Add option to output numeric data types as string. #255

shubhamdhama · 2022-11-04T07:27:00Z

Data types like numeric, real, double precision supports Infinity, -Infinity and NaN values. Currently, these values output as null because JSON specification does not recognize them as valid numeric values. This will create problems for the users of wal2json who need these values to maintain data integrity.

Tests
Added tests and tested against Postgres 9.6, 10, 11, 12, 13, and 14.

Fixes: #245

shubhamdhama · 2022-11-04T08:15:20Z

I tested this option numerics-as-string with pgcopydb and it fixes dimitri/pgcopydb#127 (comment)

Before this change

Start pgcopydb clone --follow in other window. While initial load is happening, check source and insert some rows for prefetch,

$ psql $PGCOPYDB_SOURCE_PGURI
psql (14.5 (Ubuntu 14.5-2.pgdg20.04+2), server 12.12 (Ubuntu 12.12-1.pgdg20.04+1))

postgres=# select * from table_integer;
 a  |   b    |      c      |          d
----+--------+-------------+----------------------
  9 |  32767 |  2147483647 |  9223372036854775807
 10 | -32768 | -2147483648 | -9223372036854775808
(2 rows)

postgres=# select * from table_decimal ;
     a     |         b         |                    c
-----------+-------------------+------------------------------------------
  Infinity |          Infinity |
 -Infinity |         -Infinity |
       NaN |               NaN |                                      NaN
   123.456 |  123456789.012345 |  1234567890987654321.1234567890987654321
  -123.456 | -123456789.012345 | -1234567890987654321.1234567890987654321
(5 rows)

postgres=# INSERT INTO table_integer (b, c, d) VALUES(32767, 2147483647, 9223372036854775807);
INSERT 0 1
postgres=# INSERT INTO table_integer (b, c, d) VALUES(-32768, -2147483648, -9223372036854775808);
INSERT 0 1
postgres=# INSERT INTO table_decimal (a, b) VALUES('Infinity', 'Infinity');
INSERT 0 1
postgres=# INSERT INTO table_decimal (a, b) VALUES('-Infinity', '-Infinity');
INSERT 0 1
postgres=# INSERT INTO table_decimal (a, b, c) VALUES('NaN', 'NaN', 'NaN');
INSERT 0 1
postgres=# INSERT INTO table_decimal (a, b, c) VALUES(123.456, 123456789.012345, 1234567890987654321.1234567890987654321);
INSERT 0 1
postgres=# INSERT INTO table_decimal (a, b, c) VALUES(-123.456, -123456789.012345, -1234567890987654321.1234567890987654321);
INSERT 0 1

After initial load has finished (in other window) and before apply has started, check the target. This should reflect the inital load data,

$ psql $PGCOPYDB_TARGET_PGURI
psql (14.5 (Ubuntu 14.5-2.pgdg20.04+2), server 13.8 (Ubuntu 13.8-1.pgdg20.04+1))

postgres=# select * from table_integer;
 a  |   b    |      c      |          d
----+--------+-------------+----------------------
  9 |  32767 |  2147483647 |  9223372036854775807
 10 | -32768 | -2147483648 | -9223372036854775808
(2 rows)

postgres=# select * from table_decimal ;
     a     |         b         |                    c
-----------+-------------------+------------------------------------------
  Infinity |          Infinity |
 -Infinity |         -Infinity |
       NaN |               NaN |                                      NaN
   123.456 |  123456789.012345 |  1234567890987654321.1234567890987654321
  -123.456 | -123456789.012345 | -1234567890987654321.1234567890987654321
(5 rows)

Set endpos to force apply the sql file generated during online phase,

shdhama@shdhama:~
$ $PGCOPYDB/src/bin/pgcopydb/pgcopydb stream sentinel set endpos --current -v
13:21:07 28812 INFO   Running pgcopydb version 0.9.13.g897182b.dirty from "/home/shdhama/pg/pgcopydb/src/bin/pgcopydb/pgcopydb"
13:21:07 28812 NOTICE [SOURCE] BEGIN;
13:21:07 28812 NOTICE [SOURCE] select current_setting('server_version'),        current_setting('server_version_num')::integer;
13:21:07 28812 NOTICE [SOURCE] Postgres version 12.12 (Ubuntu 12.12-1.pgdg20.04+1) (120012)
13:21:07 28812 NOTICE [SOURCE] update pgcopydb.sentinel set endpos = pg_current_wal_flush_lsn();
13:21:07 28812 NOTICE [SOURCE] COMMIT;
13:21:07 28812 NOTICE [SOURCE] select startpos, endpos, apply, write_lsn, flush_lsn, replay_lsn   from pgcopydb.sentinel;
13:21:07 28812 INFO   pgcopydb sentinel endpos has been set to 6/4E092878
6/4E092878

