Glossary
Definitions collected from published atlas sheets.
- add-wins
- The conflict policy in which an addition survives a concurrent removal that did not observe it.
- causal context
- A compact record of the dots a replica has observed.
- Claims
- A sequenced write-once key collection where the first accepted claim becomes the committed owner.
- component-wise order
- An order in which one vector is no greater than another in every component.
- concurrency
- The relation between distinct events with no happens-before path in either direction.
- dot
- A unique event identifier made from a replica ID and a local counter.
- eventual consistency
- Each replica sees only the updates it has received, so its current view can be incomplete and can differ from the others. Once updates stop and all remaining messages arrive, every replica eventually converges on the same state.
- G-counter
- A grow-only replicated counter that gives each replica its own nondecreasing component, merges each component by maximum, and sums the components for the total.
- GSet
- A grow-only replicated set whose replicas merge by taking the union of their members.
- happens-before
- The strict relation generated by local event order, message send-to-receive order, and transitivity.
- JsonOt
- A collaborative JSON document that transforms concurrent path operations before it applies them.
- Lamport clock
- A scalar logical clock that assigns a lower timestamp to every causal predecessor.
- last-writer-wins register
- A replicated register that keeps the value with the greatest timestamp and uses the writer ID to break equal-time ties.
- local history
- The ordered events executed at one replica.
- LWWMap
- A replicated map where each key keeps the value or tombstone with the greatest timestamp and writer tie-breaker.
- multi-value register
- A replicated register that retains concurrent writes until a write that observes them supersedes them.
- observation
- Information available to a replica through its own events and received messages.
- observed-remove set
- A replicated set whose remove operation targets only addition tags observed by the removing replica.
- OR-map
- An observed-remove map whose key identities let a concurrent update survive removal.
- OrderedCollection
- A sequenced work queue that grants each item to one accepted acquire operation.
- PactMap
- A coordinated map whose proposed value becomes accepted only after the expected connected clients sign off.
- PN-counter
- A replicated counter that stores increases and decreases in separate grow-only components, then subtracts their totals.
- RegisterCollection
- A sequenced collection of named registers that retains competing versions so readers can choose an atomic or latest-value policy.
- replica
- A local copy of shared data that can change independently and merge messages from other copies.
- sequencer
- A service that assigns ordered sequence numbers to accepted operations and broadcasts the numbered stream to replicas.
- sibling
- A surviving version of a register value with its own write identity and causal history.
- Sluice
- Watershed's in-memory server for one collaborative document. It connects clients, sequences their operations, and lets tests or demos control when messages are delivered.
- TaskManager
- A coordination structure that assigns one client per named task and keeps a FIFO volunteer queue for failover.
- tombstone
- Retained removal metadata that prevents an old addition from becoming live again.
- total order
- An ordering that compares every distinct pair, including pairs that are concurrent.
- TwoPSet
- A two-phase replicated set composed of two GSets: one records additions and one records permanent removals. Each value can move from absent, to present, to permanently removed.
- vector clock
- A logical clock with one counter per tracked process, updated by local increments and receive-time component maxima.