kofre.time

Efficient storage of a set of Time when most stored values are contiguous ranges.

This is a causally consistent wall-time clock that provides a total order in a distributed environment. It tries to be just wall-time clock if possible – this is the case when no updates happen “concurrently” where the concurrency window depends on the accuracy of different clocks (this is on the order of milliseconds even with well synchronized clocks). In case where there is clock drift, it may happen that the current wall-time clock is behind the observed time. Then we fall back to increasing the causal counter. This works, but has a weird “whoever does the most updates is the newest” semantics, which is undesirable in many practical use-cases. The random parameter is used for enforcing totality in cases where the causal values also happen to collide. It is slightly preferrable if random is monotonically increasing on a single replica, as then updates within the same millisecond from the same replica do not need to increment the causal counter.

A Dot is a globally unique point in time. Dots are partially ordered by their time per replicaId.

Essentially a more efficient version of a Set. It typically tracks all dots known within some scope.

This datastructure is used both for implementation of RDTs, as well as for ensuring causality during replication.