Messages

The complete headline project includes query SQL, output YAML, conditional sending and profiles. The report reference lists shared keys; the glossary defines message versus file output.

Many people who depend on a report never open the file: they want yesterday’s revenue, how many payments failed, whether the month closed. A message output turns a report’s query results into a short headline, a title plus a few lines of text, and posts it to Slack, Microsoft Teams, Google Chat or an inbox. It needs no plugin of its own: the message format is built into DRE. New in 0.3.

A message is written to the target path as a .md file too, so any file destination (a Databricks Volume, S3, SFTP, a local folder) delivers it unchanged and keeps a history of what was sent.

A daily headline

Aggregate in SQL, and let the message read the result:

-- reports/finance/daily/headline.sql
select
  sum(amount) filter (where day = date '{{ run.date.prev_day.iso }}') as revenue,
  sum(amount) filter (where day = date '{{ run.date.prev_day.iso }}') /
    nullif(sum(amount) filter (where day = date '{{ run.date.add(days=-8).iso }}'), 0) - 1 as change
from {{ source('sales', 'orders') }}

reports/finance/daily/daily.yml

queries: [headline]
output:
  format: message
  title: "Revenue {{ run.date.prev_day.iso }}"
  text: |
    Revenue yesterday: **{{ results.headline.value | currency('EUR') }}**
    ({{ results.headline.first.change | percent | signed }} on the same day last week)
  destination:
    - {profile: team_slack, channel: "#finance"}

In Slack this arrives as:

Revenue 2026-10-03 Revenue yesterday: €12,340 (+4.1% on the same day last week)

format: message alone, with no text:, already works: the default template writes one block per query (a single value as revenue: 12,340, one row as column: value lines, several rows as a list of up to ten with + N more).

What a message template reads

For each query of the output, results.<query> has:

value	the first column of the first row
first.<column>	a column of the first row
rows	the rows, each row.<column> or row[0] (at most max_rows, default 1,000)
row_count	the true number of rows, even past max_rows
columns	the column names
sets[n]	a result set by index (one .sql file makes one, so sets[-1] is the same as the query)

Values keep their types: numbers stay numbers, dates are DRE dates, nulls are none. The number filters make them readable: number, percent, signed, currency and compact, in the project’s locale:. var(), run.* and your macros work as everywhere.

The text is a small Markdown subset that every service can show: **bold**, *italic*, [links](https://...), `code` and - bullets. No headings or tables. Each destination translates it, and every value the template prints is escaped, so a customer called acme_corp* can’t break the formatting. Keep a longer template in a file with file: messages/daily.md instead of text:.

A headline plus the file

output: can be a list. The queries run once, every output is made from the same results, and file outputs are delivered before messages, so a message can link to what was delivered:

queries: [headline, detail]
output:
  - name: workbook
    format: xlsx
    queries: [detail]                 # the headline query isn't a tab in the workbook
    destination:
      - {profile: reports_s3, path: "s3://reports/daily-{{ run.date.yyyymmdd }}.xlsx"}
  - name: headline
    format: message
    queries: [headline]               # the message doesn't read the 10k-row detail
    text: |
      Revenue yesterday: **{{ results.headline.value | currency('EUR') }}**
      Full report: {{ outputs.workbook.location }}
    destination:
      - {profile: team_slack, channel: "#finance"}
      - {profile: finance_mail, to: cfo@example.com, attach: [workbook]}

outputs.<name> has location (where it was delivered), files and status. attach: sends another output’s files along with the message, to destinations that take both: the email arrives with the headline in its body and the workbook attached; in Slack the file goes in the same post. Teams and Google Chat take messages only, so link the file instead.

A per-row message (for_each) attaching a per-row output with the same driver and key sends each row its own file.

Sending only when it matters

Any output can carry when:, a condition over the results. When it’s false the output is skipped: nothing is written or sent, the run output says so, and run_results.json records it as skipped. It’s not a failure.

queries: [failed_payments]
output:
  format: message
  when: "results.failed_payments.row_count > 0"
  title: "Failed payments {{ run.date.iso }}"
  text: |
    **{{ results.failed_payments.row_count }}** payments failed:
    {% for p in results.failed_payments.rows[:10] %}- {{ p.customer }}: {{ p.amount | currency('EUR', 2) }}
    {% endfor %}
  destination: {profile: ops_teams}

A message whose text renders empty is skipped the same way, so {% if %} inside text: works as a condition. when: works on file outputs too (“only send the CSV if there are rows”).

Each run decides from its own results. DRE keeps nothing between runs: there’s no “alert once”, no deduplication and no escalation. A report that should say “revenue dropped” says it every day the data shows it. Alerting on whether a run succeeded belongs to your orchestrator or the Platform.

Before sending

invalid-output-option identifies a malformed message template or a query name absent from the output. invalid-destination-option can identify attach naming no file output, or a message-only destination that cannot accept it. A runtime render-failed often means the message expects a non-null first row; aggregate in SQL or handle none explicitly. See error codes and preview after repairing the cause.

• dre validate checks every template compiles, that results.<x> names one of the output’s queries and outputs.<x> another output, and that a file output or attach: doesn’t go to a destination that only takes messages.
• dre run <report> --preview prints each message (title, text, its length against each destination’s limit, and whether when: passed) and sends nothing. Numbers then come from the row sample.
• dre validate -s <report> lists every output with its destinations; production targets stand out.

A real run logs one line per message sent, and run_results.json keeps its full title and text.

Destinations

Destination	Takes	Limit
slack	messages and files	4,000 characters	over the limit: cut short, full .md attached
email	messages and files		HTML body with a plain-text alternative; subject from the title
teams	messages only	15,000 characters	a Workflows webhook
google_chat	messages only	4,000 characters	a space webhook
any other	the .md file		s3, sftp, databricks Volumes, local, …

See the message format for every option, and practices for advice.