> david.g.johnston@gmail.com wrote:
>
>> bryn@yugabyte.com wrote:
>>
>> «
>> You can start a session without specifying the name of the cluster role as which to authorize, its password, and the
nameof the database to which to connect, ONLY when these things are true:
>>
>> 1. The within-cluster catalog-owning role has a certain name, say "pg_system" (or "bob").
>>
>> 2. The O/S user that owns (most of) the O/S presence of the cluster and the software that accesses it has the
identicalname "pg_system" (or "bob").
>>
>> 3. The current O/S user when you make the attempt to connect is "pg_system" (or "bob").
>> »
>>
>> I want to know if my hypothesis is correct.
>
> It is not. "By default" probably, but not "only". Peer authentication means:
>
> If the local O/S user (bob) running "psql" requests to login to the database using that same* role name (bob) and the
roleexists in the cluster, accept the authentication attempt.
>
> * You can implement aliases by using an identity mapping.
>
> Nothing more, nothing less.
>
>> And, more importantly, I want to know where I can read a nicely written linear account of what *is* correct tha
definesand then uses the official terms of art.
>
> It doesn't exist, deal with it. Most people just call the "bootstrap" role "postgres" when not talking about a
specificinstalled cluster that happens to use something different. In any case, no matter what terminology is used
everyoneseems to figure out what is being referred to from context at least and largely don't make a big deal about it.
Unless you actually want to write the documentation just pick something you like and go with it.
Yes, all is clear now. Thanks again. And thanks for the dispensation to choose my term of art. I'll use "catalog-owning
role"from now on.