Glossary
The words, in one place
Every term the site leans on, defined in a sentence or two. Terms are underlined the first time they appear on a page, and hovering one shows the same definition without leaving the page.
- ABACSecurity →
- Attribute-based access control. Decisions use attributes of the session, the row, and the environment rather than role membership alone.
- ApplierConsensus →
- The code path that applies a committed log entry on a member. Every member runs the same applier, so a follower's rows come from the same operator path that produced the leader's.
- Bloom filterDatabase →
- A compact probabilistic index in segment metadata that says a value is definitely absent, so a scan skips segments that cannot contain a match.
- BranchZyronLake →
- A copy-on-write branch of a table or database. It starts from the current version, shares every unchanged file, and merges back with conflict resolution.
- Change feedChange streams →
- The per-table log of inserts, updates, and deletes. A consumer reads it and advances its position inside its own transaction, so the position moves only when the consumer commits.
- ChangesetConsensus →
- The sealed record of a write on a grouped node. Savepoints are resolved at capture time, so rows a savepoint rolled back are never part of it. One changeset is one Raft log entry.
- Column maskingSecurity →
- A policy that rewrites a column's value in the projection, with email, phone, SSN, card, hash, partial, and custom masks, for sessions not allowed the raw value.
- CompactionDatabase →
- A background thread moves committed rows from the row heap into .zyr segments, so analytical scans read encoded columns while writes keep landing in the heap.
- Embedded nodeMesh →
- A Zyron node running inside an application's own process, with no database or lake of its own, peered into the mesh like any other node.
- Foreign scanMesh →
- A scan of a remote node's data inside a local query. One SELECT can join a local heap table to a remote publication or a shared lake.
- fsyncDatabase →
- The system call that forces written bytes onto the storage device. Zyron pays one per commit batch, never one per page on the transactional path.
- Group commitPerformance →
- Concurrent transactions share one fsync of the write-ahead log. Latency stays flat as the batch grows, which is where the write throughput comes from.
- Health baselineUpgrades →
- The health measured on a node before it is upgraded. The upgraded node is compared against it and rolls back on its own if it degrades.
- Holdable cursorDatabase →
- A cursor that stays open after the transaction that created it commits, so a client can keep fetching from it.
- Hybrid logical clockZyronLake →
- A timestamp that combines wall-clock time with a logical counter, so events across nodes order consistently even when clocks drift.
- HybridScanDatabase →
- One operator that reads the row heap, .zyr segments, and lake tables in a single pass, so a query never has to choose a store.
- Late materializationDatabase →
- Only rows that survive the predicates are decoded and assembled. Everything else stays in its encoded columns.
- LinearizableConsensus →
- Every read observes the latest committed write, on any node, as if there were a single copy of the data.
- Mandatory access controlSecurity →
- Security labels on data and clearance on sessions, enforced by the engine rather than by grants.
- ManifestZyronLake →
- A periodic checkpoint of the lake log that lists every current segment. A reader opens one manifest instead of replaying the history of versions.
- Morsel-parallelDatabase →
- Work is cut into small ranges, called morsels, that worker threads pick up independently, so one query uses every core.
- MVCCDatabase →
- Multi-version concurrency control. Every row version carries its own visibility, so a reader sees a consistent snapshot without blocking a writer. Zyron's visibility checks run lock-free.
- Object storeZyronLake →
- S3-style storage that lake-holding nodes share. ZyronLake runs there, on a shared filesystem, or on local disk.
- Predicate pushdownDatabase →
- Filters move down the plan to the scan itself, so segments are pruned by their zone maps and rows are rejected before anything is decoded.
- PublicationMesh →
- Tables one instance publishes over the wire and another subscribes to, with exactly-once push. A remote publication can be joined inside a local query.
- QUICMesh →
- A UDP-based transport with TLS 1.3 built in and multiplexed streams. Zyron speaks its wire protocol over both TCP and QUIC.
- QuorumConsensus →
- A majority of the group. A write is acknowledged once a quorum has fsynced it, so it survives the loss of a minority of members.
- RaftConsensus →
- The consensus protocol database nodes use when they join a group. A leader orders writes into a replicated log, and an entry commits once a quorum of members has it.
- Raft indexChange streams →
- An entry's position in the group's log. On a group the change feed is written at the raft index, so a consumed position names the same record on every member.
- ReadIndexConsensus →
- The round trip a follower makes to the leader before answering a read. It learns the current commit index and waits for its own apply to reach it, which keeps follower reads linearizable.
- Rolling upgradeUpgrades →
- New binaries land across a Raft group one node at a time, with the leader drained before its turn and a per-node rollback if health drops below its baseline.
- Row heapDatabase →
- The MVCC row store where fresh writes land, tuned for transactional work. Recent rows live here until compaction moves them into .zyr segments.
- Row-level securitySecurity →
- A per-table predicate the planner adds to every query, so a session only sees the rows its policy allows. It is part of the plan, not a filter in an application tier.
- SCD type 1 and type 2Change streams →
- How APPLY CHANGES lands a stream into a target. Type 1 overwrites the row in place. Type 2 keeps history, closing the old row's validity window and opening a new one.
- Schema epochDatabase →
- A version stamp in each tuple slot that lets ALTER TABLE add, drop, or retype a column without rewriting a single row.
- Snapshot isolationDatabase →
- Every transaction reads from a consistent snapshot taken when it started and never sees another transaction's uncommitted or later writes.
- Three-state privilegesSecurity →
- GRANT, DENY, or unset on every object. DENY always wins, and a column-level setting overrides the table-level one.
- Time travelZyronLake →
- Reading a table as it was, by timestamp with AS OF TIMESTAMP or by committed version with VERSION AS OF.
- Transaction logZyronLake →
- ZyronLake's append-only log, one numbered version file per commit. Data files are immutable, so the log is the only thing that grows.
- Vectorized executionDatabase →
- Operators work on batches of values at a time rather than one row at a time, which keeps the CPU's pipelines and caches busy.
- Verifiable tableSecurity →
- A table whose commits are chained by SHA-256, one entry per commit, with VERIFY TABLE to check the chain and anchored heads to prove it outside the database.
- Versioned envelopeUpgrades →
- The header every persistent file carries, naming its format kind and version, so every reader dispatches on both and older formats migrate on the policy that suits them.
- Write-ahead logDatabase →
- Every commit is written and fsynced to the append-only log before the client is acknowledged. Heap and index pages reach disk later through a background writer, never on the commit path.
- Z-order clusteringZyronLake →
- Space-filling-curve clustering that keeps rows close on several columns at once, so pruning works for more than one predicate.
- Zone mapDatabase →
- The minimum and maximum of each column in a segment, kept in its metadata. A scan skips every segment whose range cannot match the predicate.
- .zyrDatabase →
- Zyron's columnar format. Columns are encoded with dictionary, RLE, bit-pack, and FastLanes codings, filtered without being decoded, and carry bloom filters and zone maps in their segment metadata.
Running in the next five minutes
Prebuilt binaries for Linux and Windows, no external dependencies, one command to a running server.