Documentation
¶
Overview ¶
Package attribution carries who an invocation is for, so that a log line, a span or an audit record can name them.
It is tracing baggage and nothing else. No command reads it, the query layer does not read it, and it never reaches the compiler — a depguard rule in .golangci.yml enforces that rather than trusting this comment.
The distinction from internal/dataperm is the whole reason this is a separate package, because the two look alike and must not merge. An attribution field named "user_id" and a dataperm attribute of the same name may hold the same string with entirely different standing: the first is for a human reading logs, the second changes which rows a query returns. A host that wants the caller's email to drive a row rule puts it in dataperm.Attributes explicitly.
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
This section is empty.
Types ¶
type Fields ¶
Fields names the caller of one invocation, as key/value pairs destined for a log or span.
It is a map rather than a struct because the engine genuinely does not interpret any of it, and a host should not need an engine change to carry one more identifier. Hosts put in whatever they can attribute by: a tenant, a user, a session, a request id.
The tenant is also command.CommandContext.TenantID, which is not attribution: canal-query isolates warehouse connections by it, so it changes how a query runs, and is a field the engine passes on, not a label.
Values are strings because the destination is a log field. Rendering is the host's job, which keeps log output predictable and stops a live object being carried across the boundary by accident.