Now again check target,

shdhama@shdhama:~
$ psql $PGCOPYDB_TARGET_PGURI
psql (14.5 (Ubuntu 14.5-2.pgdg20.04+2), server 13.8 (Ubuntu 13.8-1.pgdg20.04+1))
SSL connection (protocol: TLSv1.3, cipher: TLS_AES_256_GCM_SHA384, bits: 256, compression: off)
Type "help" for help.

postgres=# select * from table_decimal ;
     a     |         b         |                    c
-----------+-------------------+------------------------------------------
  Infinity |          Infinity |
 -Infinity |         -Infinity |
       NaN |               NaN |                                      NaN
   123.456 |  123456789.012345 |  1234567890987654321.1234567890987654321
  -123.456 | -123456789.012345 | -1234567890987654321.1234567890987654321
           |                   |
           |                   |
           |                   |
   123.456 |         123457000 |                      1234570000000000000
  -123.456 |        -123457000 |                     -1234570000000000000
(10 rows)

postgres=# select * from table_integer;
 a  |   b    |      c      |          d
----+--------+-------------+----------------------
  9 |  32767 |  2147483647 |  9223372036854775807
 10 | -32768 | -2147483648 | -9223372036854775808
 11 |  32767 |  2147480000 |  9223370000000000000
 12 | -32768 | -2147480000 | -9223370000000000000
(4 rows)

In table_decimal, the first 4 rows were migrated during inital load and last 4 rows were migrated online. We can see NaN, Infinity and -Infinity are coming as null.

After

Start pgcopydb clone --follow in other window. While initial load is happening, check source and insert some rows for prefetch,

$ psql $PGCOPYDB_SOURCE_PGURI
psql (14.5 (Ubuntu 14.5-2.pgdg20.04+2), server 12.12 (Ubuntu 12.12-1.pgdg20.04+1))

postgres=# select * from table_decimal ;
     a     |         b         |                    c
-----------+-------------------+------------------------------------------
  Infinity |          Infinity |
 -Infinity |         -Infinity |
       NaN |               NaN |                                      NaN
   123.456 |  123456789.012345 |  1234567890987654321.1234567890987654321
  -123.456 | -123456789.012345 | -1234567890987654321.1234567890987654321
(5 rows)

postgres=# select * from table_integer;
 a |   b    |      c      |          d
---+--------+-------------+----------------------
 5 |  32767 |  2147483647 |  9223372036854775807
 6 | -32768 | -2147483648 | -9223372036854775808
(2 rows)

postgres=# INSERT INTO table_integer (b, c, d) VALUES(32767, 2147483647, 9223372036854775807);
INSERT 0 1
postgres=# INSERT INTO table_integer (b, c, d) VALUES(-32768, -2147483648, -9223372036854775808);
INSERT 0 1
postgres=# INSERT INTO table_decimal (a, b) VALUES('Infinity', 'Infinity');
INSERT 0 1
postgres=# INSERT INTO table_decimal (a, b) VALUES('-Infinity', '-Infinity');
INSERT 0 1
postgres=# INSERT INTO table_decimal (a, b, c) VALUES('NaN', 'NaN', 'NaN');
INSERT 0 1
postgres=# INSERT INTO table_decimal (a, b, c) VALUES(123.456, 123456789.012345, 1234567890987654321.1234567890987654321);
INSERT 0 1
postgres=# INSERT INTO table_decimal (a, b, c) VALUES(-123.456, -123456789.012345, -1234567890987654321.1234567890987654321);
INSERT 0 1

After initial load has finished (in other window) and before apply has started, check the target. This should reflect the inital load data,

shdhama@shdhama:~
$ psql $PGCOPYDB_TARGET_PGURI
psql (14.5 (Ubuntu 14.5-2.pgdg20.04+2), server 13.8 (Ubuntu 13.8-1.pgdg20.04+1))

postgres=# select * from table_decimal ;
     a     |         b         |                    c
-----------+-------------------+------------------------------------------
  Infinity |          Infinity |
 -Infinity |         -Infinity |
       NaN |               NaN |                                      NaN
   123.456 |  123456789.012345 |  1234567890987654321.1234567890987654321
  -123.456 | -123456789.012345 | -1234567890987654321.1234567890987654321
(5 rows)

postgres=# select * from table_integer;
 a |   b    |      c      |          d
---+--------+-------------+----------------------
 5 |  32767 |  2147483647 |  9223372036854775807
 6 | -32768 | -2147483648 | -9223372036854775808
(2 rows)

