, he tried to recover the data on the last inserted transaction by using the recovery_target_time.
The issue is the database is present in the system catalog but the directory was still deleted.
Here the technical information of the database
version 11
default postgresql.conf except for this options
wal_level = replica
archive_mode = on
archive_command = 'cp %p /tmp/wal_archive/%f '
log_statement = 'all'
log_min_messages = debug5
The following method was used
create cluster
create database
create 1 table
create 1 index on 1 column
insert 1 rows
backup with pg_base_backup
insert 2 rows
autocommit?
drop database
force?
Change recovery behaviour in that case to prevent all xact operation to perform until COMMIT timestamp is checked against recovery_time bound (but it seems to be difficult as state https://www.postgresql.org/message-id/flat/20141125160629.GC21475%40msg.df7cb.de which also identifies the problem and tries to give some solutions. Maybe another way, as a trivial guess (all apologises) is to buffer immediate xacts until we have the commit for each and apply the whole buffer xact once the timestamp known (and checked agains recovery_target_time value);
