Create a sequence of tasks with a task graph¶

CREATE TASK task_finalizer
  FINALIZE = task_root
  AS SELECT 1;

CREATE OR REPLACE TASK task_root
  SCHEDULE = '1 MINUTE'
  TASK_AUTO_RETRY_ATTEMPTS = 2   --  Failed task graph retries up to 2 times
  SUSPEND_TASK_AFTER_NUM_FAILURES = 3   --  Task graph suspends after 3 consecutive failures
  AS SELECT 1;

CREATE OR REPLACE TASK task_root
  SCHEDULE = '1 MINUTE'
  USER_TASK_TIMEOUT_MS = 60000
  CONFIG='{"environment": "production", "path": "/prod_directory/"}'
  AS SELECT 1;

CREATE OR REPLACE TASK task_a
  USER_TASK_TIMEOUT_MS = 600000
  AFTER task_root
  AS
    BEGIN
      LET VALUE := (SELECT SYSTEM$GET_TASK_GRAPH_CONFIG('path'));
      CREATE TABLE IF NOT EXISTS demo_table(NAME VARCHAR, VALUE VARCHAR);
      INSERT INTO demo_table VALUES('task c path',:value);
    END;

CREATE OR REPLACE TASK task_c
  SCHEDULE = '1 MINUTE'
  USER_TASK_TIMEOUT_MS = 60000
  AS
    BEGIN
      CALL SYSTEM$SET_RETURN_VALUE('task_c successful');
    END;

CREATE OR REPLACE TASK task_d
  USER_TASK_TIMEOUT_MS = 60000
  AFTER task_c
  AS
    BEGIN
      LET VALUE := (SELECT SYSTEM$GET_PREDECESSOR_RETURN_VALUE('task_c'));
      CREATE TABLE IF NOT EXISTS demo_table(NAME VARCHAR, VALUE VARCHAR);
      INSERT INTO demo_table VALUES('Value from predecessor task_c', :value);
    END;

-- Updates the date/time table after the root task completes.
CREATE OR REPLACE TASK task_date_time_table
  USER_TASK_TIMEOUT_MS = 60000
  AFTER task_root
  AS
    BEGIN
      LET VALUE := (SELECT SYSTEM$TASK_RUNTIME_INFO('CURRENT_TASK_GRAPH_ORIGINAL_SCHEDULED_TIMESTAMP'));
      INSERT INTO date_time_table VALUES('order_date',:value);
    END;

-- Create a notebook in the public schema
-- USE DATABASE <database name>;
-- USE SCHEMA <schema name>;

-- task_a: Root task. Starts the task graph and sets basic configurations.
CREATE OR REPLACE TASK task_a
  SCHEDULE = '1 MINUTE'
  TASK_AUTO_RETRY_ATTEMPTS = 2
  SUSPEND_TASK_AFTER_NUM_FAILURES = 3
  USER_TASK_TIMEOUT_MS = 60000
  CONFIG='{"environment": "production", "path": "/prod_directory/"}'
  AS
    BEGIN
      CALL SYSTEM$SET_RETURN_VALUE('task_a successful');
    END;
;

-- task_customer_table: Updates the customer table.
--   Runs after the root task completes.
CREATE OR REPLACE TASK task_customer_table
  USER_TASK_TIMEOUT_MS = 60000
  AFTER task_a
  AS
    BEGIN
      LET VALUE := (SELECT customer_id FROM ref_cust_table
        WHERE cust_name = "Jane Doe";);
      INSERT INTO customer_table VALUES('customer_id',:value);
    END;
;

-- task_product_table: Updates the product table.
--   Runs after the root task completes.
CREATE OR REPLACE TASK task_product_table
  USER_TASK_TIMEOUT_MS = 60000
  AFTER task_a
  AS
    BEGIN
      LET VALUE := (SELECT product_id FROM ref_item_table
        WHERE PRODUCT_NAME = "widget";);
      INSERT INTO product_table VALUES('product_id',:value);
    END;
;

-- task_date_time_table: Updates the date/time table.
--   Runs after the root task completes.
CREATE OR REPLACE TASK task_date_time_table
  USER_TASK_TIMEOUT_MS = 60000
  AFTER task_a
  AS
    BEGIN
      LET VALUE := (SELECT SYSTEM$TASK_RUNTIME_INFO('CURRENT_TASK_GRAPH_ORIGINAL_SCHEDULED_TIMESTAMP'));
      INSERT INTO "date_time_table" VALUES('order_date',:value);
    END;
;

