Reference
Glossary
Look up the terms used in SQL Flow and these docs, in alphabetical order, with links to the pages that explain them.
Terms used in SQL Flow and in these docs, in alphabetical order. For the ideas behind them, see Core concepts.
A
Account. The unit that owns agents, brokers, connections and pipelines. A user can belong to several accounts. See Accounts and team.
Agent. The SqlFlow.Agent service that runs pipelines on your machine: initial loads, capture, the spool and apply. See Agents.
Air-gapped deployment. A deployment in which you run the control plane on your own servers, so nothing needs to reach the internet. See Air-gapped deployment.
Always Encrypted. SQL Server's client-side column encryption. See Encrypted data support for what capture and targets do with it.
Apply. Writing captured changes to a target. See How SQL Flow works.
Apply agent. In a relayed pipeline, the agent near the target that receives changes from the capture agent and writes the target. See How relay works.
Approval. The step in which a user accepts a newly registered agent or log broker on the Agents page. Until then it runs nothing. See Agents.
Atomic transaction limits. The Maximum records per transaction and Maximum transaction size (MiB) options of a database target. Apply stops at a larger source transaction rather than split it.
Automatic restart. The agent's restart of a failed pipeline after a growing delay, up to 5 restarts in 30 minutes. See Automatic restart.
B
Binlog. MySQL's binary log, and the MySQL capture mode that reads it. See Binlog.
Bookkeeping schema. The schema in a target database where SQL Flow keeps its own tables, such as the committed position. It is sqlflow by default. See SQL Server target.
C
Capture. Reading changes from the source. See How SQL Flow works.
Capture agent. In a relayed pipeline, the agent near the source that captures changes and ships them to the apply agent. See How relay works.
Capture instance. SQL Server's CDC object for one captured table, with its change table. Native CDC reads it; DirectLog creates one to mark a table. See Native CDC.
Capture mode. How SQL Flow reads changes from a source, for example DirectLog, LogMiner or Native CDC. See Choosing a capture mode.
CDC (change data capture). Capturing the changes made to a database, rather than copying whole tables. SQL Server's own feature of that name is what the Native CDC mode reads.
Collapse row versions (faster). A target option that writes only each row's final state per batch. Untick it when a target must see every change. See PostgreSQL target: Options.
Committed position. The point in the source's change stream that a target has committed. SQL Flow records it on the target and resumes from it after any restart. See High availability: Agents.
Compare. A check of every row of a target against the source, while the pipeline streams. See Compare.
Configure. The dialog in which you set a target's table and column names and its processing rules. See Tables and configure.
Connection. The host, database, login and options SQL Flow uses to reach one database or service. See Saved connections.
Control plane. The SqlFlow.App web application that holds your configuration and sends commands to agents and brokers. See How SQL Flow works.
D
Data directory. The folder in which an agent or broker keeps its identity, keys and, for an agent, the spool. Keep it across restarts. See Agents: The data directory.
Diagnostics. A file for SQL Flow support from Download diagnostics on a pipeline's Actions menu, with no row data, passwords or keys. See Diagnostics for support.
DirectLog. The capture mode that reads the SQL Server transaction log, Oracle redo or the Sybase ASE log through a log broker. See SQL Server DirectLog, Oracle DirectLog and Sybase ASE DirectLog.
Distributed acceleration. Splitting a pipeline between a capture agent and an apply agent linked by a relay. See Distributed acceleration.
Drift. Rows that differ between a target and its source. See Compare.
E
Encryption. A database connection's transport setting. Require (default) encrypts the connection and validates the server certificate; Prefer doesn't validate it, and for some engines can connect unencrypted. Azure SQL and SQL database in Fabric connections always validate. See Encrypted data support.
Epoch. A number that increases each time a pipeline role moves to another agent. A target refuses writes from an older epoch. See High availability overview.
H
Held schema change. A source schema change that a target couldn't apply by itself. Its table pauses for your review while the other tables keep streaming. See Schema changes.
Hybrid deployment. A deployment in which the control plane is hosted on the web and agents and brokers run in your network. See Hybrid deployment.
I
In-flight processing. Rules per target that filter rows, choose and compute columns, and mask values. See In-flight processing.
Initial load. The copy of the existing rows of each selected table, before streaming starts. See Initial load.
J
Join code. A code from Add agent or Add broker with which a new agent or broker registers. It is valid for 15 minutes. See Agents.
K
Key repair. A repair that copies the rows with the keys you list from the source to the target, or deletes them there when the source no longer has them. The UI calls it Repair keys. See Repairing drift.
L
Lag. How far a target is behind the source. See Lag, spool and log retention.
Lag goal. The lag a pipeline is judged against, which sets when its lag figure turns amber or red. Set it on the Lag goal card. See Lag, spool and log retention: Lag goals and colors.
Lease. A role's 30-second grant to one agent, renewed with every heartbeat. See High availability overview.