Set endpos to force apply the sql file generated during online phase,

shdhama@shdhama:~
$ $PGCOPYDB/src/bin/pgcopydb/pgcopydb stream sentinel set endpos --current -v
13:14:41 28041 INFO   Running pgcopydb version 0.9.13.g897182b.dirty from "/home/shdhama/pg/pgcopydb/src/bin/pgcopydb/pgcopydb"
13:14:41 28041 INFO   pgcopydb sentinel endpos has been set to 6/4E072488
6/4E072488

Now again check target,

shdhama@shdhama:~
$ psql $PGCOPYDB_TARGET_PGURI
psql (14.5 (Ubuntu 14.5-2.pgdg20.04+2), server 13.8 (Ubuntu 13.8-1.pgdg20.04+1))

postgres=# select * from table_decimal ;
     a     |         b         |                    c
-----------+-------------------+------------------------------------------
  Infinity |          Infinity |
 -Infinity |         -Infinity |
       NaN |               NaN |                                      NaN
   123.456 |  123456789.012345 |  1234567890987654321.1234567890987654321
  -123.456 | -123456789.012345 | -1234567890987654321.1234567890987654321
  Infinity |          Infinity |
 -Infinity |         -Infinity |
       NaN |               NaN |                                      NaN
   123.456 |  123456789.012345 |  1234567890987654321.1234567890987654321
  -123.456 | -123456789.012345 | -1234567890987654321.1234567890987654321
(10 rows)

postgres=# select * from table_integer;
 a |   b    |      c      |          d
---+--------+-------------+----------------------
 5 |  32767 |  2147483647 |  9223372036854775807
 6 | -32768 | -2147483648 | -9223372036854775808
 7 |  32767 |  2147483647 |  9223372036854775807
 8 | -32768 | -2147483648 | -9223372036854775808
(4 rows)

NaN, Infinity and -Infinity is replicated properly 🎉

eulerto · 2022-11-05T14:56:00Z

Based on the discussing we have in #245 , this PR needs some adjustments. It should cover all numeric data types. Use only one test file. There are some unrelated changes (blank spaces), remove them. A good name for the new parameter is numeric-data-types-as-string. The README should mention it such as:

numeric-data-types-as-string: use strings for numeric data types. JSON specification does not recognize Infinity and NaN as valid numeric values. There might be potential interoperability problems for double precision numbers. Default is false.

Data types like `numeric`, `real`, `double precision` supports `Infinity`, `-Infinity` and `NaN` values. Currently these values output as `null` because JSON specification does not recognize them as valid numeric values. This will create problems for the users of wal2json who need these values to maintain data integerity.

shubhamdhama · 2022-11-07T09:33:18Z

@eulerto I have updated the PR, please review.

It should cover all numeric data types.

I think the current switch case covers all numeric data types. Please correct me if I'm wrong.

There are some unrelated changes (blank spaces), remove them. A good name for the new parameter is numeric-data-types-as-string. The README should mention

Fixed.

shubhamdhama · 2022-11-11T05:31:26Z

hi @eulerto, gentle ping on this. Thanks!

nakatlam · 2022-11-15T00:53:12Z

sql/numeric_data_types_as_string.sql

+i	tsvector
+);
+
+SELECT 'init' FROM pg_create_logical_replication_slot('regression_slot', 'wal2json');


You may want to add illustrations/outputs of these queries with and without using the option 'numeric-data-types-as-string'.

nakatlam · 2022-11-15T01:00:28Z

wal2json.c

@@ -1257,8 +1272,9 @@ tuple_to_stringinfo(LogicalDecodingContext *ctx, TupleDesc tupdesc, HeapTuple tu
 			 * Data types are printed with quotes unless they are number, true,


update comment

eulerto · 2022-11-15T18:40:19Z

sql/numeric_data_types_as_string.sql

+COMMIT;
+
+SELECT data FROM pg_logical_slot_peek_changes('regression_slot', NULL, NULL, 'format-version', '1', 'pretty-print', '1', 'numeric-data-types-as-string', '1');
+SELECT data FROM pg_logical_slot_peek_changes('regression_slot', NULL, NULL, 'format-version', '2', 'pretty-print', '1', 'numeric-data-types-as-string', '1');


The option pretty-print has no effect for v2. Remove it. You should duplicate both pg_logical_slot_peek_changes to provide the output without this option. It is also recommended to drop the tables you created at the end.