-- task_sales_table: Aggregates changes from other tables.
--   Runs only after updates are complete to all three other tables.
CREATE OR REPLACE TASK task_sales_table
  USER_TASK_TIMEOUT_MS = 60000
  AFTER task_customer_table, task_product_table, task_date_time_table
  AS
    BEGIN
      LET VALUE := (SELECT sales_order_id FROM ORDERS);
      JOIN CUSTOMER_TABLE ON orders.customer_id=customer_table.customer_id;
      INSERT INTO sales_table VALUES('sales_order_id',:value);
    END;
;

CREATE OR REPLACE TASK notify_finalizer
  USER_TASK_TIMEOUT_MS = 60000
  FINALIZE = task_root
AS
  DECLARE
    my_root_task_id STRING;
    my_start_time TIMESTAMP_LTZ;
    summary_json STRING;
    summary_html STRING;
  BEGIN
    --- Get root task ID
    my_root_task_id := (SELECT SYSTEM$TASK_RUNTIME_INFO('CURRENT_ROOT_TASK_UUID'));
    --- Get root task scheduled time
    my_start_time := (SELECT SYSTEM$TASK_RUNTIME_INFO('CURRENT_TASK_GRAPH_ORIGINAL_SCHEDULED_TIMESTAMP')::timestamp_ltz);
    --- Combine all task run info into one JSON string
    summary_json := (SELECT get_task_graph_run_summary(:my_root_task_id, :my_start_time));
    --- Convert JSON into HTML table
    summary_html := (SELECT HTML_FROM_JSON_TASK_RUNS(:summary_json));

    --- Send HTML to email
    CALL SYSTEM$SEND_EMAIL(
        'email_notification',
        'admin@snowflake.com',
        'notification task run summary',
        :summary_html,
        'text/html');
    --- Set return value for finalizer
    CALL SYSTEM$SET_RETURN_VALUE('✅ Graph run summary sent.');
  END

CREATE OR REPLACE FUNCTION get_task_graph_run_summary(my_root_task_id STRING, my_start_time TIMESTAMP_LTZ)
  RETURNS STRING
AS
$$
  (SELECT
    ARRAY_AGG(OBJECT_CONSTRUCT(
      'task_name', name,
      'run_status', state,
      'return_value', return_value,
      'started', query_start_time,
      'duration', duration,
      'error_message', error_message
      )
    ) AS GRAPH_RUN_SUMMARY
  FROM
    (SELECT
      NAME,
      CASE
        WHEN STATE = 'SUCCEED' then '🟢 Succeeded'
        WHEN STATE = 'FAILED' then '🔴 Failed'
        WHEN STATE = 'SKIPPED' then '🔵 Skipped'
        WHEN STATE = 'CANCELLED' then '🔘 Cancelled'
      END AS STATE,
      RETURN_VALUE,
      TO_VARCHAR(QUERY_START_TIME, 'YYYY-MM-DD HH24:MI:SS') AS QUERY_START_TIME,
      CONCAT(TIMESTAMPDIFF('seconds', query_start_time, completed_time),
        ' s') AS DURATION,
      ERROR_MESSAGE
    FROM
      TABLE(my-database.information_schema.task_history(
        ROOT_TASK_ID => my_root_task_id ::STRING,
        SCHEDULED_TIME_RANGE_START => my_start_time,
        SCHEDULED_TIME_RANGE_END => current_timestamp()
      ))
    ORDER BY
      SCHEDULED_TIME)
  )::STRING
$$
;

CREATE OR REPLACE FUNCTION HTML_FROM_JSON_TASK_RUNS(JSON_DATA STRING)
  RETURNS STRING
  LANGUAGE PYTHON
  RUNTIME_VERSION = '3.9'
  HANDLER = 'GENERATE_HTML_TABLE'
AS
$$
import json

def GENERATE_HTML_TABLE(JSON_DATA):
    column_widths = ["320px", "120px", "400px", "160px", "80px", "480px"]

    DATA = json.loads(JSON_DATA)
    HTML = f"""
    <img src="https://docs.snowflake.com/images/logo-sample.png"
      alt="Sample organization logo" height="72">
    <p><strong>Task Graph Run Summary</strong>
      <br>Sign in to Snowsight to see more details.</p>
    <table border="1" style="border-color:#DEE3EA"
      cellpadding="5" cellspacing="0">
      <thead>
        <tr>
    """
    headers = ["Task name", "Run status", "Return value", "Started", "Duration", "Error message"]
    for i, header in enumerate(headers):
        HTML += f'<th scope="col" style="text-align:left; width: {column_widths[i]}">{header.capitalize()}</th>'

