Giters

2ndQuadrant / pglogical

Logical Replication extension for PostgreSQL 15, 14, 13, 12, 11, 10, 9.6, 9.5, 9.4 (Postgres), providing much faster replication than Slony, Bucardo or Londiste, as well as cross-version upgrades.

Home Page:http://2ndquadrant.com/en/resources/pglogical/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

pg_logical_slot_peek_changes with slots created with pglogical_output missed DML records

Jamespsql opened this issue · comments

commented

we installed from pglogical_13-2.4.2-1.rhel8.x86_64.rpm , database version is 13.4 , 13.6 two-way replication
We test pglogical replication, with pglogical.conflict_resolution=error, and from source database run
Insert into pgbench_branches values(109,0,’test’);
Commit;

At target database, since already exist bid=109, so it failed there and from the error log:
2023-01-10 13:06:33.161 UTC::@:[1700321]:[30-1]::2023-01-10 11:22:38 UTCLOG: background worker "pglogical apply 16389:2210095472" (PID 1716767) exited with exit code 1
2023-01-10 13:06:38.168 UTC::[unknown]@pgddweb:[1723585]:[1-1]:pglogical apply 16389:2210095472:2023-01-10 13:06:38 UTCLOG: starting apply for subscription topgddweb
2023-01-10 13:06:38.606 UTC::[unknown]@pgddweb:[1723585]:[2-1]:pglogical apply 16389:2210095472:2023-01-10 13:06:38 UTCERROR: cannot apply conflicting row
2023-01-10 13:06:38.606 UTC::[unknown]@pgddweb:[1723585]:[3-1]:pglogical apply 16389:2210095472:2023-01-10 13:06:38 UTCCONTEXT: apply INSERT from remote relation pgbench.pgbench_branches in commit before 0/FA002200, xid 1147 committed at 2023-01-10 13:06:33.156494+00 (action #2) from node replorigin 1

Go back to source database, and check details about pg wal records,

SELECT *
FROM pg_logical_slot_peek_changes('pgl_pgddweb_pgtest1_topgddweb', NULL, 100,
'min_proto_version', '1', 'max_proto_version', '1',
'startup_params_format', '1', 'proto_format', 'json');

lsn     | xid  |

0/FA002110 | 1147 | {"action":"S", "params": {"max_proto_version":"1","min_proto_version":"1","coltypes":"f","pg_version_num":"130008","pg_version":"13.8","pg_catversion":"202007201","da
tabase_encoding":"LATIN1","encoding":"SQL_ASCII","forward_changeset_origins":"t","walsender_pid":"1685609","pglogical_version":"2.4.2","pglogical_version_num":"20402","binary.internal_bas
etypes":"f","binary.binary_basetypes":"f","binary.basetypes_major_version":"1300","binary.sizeof_int":"4","binary.sizeof_long":"8","binary.sizeof_datum":"8","binary.maxalign":"8","binary.
bigendian":"f","binary.float4_byval":"f","binary.float8_byval":"t","binary.integer_datetimes":"f","binary.binary_pg_version":"1300","no_txinfo":"f"}}
0/FA002110 | 1147 | {"action":"B", "has_catalog_changes":"f", "xid":"1147", "first_lsn":"0/FA002110", "commit_time":"2023-01-10 13:06:33.156494+00"}
0/FA002230 | 1147 | {"action":"C", "final_lsn":"0/FA002200", "end_lsn":"0/FA002230"}

We only get ‘B’ and ‘C’ but no ‘I’ fo the failed insert statement in transaction 1147.

We create another test slot use test_decoding plugin, it works, it will print all sql statements records together ‘B’, ‘C’.

select * from pg_replication_slots;
slot_name | plugin | slot_type | datoid | database | temporary | active | active_pid | xmin | catalog_xmin | restart_lsn | confirmed_flush_lsn | wal_status
| safe_wal_size
-------------------------------+------------------+-----------+--------+----------+-----------+--------+------------+------+--------------+-------------+---------------------+------------
+---------------
pgl_pgddweb_pgtest1_topgddweb | pglogical_output | logical | 25634 | pgtest1 | f | f | | | 1147 | 0/FA0020D8 | 0/FA002110 | reserved
|
test | test_decoding | logical | 25634 | pgtest1 | f | f | | | 1147 | 0/FA0027C0 | 0/FA0027F8 | reserved
|

Could you help check why pglogical_output does not ?

commented

both 13.4 and 13.6 database slot saw this issue

commented

Seeing same issue . If it is resolved then please post .