SELECT data FROM pg_logical_slot_peek_changes('regression_slot', NULL, NULL, 'format-version', '1', 'pretty-print', '1', 'numeric-data-types-as-string', '1'); SELECT data FROM pg_logical_slot_peek_changes('regression_slot', NULL, NULL, 'format-version', '1', 'pretty-print', '1'); SELECT data FROM pg_logical_slot_peek_changes('regression_slot', NULL, NULL, 'format-version', '2', 'numeric-data-types-as-string', '1'); SELECT data FROM pg_logical_slot_peek_changes('regression_slot', NULL, NULL, 'format-version', '2');

Make sense, done.

eulerto · 2022-11-22T12:04:15Z

I have a few additional comments:

README: use strings for numeric... -> use string for numeric ...
README: there is an additional line that should be removed: output numeric data types as string. Default is false.
tests: remove table table_others. It doesn't add nothing. It is already covered by other test cases.
comment: "Data types are printed..." You change this comment but doesn't add nothing. Revert it. Keep the last paragraph "Except to this ...".
spaces: don't mix spaces and tabs. JsonDecodingData and other variables should be aligned with tabs.

shubhamdhama · 2022-11-22T13:02:25Z

@eulerto thank you for the review, I've addressed all the comments.

Add option to output numeric data types as string. (eulerto#255)

thecatontheflat · 2023-05-20T15:16:57Z

This fix is not available in the Debian/Ubuntu package right? Needs to be built from the source?

eulerto · 2023-05-21T22:47:49Z

No. It will be in the next release.

hanefi · 2023-11-30T20:12:45Z

@eulerto Any timeline of when the next release will be?

This change adds a new option `--wal2json-numeric-as-string` that changes wal2json plugin output format to print numeric data types as strings. This is accomplished by passing the `--numeric-data-types-as-string` option to wal2json plugin. This is useful to prevent precision loss when using wal2json plugin to stream changes from a database that uses numeric data types. wal2json plugin version that supports `--numeric-data-types-as-string` option is required to use this pgcopydb option. As of today there is no official wal2json release that supports this option, but it is available on master branch of the project. Relevant changes in wal2json plugin is at eulerto/wal2json#255

* Add option to output numeric as string on wal2json This change adds a new option `--wal2json-numeric-as-string` that changes wal2json plugin output format to print numeric data types as strings. This is accomplished by passing the `--numeric-data-types-as-string` option to wal2json plugin. This is useful to prevent precision loss when using wal2json plugin to stream changes from a database that uses numeric data types. wal2json plugin version that supports `--numeric-data-types-as-string` option is required to use this pgcopydb option. As of today there is no official wal2json release that supports this option, but it is available on master branch of the project. Relevant changes in wal2json plugin is at eulerto/wal2json#255 * Add env to output numeric as string on wal2json PGCOPYDB_WAL2JSON_NUMERIC_AS_STRING can be set to a boolean value that will be used to determine if pgcopydb should set the wal2json option `--numeric-data-types-as-string`. In passing, also add the PGCOPYDB_OUTPUT_PLUGIN env variable to all relevant pages of our documentation.

apduvuri · 2023-12-28T11:48:46Z

@eulerto Any timeline of when the next release will be. Most of the customers want to install wal2json from the Linux repositories for their production environment workloads and it is a blocker for them

hsinghjkaur · 2024-01-09T22:26:45Z

@eulerto - I would really appreciate if you could share the timeline for the upcoming release that includes this fix as many customers experience it and is long awaited.

eulerto · 2024-01-11T01:30:48Z

New version will be released after #273 is fixed.

hsinghjkaur · 2024-01-25T16:36:00Z

New version will be released after #273 is fixed.

