Files
greptimedb/tests/cases/standalone/common/flow/flow_incremental_partitioned.result
discord9 28fd796f4e fix(flow): harden incremental read correctness (#8196)
* fix(flow): harden incremental read correctness

Signed-off-by: discord9 <discord9@163.com>

* fix(flow): propagate dirty window options

Signed-off-by: discord9 <discord9@163.com>

* test: more

Signed-off-by: discord9 <discord9@163.com>

* chore: test config api

Signed-off-by: discord9 <discord9@163.com>

* refactor: split gen

Signed-off-by: discord9 <discord9@163.com>

* chore: per review

Signed-off-by: discord9 <discord9@163.com>

* fix: allowlist key

Signed-off-by: discord9 <discord9@163.com>

---------

Signed-off-by: discord9 <discord9@163.com>
2026-06-01 02:48:00 +00:00

111 lines
3.2 KiB
Plaintext

-- Validate that a flow performing an incremental aggregate read on a
-- partitioned source table (multiple regions) only reads memtable data
-- and does NOT re-read source rows that have already been flushed to SST.
CREATE TABLE flow_incr_part_input (
host_id INT,
n INT,
ts TIMESTAMP TIME INDEX,
PRIMARY KEY(host_id)
)
PARTITION ON COLUMNS (host_id) (
host_id < 3,
host_id >= 3
)
WITH (
append_mode = 'true'
);
Affected Rows: 0
CREATE FLOW flow_incr_part SINK TO flow_incr_part_sink
WITH (experimental_enable_incremental_read = 'true')
AS
SELECT
sum(n) AS total,
min(n) AS min_n,
max(n) AS max_n,
date_bin(INTERVAL '1 minute', ts, '2024-01-01 00:00:00') AS time_window
FROM
flow_incr_part_input
GROUP BY
time_window;
Affected Rows: 0
-- ==== Phase 1: initial insert across both partitions ====
INSERT INTO flow_incr_part_input VALUES
(1, 10, '2024-01-01 00:00:00'),
(4, 20, '2024-01-01 00:00:30');
Affected Rows: 2
-- SQLNESS REPLACE (ADMIN\sFLUSH_FLOW\('\w+'\)\s+\|\n\+-+\+\n\|\s+)[0-9]+\s+\| $1 FLOW_FLUSHED |
ADMIN FLUSH_FLOW('flow_incr_part');
+------------------------------------+
| ADMIN FLUSH_FLOW('flow_incr_part') |
+------------------------------------+
| FLOW_FLUSHED |
+------------------------------------+
SELECT total, min_n, max_n, time_window FROM flow_incr_part_sink ORDER BY time_window;
+-------+-------+-------+---------------------+
| total | min_n | max_n | time_window |
+-------+-------+-------+---------------------+
| 30 | 10 | 20 | 2024-01-01T00:00:00 |
+-------+-------+-------+---------------------+
-- ==== Phase 2: flush source table to SST ====
-- Move already checkpointed source rows into SST so the next incremental run
-- must skip them.
ADMIN FLUSH_TABLE('flow_incr_part_input');
+-------------------------------------------+
| ADMIN FLUSH_TABLE('flow_incr_part_input') |
+-------------------------------------------+
| 0 |
+-------------------------------------------+
-- ==== Phase 3: insert new delta across both partitions, same time window ====
INSERT INTO flow_incr_part_input VALUES
(2, 30, '2024-01-01 00:00:15'),
(3, 40, '2024-01-01 00:00:45');
Affected Rows: 2
-- ==== Phase 4: flush flow again (incremental read) ====
-- The flow must only read the new memtable delta from both regions and merge
-- with the existing sink aggregate. If it mistakenly re-reads the SST, the
-- result will be inflated (initial data counted twice).
-- SQLNESS REPLACE (ADMIN\sFLUSH_FLOW\('\w+'\)\s+\|\n\+-+\+\n\|\s+)[0-9]+\s+\| $1 FLOW_FLUSHED |
ADMIN FLUSH_FLOW('flow_incr_part');
+------------------------------------+
| ADMIN FLUSH_FLOW('flow_incr_part') |
+------------------------------------+
| FLOW_FLUSHED |
+------------------------------------+
SELECT total, min_n, max_n, time_window FROM flow_incr_part_sink ORDER BY time_window;
+-------+-------+-------+---------------------+
| total | min_n | max_n | time_window |
+-------+-------+-------+---------------------+
| 100 | 10 | 40 | 2024-01-01T00:00:00 |
+-------+-------+-------+---------------------+
-- Clean up
DROP FLOW flow_incr_part;
Affected Rows: 0
DROP TABLE flow_incr_part_input;
Affected Rows: 0
DROP TABLE flow_incr_part_sink;
Affected Rows: 0