on one of our Mac servers an update (Remote Desktop Client) killed yesterday the postmaster process. Apparently this did not have any influence on existing connections and therefore was not detected until some time later, when no connection for a backup could be made.
I have then closed all apps with connections to the database. This brought the whole cluster down. It restarted then with some transaction log rollback messages and seems to be running fine since then.
Can I now expect that the database is in a consistent state, or must I assume the database is corrupted? I could run a dump-all without problems and there are rows created after the death the postmaster.
It depends on the signal which was sent to the postgres process.
AFAIK, only SIGKILL (unconditional kill) can make some damage to the database.