Execution messages

Report business information, warnings and errors directly from node SQL. For persistent quality checks with pass/fail counts, use data tests.

Write a message

Add -- @message in node code, or in a shared template to apply it to every linked node. Each block contains one read-only query:

-- @input source
-- @param status: text = "paid"
-- @output orders
SELECT * FROM {{ source }} WHERE status = {{ status }};
-- @message
SELECT 'warning' AS level, count(*) || ' orders without an amount' AS message
FROM {{ orders }} WHERE amount IS NULL HAVING count(*) > 0;

Result contract

The query must return unique level and message columns. Every row emits one message; zero rows emit none. Levels are exactly info, warning, or error; messages must be non-null text. Missing columns, unknown levels and invalid messages fail the node with an explicit error. Additional columns are ignored.

Execution and failure

All output blocks run first. Message blocks then run in declaration order and can reference declared inputs, parameters and any of the newly computed outputs, even if that output is declared later in the file. They obey the node’s declared data access contract and cannot write data or introduce side effects. A source can use its declared connections and resources. On an export, message queries read its inputs and run before any external write, even though the COPY or INSERT statement is written before -- @message in the file.

An error result prevents publication of every new output, retains the previous successful outputs as stale, and blocks descendants. All message blocks are checked before publication or export; warnings alone leave the run in success. Changing a message block invalidates results just like changing other SQL.

Read messages

The Log tab of the results panel lists the messages of the run behind the displayed result; a stale retained result keeps its original messages alongside the latest failure. The canvas shows a warning badge and count; selecting the badge opens the Log. History retains messages and warning totals for each run, including failures and exports, after a restart. A node outside the scope of a later launch keeps its previous result’s messages.

Limits and server logs

Each node execution retains the first 100 messages, followed by “… and N more”. All additional rows are still validated and counted: an error beyond the limit still fails the node. Server logs emit one JSON line per actual message, including run, project, branch, environment, node, level, message and origin. Multiline text is escaped within that line. Connection secrets are redacted in messages, errors, saved history and logs.

See SQL syntax for declarations and results and logs for the execution timeline.