|"CONCURRENT\n(neither dominates)"| V2b
def compare(v1, v2):
    less_or_equal = all(v1[n] <= v2[n] for n in v1)
    strictly_less_somewhere = any(v1[n] < v2[n] for n in v1)
    if less_or_equal and strictly_less_somewhere:
        return "v1 happened-before v2"

    v1_greater_or_equal = all(v1[n] >= v2[n] for n in v1)
    strictly_greater_somewhere = any(v1[n] > v2[n] for n in v1)
    if v1_greater_or_equal and strictly_greater_somewhere:
        return "v2 happened-before v1"

    return "concurrent"  # neither dominates the other

Why "concurrent" is a real, meaningful third answer

flowchart LR Concurrent["Two writes are\nCONCURRENT (vector\ncomparison says so)"] --> Meaning["Neither write knew about\nthe other when it happened -\nthey are GENUINELY\nindependent, not just\n'happened close in time'"]

Unlike wall-clock comparison (which always produces some answer, even a meaningless one for events that had no causal relationship), vector clock comparison can correctly say "these two events are concurrent" — a precise, meaningful statement that neither event's originator had any knowledge of the other when it acted. This is exactly the signal a distributed data store needs to detect a genuine write conflict (see BASE & Eventual Consistency — senior): two concurrent writes to the same key must be reconciled (LWW, application- level merge, or a CRDT), because neither can claim to have "come after" the other in any meaningful sense.

🎯 Senior takeaway: vector clocks don't just fix wall-clock comparison's accuracy problem — they add a genuinely new capability: correctly identifying when two events have no causal relationship at all, which a total-ordering mechanism (any clock, logical or physical) cannot express, because a total order always puts one event "before" the other even when that's not a meaningful or true statement about their relationship.

Test yourself

  1. Given V1 = {A:2, B:3, C:1} and V2 = {A:2, B:3, C:2}, is one happened-before the other, and if so which? Justify it using the comparison rule.
  2. Given V1 = {A:2, B:1} and V2 = {A:1, B:2}, why are these concurrent, and what does that concretely mean about the two events they represent?
  3. Why is "concurrent" a capability that a single logical counter (a Lamport clock, which produces a total order) cannot express, while a vector clock can?

Continue to professional.md to see how production systems (Dynamo, Riak) use vector clocks and manage their size-growth problem.