The following bug has been logged on the website:
Bug reference: 17914
Logged by: Stan S
Email address: stashukny@gmail.com
PostgreSQL version: 14.3
Operating system: AWS RDS - (Red Hat 7.3.1-12), 64-bit
Description:
Seeing the same issue as reported here:
https://www.postgresql.org/message-id/CAMnUB3oYugXCBLSkih%2BqNsWQPciEwos6g_AMbnz_peNoxfHwyw%40mail.gmail.com
In our case, logical_decoding_work_mem is set to 64 MB, yet walsenders can
each take up 30 GB of memory, crashing our database.
Able to reproduce it locally in docker, using Postgres 14.7 (latest minor
version of 14) following the same instructions as mentioned in
https://www.postgresql.org/message-id/CAMnUB3oYugXCBLSkih%2BqNsWQPciEwos6g_AMbnz_peNoxfHwyw%40mail.gmail.com
which is basically running parallelized update statements on one table on
the publisher, updating it's json column like this:
UPDATE tab1 SET data = data || '{"test_0": "1", "test_1": "1", "test_2":
"1",
"test_3": "1", "test_4": "1", "test_5": "1", "test_6": "1", "test_7": "1",
"test_8":
"1", "test_9": "1", "test_a": "1", "test_b": "1", "test_c": "1", "test_d":
"1",
"test_e": "1", "test_f": "1"}' #- '{test_0}'
The end result is each walsender's memory usage jumping up many times the
logical_decoding_work_mem limit.