Search before asking

• I had searched in the issues and found no similar issues.

Version

main e9b49fd (source build, v1.2.953-nightly-e9b49fde); also v1.2.952-nightly-8862a6285b and v1.2.925-patch-15.

What's Wrong?

With ON_ERROR = ABORT_N, rows that fail to parse are skipped as long as fewer than N errors are seen. That part is
intended. But the result of COPY INTO then reports the file with ERRORS_SEEN = 0 and FIRST_ERROR = NULL, even
though rows were dropped (the docs define ERRORS_SEEN as "Number of error rows in the source file"). With RETURN_FAILED_ONLY = TRUE the file is not listed at all. The same file loaded with
ON_ERROR = CONTINUE correctly reports ERRORS_SEEN = 2 and the first error. Load pipelines that check ERRORS_SEEN /
FIRST_ERROR (or use RETURN_FAILED_ONLY) therefore see a clean load, the file is marked as loaded, and nothing
records that rows are missing. NDJSON behaves the same.

Cause: on_error in src/query/storages/stage/src/read/error_handler.rs. In the OnErrorMode::Continue branch it
calls file_status.add_error(e.error, line). In the OnErrorMode::AbortNum(n) branch, when the error is tolerated, it
only increments on_error_count and returns Ok(()) without recording the error in file_status. Calling
file_status.add_error(...) there as well would make ERRORS_SEEN/FIRST_ERROR (and RETURN_FAILED_ONLY) correct.

How to Reproduce?

CREATE STAGE IF NOT EXISTS s115;
COPY INTO @s115/a.csv FROM (SELECT number::STRING AS id, 'v' AS s FROM numbers(10)
                            UNION ALL SELECT 'bad1', 'x' UNION ALL SELECT 'bad2', 'y')
     FILE_FORMAT = (TYPE = CSV) SINGLE = TRUE USE_RAW_PATH = TRUE;      -- 12 rows, 2 of them not INT

CREATE OR REPLACE TABLE t_abort (id INT, s STRING);
COPY INTO t_abort FROM @s115 FILES = ('a.csv') FILE_FORMAT = (TYPE = CSV) ON_ERROR = ABORT_3;
-- a.csv | 10 | 0 | NULL | NULL          <-- rows_loaded 10, errors_seen 0, first_error NULL
SELECT count(*) FROM t_abort;            -- 10

CREATE OR REPLACE TABLE t_continue (id INT, s STRING);
COPY INTO t_continue FROM @s115 FILES = ('a.csv') FILE_FORMAT = (TYPE = CSV) ON_ERROR = CONTINUE;
-- a.csv | 10 | 2 | Invalid value 'bad1' for column 0 (id Int32 NULL): invalid text for number | 11

CREATE OR REPLACE TABLE t_abort2 (id INT, s STRING);
COPY INTO t_abort2 FROM @s115 FILES = ('a.csv') FILE_FORMAT = (TYPE = CSV) ON_ERROR = ABORT_3 RETURN_FAILED_ONLY = TRUE;
-- (empty result)

Expected: ABORT_3 reports errors_seen = 2 and the first error (as CONTINUE does), and RETURN_FAILED_ONLY lists
the file.

Single node, default settings, HTTP handler.

Are you willing to submit PR?

• Yes I am willing to submit a PR!