Thank you for the update! Do you have any ETA?

eulerto · 2024-04-25T15:30:20Z

It took some time but the new version was released including this feature. Enjoy!

eulerto mentioned this pull request Nov 4, 2022

Output NUMERIC type as string in JSON #245

Closed

eulerto added the feature label Nov 5, 2022

shubhamdhama force-pushed the add-numerics-as-strings-option branch from 1a25952 to 94254f8 Compare November 7, 2022 09:15

nakatlam reviewed Nov 15, 2022

View reviewed changes

eulerto reviewed Nov 15, 2022

View reviewed changes

shubhamdhama force-pushed the add-numerics-as-strings-option branch from f4e89d4 to cc73952 Compare November 17, 2022 16:03

shubhamdhama force-pushed the add-numerics-as-strings-option branch from bc26a84 to a54d252 Compare November 22, 2022 12:59

Address review comments.

b598c1e

shubhamdhama force-pushed the add-numerics-as-strings-option branch from a54d252 to b598c1e Compare November 22, 2022 13:04

eulerto merged commit bb7cd50 into eulerto:master Nov 22, 2022

shubhamdhama deleted the add-numerics-as-strings-option branch November 22, 2022 13:45

alejandrosanchezcabana added a commit to streamsets/wal2json that referenced this pull request Jan 18, 2023

Merge pull request #10 from eulerto/master

7bd77a4

Add option to output numeric data types as string. (eulerto#255)

eulerto mentioned this pull request Feb 9, 2023

NaN values are received as "null" for NUMERIC type while capturing change data #265

Open

hanefi mentioned this pull request Nov 30, 2023

Possible precision loss for numeric columns dimitri/pgcopydb#552

Closed

hanefi mentioned this pull request Dec 13, 2023

Official instructions to build from source for production use #274

Closed

hanefi mentioned this pull request Dec 13, 2023

Add option to output numeric as string on wal2json dimitri/pgcopydb#576

Merged

dimitri mentioned this pull request Jan 10, 2024

pgcopydb v0.14: ERRORS with pgcopydb clone --follow --plugin test_decoding dimitri/pgcopydb#623

Open

arajkumar mentioned this pull request Jan 12, 2024

Problem: float values loses precision after SQL transformation dimitri/pgcopydb#631

Closed

yichao-figma mentioned this pull request Apr 12, 2024

include-pk can't work with identity full? #273

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add option to output numeric data types as string. #255

Add option to output numeric data types as string. #255

shubhamdhama commented Nov 4, 2022 •

edited

shubhamdhama commented Nov 4, 2022

eulerto commented Nov 5, 2022

shubhamdhama commented Nov 7, 2022 •

edited

shubhamdhama commented Nov 11, 2022

nakatlam Nov 15, 2022 •

edited

shubhamdhama Nov 15, 2022

nakatlam Nov 15, 2022

shubhamdhama Nov 15, 2022

eulerto Nov 15, 2022

shubhamdhama Nov 17, 2022

eulerto commented Nov 22, 2022

shubhamdhama commented Nov 22, 2022

thecatontheflat commented May 20, 2023

eulerto commented May 21, 2023

hanefi commented Nov 30, 2023

apduvuri commented Dec 28, 2023

hsinghjkaur commented Jan 9, 2024

eulerto commented Jan 11, 2024

hsinghjkaur commented Jan 25, 2024

eulerto commented Apr 25, 2024

		@@ -1257,8 +1272,9 @@ tuple_to_stringinfo(LogicalDecodingContext *ctx, TupleDesc tupdesc, HeapTuple tu
		* Data types are printed with quotes unless they are number, true,

Add option to output numeric data types as string. #255

Add option to output numeric data types as string. #255

Conversation

shubhamdhama commented Nov 4, 2022 • edited

shubhamdhama commented Nov 4, 2022

Before this change

After

eulerto commented Nov 5, 2022

shubhamdhama commented Nov 7, 2022 • edited

shubhamdhama commented Nov 11, 2022

nakatlam Nov 15, 2022 • edited

Choose a reason for hiding this comment

shubhamdhama Nov 15, 2022

Choose a reason for hiding this comment

nakatlam Nov 15, 2022

Choose a reason for hiding this comment

shubhamdhama Nov 15, 2022

Choose a reason for hiding this comment

eulerto Nov 15, 2022

Choose a reason for hiding this comment

shubhamdhama Nov 17, 2022

Choose a reason for hiding this comment

eulerto commented Nov 22, 2022

shubhamdhama commented Nov 22, 2022

thecatontheflat commented May 20, 2023

eulerto commented May 21, 2023

hanefi commented Nov 30, 2023

apduvuri commented Dec 28, 2023

hsinghjkaur commented Jan 9, 2024

eulerto commented Jan 11, 2024

hsinghjkaur commented Jan 25, 2024

eulerto commented Apr 25, 2024

shubhamdhama commented Nov 4, 2022 •

edited

shubhamdhama commented Nov 7, 2022 •

edited

nakatlam Nov 15, 2022 •

edited