replication with clickhouse as a subscriber causes a crash of the compute node

[save-buffer]

Steps to reproduce

Follow this instruction, but don't create a role and a database in Postgres, don't modify pg_hba.conf etc, use the neon default user and the default database instead. Crash will happen after the database connection creation in the ClickHouse.

Expected result

No crash

Actual result

Crash

Environment

Staging (ep-holy-fog-w2aoqekg)

Logs, links

Discussion: https://neondb.slack.com/archives/C0756AKBMU0/p1720032475503459

[a-masterov]

To reproduce follow this instruction, but don't create a role and a database in Postgres, don't modify pg_hba.conf etc, use the neon default user and the default database instead. Crash will happen after the database connection creation in the ClickHouse.

[knizhnik]

Were you able to get core dump?

[a-masterov]

Were you able to get core dump?

Unfortunately https://github.com/neondatabase/neon/pull/8272 is not yet merged, still needs an approve from the code owners

[a-masterov]

The backtrace is attached to the slack discussion.

[kelvich]

#1  0x00007f5505fbeecf in NeonWALReadRemote (state=0x5652093cd200, buf=0x5652093cf200 "s", buf@entry=0x5652093cd200 "\023\321\002", startptr=140737453845920, startptr@entry=16777216, count=140737453845916, count@entry=8192, tli=tli@entry=1) at neon_walreader.c:373

so in https://github.com/neondatabase/neon/blob/main/pgxn/neon/neon_walreader.c#L373

[save-buffer]

With https://github.com/neondatabase/neon/pull/8360 merged, we should try again and confirm it doesn't crash with clickhouse

[a-masterov]

I confirm that starting a logical replication using ClickHouse doesn't cause crashes anymore.