Log broker. The SqlFlow.LogBroker service on the database host that serves allow-listed log files, read-only, over TLS, only to the agents you allow. See Log brokers.
Log hold. SQL Server DirectLog's hold on the source log, which keeps log records until the target commits them. A pause releases it. See Source log retention.
Logical replication. PostgreSQL's streaming of decoded WAL changes, and the PostgreSQL capture mode that uses it. See Logical replication.
LogMiner. The Oracle capture mode that mines redo with Oracle's DBMS_LOGMNR package over the ordinary connection. See LogMiner.
M
Marking. Enabling SQL Server CDC on a table with sys.sp_cdc_enable_table, or marking a Sybase ASE table for replication with sp_setreptable, so that DirectLog gets full row images in the log. See SQL Server DirectLog and Sybase ASE DirectLog.
Masking. Replacing a value before it reaches a target with a partial mask, a keyed hash or a constant. See Masking.
N
Native CDC. The capture mode that reads SQL Server's or Azure SQL's own CDC change tables. See SQL Server Native CDC.
O
Output columns. The columns a target receives when a rule lists them: kept, computed or masked. See Output and computed columns.
P
Pair & verify. The wizard action that pairs a capture agent with an apply agent and tests the relay connection. See Relay.
Page-header helper. The signed procedure dbo.sqlflow_page_header that a sysadmin installs once in a SQL Server 2016 SP2 or 2017 source database for Reconstruction, and for DirectLog tables with LOB columns or long rows. It reads page headers, never row data. See SQL Server Reconstruction.
Parked. The state of a pipeline whose fault kept coming back after automatic restarts. Its card shows Failed: needs operator, and its banner Failed, needs operator:. See Automatic restart.
Pipeline. One source, the tables selected from it, and one or more targets, run by an agent. See Lifecycle.
Polling. The SQL Server capture mode that reads log records with sys.fn_dblog over the ordinary connection. See Polling.
Prepare target schema. The wizard action that creates missing target tables and keys. SQL Flow never creates tables during a load. See Prepare.
Project. A named group of related pipelines. See Projects.
R
Raw log bytes export. A file of the raw SQL Server log a pipeline stopped at, for SQL Flow support, which an Account admin saves with Save raw log bytes on agent (contains row data)…. It stays on the agent for up to 24 hours. See Diagnostics for support: Save raw log bytes on the agent.
Reconstruction. A capture mode that rebuilds changes from the log without CDC or table marking. An update can carry the row's current value, so the target converges to the source rather than replaying every intermediate version. See SQL Server Reconstruction and Sybase ASE Reconstruction.
Relay. The compressed, mutually authenticated link between a capture agent and an apply agent. See How relay works.
Reload. Reload from source empties a pipeline's target tables and runs the initial load again, from a fresh source position. See Lifecycle.
Replication slot. The PostgreSQL object that keeps WAL until SQL Flow confirms it. SQL Flow creates one per set of captured tables. See Logical replication.
Resync. Bringing one table back in line with the source while the other tables keep streaming. See Repairing drift: Resync a table.
Revoke. The action on the Agents page, or sqlflow agent revoke, that refuses an agent's or broker's credential from then on. Use it for an agent that is lost for good: its pipelines move to its standby or pause. See Agents and brokers.
Row filter. A condition, set in Keep rows where, that decides which rows a target receives. See Row filtering.
S
Schema evolution. How SQL Flow carries source table structure changes to every target, or holds a table for your review. See Schema evolution.
Secret reference. A password source, env:<name> or file:<name>, that the agent resolves on its own host, so the password never reaches the control plane. See Database credentials and secret references.
Source. The database SQL Flow captures changes from. See Supported sources and targets.
Source monitor. A pipeline that captures changes without a target. See Source monitor.
Spool. The agent's durable, encrypted store of captured changes on local disk. See Lag, spool and log retention: The spool.
Standby. A second agent or log broker that takes over when the first one fails. See High availability: Agents and High availability: Log brokers.
Supplemental logging. Oracle's setting that adds column values to redo. DirectLog needs an (ALL) COLUMNS group on every table. See Oracle DirectLog.
Support window. The 12 months after its release date during which an agent or broker release gets new work. See Upgrades and restarts.
T
Target. Where a pipeline writes: a relational database, an event streaming or messaging service, or a data platform. See Supported sources and targets.
TDE (Transparent Data Encryption). Encryption of a database's files on disk. Capture modes that read through a log broker can't decrypt it. See Encrypted data support.
Truncation point. The position before which a database may discard its log. SQL Server DirectLog holds the replication truncation point; Sybase ASE DirectLog and Reconstruction hold the secondary truncation point. See Source log retention.
W
WAL (write-ahead log). PostgreSQL's transaction log. See Logical replication.
Wizard. The New pipeline flow that creates a pipeline step by step: source, target, tables, prepare and review. See Source.