    HTML += """
        </tr>
      </thead>
      <tbody>
    """
    for ROW_DATA in DATA:
        HTML += "<tr>"
        for header in headers:
            key = header.replace(" ", "_").upper()
            CELL_DATA = ROW_DATA.get(key, "")
            HTML += f'<td style="text-align:left; width: {column_widths[headers.index(header)]}">{CELL_DATA}</td>'
        HTML += "</tr>"
    HTML += """
      </tbody>
    </table>
    """
    return HTML
$$
;

-- Configuration
-- By default, the notebook creates the objects in the public schema.
-- USE DATABASE <database name>;
-- USE SCHEMA <schema name>;

-- 1. Set the default configurations.
--    Creates a root task ("task_a"), and sets the default configurations
--    used throughout the task graph.
--    Configurations include:
--    * Each task runs after one minute, with a 60-second timeout.
--    * If a task fails, retry it twice. if it fails twice,
--      the entire task graph is considered as failed.
--    * If the task graph fails consecutively three times, suspend the task.
--    * Other environment values are set.

CREATE OR REPLACE TASK task_a
  SCHEDULE = '1 MINUTE'
  USER_TASK_TIMEOUT_MS = 60000
  TASK_AUTO_RETRY_ATTEMPTS = 2
  SUSPEND_TASK_AFTER_NUM_FAILURES = 3
  AS
    BEGIN
      CALL SYSTEM$SET_RETURN_VALUE('task a successful');
    END;
;

-- 2. Use a runtime reflection variable.
--    Creates a child task ("task_b").
--    By design, this example fails the first time it runs, because
--    it writes to a table ("demo_table") that doesn’t exist.
CREATE OR REPLACE TASK task_b
  USER_TASK_TIMEOUT_MS = 60000
  AFTER task_a
  AS
    BEGIN
      LET VALUE := (SELECT SYSTEM$TASK_RUNTIME_INFO('current_task_name'));
      INSERT INTO demo_table VALUES('task b name',:VALUE);
    END;
;

-- 3. Get a task graph configuration value.
--    Creates the child task ("task_c").
--    By design, this example fails the first time it runs, because
--    the predecessor task ("task_b") fails.
CREATE OR REPLACE TASK task_c
  USER_TASK_TIMEOUT_MS = 60000
  AFTER task_b
  AS
    BEGIN
      CALL SYSTEM$GET_TASK_GRAPH_CONFIG('path');
      LET VALUE := (SELECT SYSTEM$GET_TASK_GRAPH_CONFIG('path'));
      INSERT INTO demo_table VALUES('task c path',:value);
    END;
;

-- 4. Get a value from a predecessor.
--    Creates the child task ("task_d").
--    By design, this example fails the first time it runs, because
--    the predecessor task ("task_c") fails.
CREATE OR REPLACE TASK task_d
  USER_TASK_TIMEOUT_MS = 60000
  AFTER task_c
  AS
    BEGIN
      LET VALUE := (SELECT SYSTEM$GET_PREDECESSOR_RETURN_VALUE('TASK_A'));
      INSERT INTO demo_table VALUES('task d: predecessor return value', :value);
    END;
;

-- 5. Create the finalizer task ("task_f"), which creates the missing demo table.
--    After the finalizer completes, the task should automatically retry
--    (see task_a: task_auto_retry_attempts).
--    On retry, task_b, task_c, and task_d should complete successfully.
CREATE OR REPLACE TASK task_f
  USER_TASK_TIMEOUT_MS = 60000
  FINALIZE = task_a
  AS
    BEGIN
      CREATE TABLE IF NOT EXISTS demo_table(NAME VARCHAR, VALUE VARCHAR);
    END;
;

-- 6. Resume the finalizer. Upon creation, tasks start in a suspended state.
--    Use this command to resume the finalizer.
ALTER TASK task_f RESUME;
SELECT SYSTEM$TASK_DEPENDENTS_ENABLE('task_a');

-- 7. Query the task history
SELECT
    name, state, attempt_number, scheduled_from
  FROM
    TABLE(information_schema.task_history(task_name=> 'task_b'))
  LIMIT 5;
;

-- 8. Suspend the task graph to stop incurring costs
--    Note: To stop the task graph, you only need to suspend the root task
--    (task_a). Child tasks don’t run unless the root task is run.
--    If any child tasks are running, they have a limited duration
--    and will end soon.
ALTER TASK task_a SUSPEND;
DROP TABLE demo_table;

-- 9. Check tasks during execution (optional)
--    Run this command to query the demo table during execution
--    to check which tasks have run.
SELECT * FROM demo_table;

-- 10. Demo reset (optional)
--     Run this command to remove the demo table.
--     This causes task_b to fail during its first run.
--     After the task graph retries, task_b will succeed.
DROP TABLE demo_table;