Re: pg_upgrade fails with non-standard ACL

From: Bruce Momjian <bruce(at)momjian(dot)us>
To: Anastasia Lubennikova <a(dot)lubennikova(at)postgrespro(dot)ru>
Cc: Stephen Frost <sfrost(at)snowman(dot)net>, pgsql-hackers(at)lists(dot)postgresql(dot)org
Subject: Re: pg_upgrade fails with non-standard ACL
Date: 2019-08-21 19:47:38
Views: Raw Message | Whole Thread | Download mbox | Resend email
Lists: pgsql-hackers

On Tue, Aug 20, 2019 at 04:38:18PM +0300, Anastasia Lubennikova wrote:
> > Solving this in pg_upgrade does seem like it's probably the better
> > approach rather than trying to do it in pg_dump. Unfortunately, that
> > likely means that all we can do is have pg_upgrade point out to the user
> > when something will fail, with recommendations on how to address it, but
> > that's also something users are likely used to and willing to accept,
> > and puts the onus on them to consider their ACL decisions when we're
> > making catalog changes, and it keeps these issues out of pg_dump.
> I wrote a prototype to check API and ACL compatibility (see attachment).
> In the current implementation it fetches the list of system procedures for
> both old and new clusters
> and reports deleted functions or ACL changes during pg_upgrade check:
> Performing Consistency Checks
> -----------------------------
> ...
> Checking for system functions API compatibility
> dbname postgres : check procsig is equal pg_stop_backup(), procacl not equal
> {anastasia=X/anastasia,backup=X/anastasia} vs {anastasia=X/anastasia}
> dbname postgres : procedure pg_stop_backup(exclusive boolean, OUT lsn
> pg_lsn, OUT labelfile text, OUT spcmapfile text) doesn't exist in
> new_cluster
> dbname postgres : procedure pg_switch_xlog() doesn't exist in new_cluster
> dbname postgres : procedure pg_xlog_replay_pause() doesn't exist in
> new_cluster
> dbname postgres : procedure pg_xlog_replay_resume() doesn't exist in
> new_cluster
> ...
> I think it's a good first step in the right direction.
> Now I have questions about implementation details:
> 1) How exactly should we report this incompatibility to a user?
> I think it's fine to leave the warnings and also write some hint for the
> user by analogy with other checks.
> "Reset ACL on the problem functions to default in the old cluster to
> continue"

Yes, I think it is good to at least throw an error during --check so
they don't have to find out during a live upgrade. Odds are it will
require manual repair.

Bruce Momjian <bruce(at)momjian(dot)us>

+ As you are, so once was I. As I am, so you will be. +
+ Ancient Roman grave inscription +

In response to


Browse pgsql-hackers by date

  From Date Subject
Next Message Sergei Kornilov 2019-08-21 20:01:25 Re: Change ereport level for QueuePartitionConstraintValidation
Previous Message Andrew Dunstan 2019-08-21 19:40:15 Re: "ago" times on buildfarm status page