Query timestamp in append mode yields incorrect result

[v0y4g3r]

What type of bug is this?

Incorrect result

What subsystems are affected?

Standalone mode

Minimal reproduce step

1. Create table with append mode:

create table t (
    ts timestamp time index,
    host string primary key,
    not_pk string,
    val double,
) with (append_mode='true', 'compaction.type'='twcs','compaction.twcs.max_active_window_files'='8','compaction.twcs.max_inactive_window_files'='8');

2. Insert 9 rows and flush

insert into t values
    (0, 'a', '🌕', 1.0),
    (1, 'b', '🌖', 2.0),
    (1, 'a', '🌗', 3.0),
    (1, 'c', '🌘', 4.0),
    (2, 'a', '🌑', 5.0),
    (2, 'b', '🌒', 6.0),
    (2, 'a', '🌓', 7.0),
    (3, 'c', '🌔', 8.0),
    (3, 'd', '🌕', 9.0);

admin flush_table('t');    

3. Insert another 9 rows and flush

insert into t values
    (10, 'a', '🌕', 1.0),
    (11, 'b', '🌖', 2.0),
    (11, 'a', '🌗', 3.0),
    (11, 'c', '🌘', 4.0),
    (12, 'a', '🌑', 5.0),
    (12, 'b', '🌒', 6.0),
    (12, 'a', '🌓', 7.0),
    (13, 'c', '🌔', 8.0),
    (13, 'd', '🌕', 9.0);
admin flush_table('t');    

4. Then select num of total rows and all timestamps

select count(ts) from t;    

select ts from t;

What did you expect to see?

• select count(ts) from t should give 18 since 18 rows are written

+-------------+
| COUNT(t.ts) |
+-------------+
| 18          |
+-------------+

• select ts from t should list all inserted timestamps

+-------------------------+
| ts                      |
+-------------------------+
| 1970-01-01T00:00:00     |
| 1970-01-01T00:00:00.001 |
| 1970-01-01T00:00:00.002 |
| 1970-01-01T00:00:00.002 |
| 1970-01-01T00:00:00.010 |
| 1970-01-01T00:00:00.011 |
| 1970-01-01T00:00:00.012 |
| 1970-01-01T00:00:00.012 |
| 1970-01-01T00:00:00.001 |
| 1970-01-01T00:00:00.002 |
| 1970-01-01T00:00:00.011 |
| 1970-01-01T00:00:00.012 |
| 1970-01-01T00:00:00.001 |
| 1970-01-01T00:00:00.003 |
| 1970-01-01T00:00:00.011 |
| 1970-01-01T00:00:00.013 |
| 1970-01-01T00:00:00.003 |
| 1970-01-01T00:00:00.013 |
+-------------------------+

What did you see instead?

• select count(ts) from t gives 18:

+-------------+
| COUNT(t.ts) |
+-------------+
| 18          |
+-------------+

• But select ts from t only returns 9 rows:

+----------------------------+
| ts                         |
+----------------------------+
| 1970-01-01 00:00:00.010000 |
| 1970-01-01 00:00:00.011000 |
| 1970-01-01 00:00:00.012000 |
| 1970-01-01 00:00:00.012000 |
| 1970-01-01 00:00:00.011000 |
| 1970-01-01 00:00:00.012000 |
| 1970-01-01 00:00:00.011000 |
| 1970-01-01 00:00:00.013000 |
| 1970-01-01 00:00:00.013000 |
+----------------------------+
9 rows in set (0.07 sec)

What operating system did you use?

NA

What version of GreptimeDB did you use?

0.9.2

Relevant log output and stack trace

No response

[v0y4g3r]

This bug was introduced in #4627