requested WAL segment has already been removed

[alesz]

Submission type

• Bug report

Environment

• RHEL 7.5
• PGSQL 10.3

Stolon version

0.10.0

Expected behaviour you didn't see

Slave node synchronizes successfully

Unexpected behaviour you saw

pg_basebackup fails due to WAL segments already missing by the time it is done.

Steps to reproduce the problem

Stolon executes pg_basebackup as:

/usr/pgsql-10/bin/pg_basebackup -R -Xf -D /var/lib/pgsql/10/data/postgres -d application_name=stolon_33a935d2 options=-c\ synchronous_commit=off sslmode=disable user=repluser host=x.x.x.x port=5431

According to docs, WAL segments will rotate even if replication haven't caught up when using -Xf flag.

-Xs flag should fix the issue.

[sgotti]

@alesz I think we should switch to wal streaming in pg_basebackup. This will require doubling the number of wal senders to be on the safe side.

If I got the doc correctly in postgres 10 a temporary replication slot will be created if --slot is not provided. We could directly provide --slot to make it work also on older version but a side effect will be that sometimes the initial pg_basebackup will fail if the slot is not yet created on the source instance by the keeper, but a second try will succeed (so user will see a pg_basebackup error but this is better that not using a replication slot).

There is another similar issues #372

[sgotti]

Created PR #477