Pipelines and steps

A pipeline is an ordered sequence of steps declared inside an endpoint block. Each step performs one unit of work and writes its result into the execution context for subsequent steps to read.

Step	Function	Exported outputs
`sql`	Parameterized database queries and mutations	`rows`, `row`, `rows_affected`
`redis`	Cache reads, writes, deletions, counters	`value`
`starlark`	Sandboxed data transformation	`result`
`go`	Invokes a registered native Go function	`result`
`transaction`	Atomic multi-statement SQL execution	Nested SQL outputs
`parallel`	Concurrent branch execution	Nested branch outputs
`respond`	Terminates the pipeline	None

Step execution rules

Unique step labels

Every step except respond and parallel requires a unique name. The label defines the namespace under steps.<name>.

Read-only context history

A step may read outputs written by prior steps, but cannot modify prior outputs or mutate ctx.request.

Fail-fast errors

An unhandled error in any step immediately halts the pipeline and triggers the error handler.

Early termination

Execution stops at the first respond whose condition evaluates to true, or at the first unconditional respond.

Pipelines and steps#

#Step execution rules

#Unique step labels

#Read-only context history