Обсуждение: I'm surprised to see the word master here

The following documentation comment has been logged on the website:

Page: https://www.postgresql.org/docs/11/logical-replication-publication.html
Description:

I was under the impression we were removing the use of that word from our
docs ?

On 2019-09-24 21:58, PG Doc comments form wrote:
> The following documentation comment has been logged on the website:
> 
> Page: https://www.postgresql.org/docs/11/logical-replication-publication.html
> Description:
> 
> I was under the impression we were removing the use of that word from our
> docs ?

$ git grep -w master | wc -l
611

You might be thinking of "slave".

-- 
Peter Eisentraut              http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

I was under the impression that both terms were being deprecated. Is that only the case when they appear in tandem?

> On Sep 24, 2019, at 6:05 PM, Peter Eisentraut <peter.eisentraut@2ndquadrant.com> wrote:
>
>> On 2019-09-24 21:58, PG Doc comments form wrote:
>> The following documentation comment has been logged on the website:
>>
>> Page: https://www.postgresql.org/docs/11/logical-replication-publication.html
>> Description:
>>
>> I was under the impression we were removing the use of that word from our
>> docs ?
>
> $ git grep -w master | wc -l
> 611
>
> You might be thinking of "slave".
>
> --
> Peter Eisentraut              http://www.2ndQuadrant.com/
> PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
>
>

On 2019-09-25 00:28, Dave Cramer wrote:
> Ya, I was under that impression as well. 
> 
> Dave Cramer
> 
> 
> On Tue, 24 Sep 2019 at 18:18, Renee <renee.phillips@gmail.com
> <mailto:renee.phillips@gmail.com>> wrote:
> 
>     I was under the impression that both terms were being deprecated. Is
>     that only the case when they appear in tandem?

Again, you might be confusing this.  I don't recall any such initiative
nor do I see any commits to that effect.

-- 
Peter Eisentraut              http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

Greetings,

* Peter Eisentraut (peter.eisentraut@2ndquadrant.com) wrote:
> On 2019-09-25 00:28, Dave Cramer wrote:
> > Ya, I was under that impression as well. 
> >
> > Dave Cramer
> >
> >
> > On Tue, 24 Sep 2019 at 18:18, Renee <renee.phillips@gmail.com
> > <mailto:renee.phillips@gmail.com>> wrote:
> >
> >     I was under the impression that both terms were being deprecated. Is
> >     that only the case when they appear in tandem?
>
> Again, you might be confusing this.  I don't recall any such initiative
> nor do I see any commits to that effect.

Alright then, given we have multiple people asking about this- should we
be considering adopting different language, even if we hadn't previously
had such an initiative?

I know that I tend towards primary/replica when discussing physical
replication, and we do that quite a bit in the documentation (consider
https://www.postgresql.org/docs/11/warm-standby.html where we seem to be
pretty confused about if we want to talk about the system as a 'primary'
or as a 'master'- but *clearly* primary is winning the war there).

Even if we aren't avoiding the term for its negative connotations
explicitly, having some consistency here strikes me as worthwhile.

Thanks,

Stephen

On 2019-10-02 10:21, Magnus Hagander wrote:
> Exactly. Both might be accurate, but one comes with a lot less baggage.
> 
>     I support a search and replace. 
> 
> I think it'll take a bit more than just a simple "sed script to
> replace", if that's what you mean. But probably not all that much -- but
> there can certainly be cases where nearby langaugae also has to be
> changed to make it work properly. But I have a hard time seeing it as
> being a *huge* undertaking.

I find this proposal to be dubious and unsubstantiated.  Do we need to
get rid of "multimaster", "postmaster"?

-- 
Peter Eisentraut              http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

On 2019-10-02 12:46, Peter Eisentraut wrote:
> On 2019-10-02 10:21, Magnus Hagander wrote:
>> Exactly. Both might be accurate, but one comes with a lot less 
>> baggage.
>> 
>>     I support a search and replace. 
>> 
>> I think it'll take a bit more than just a simple "sed script to
>> replace", if that's what you mean. But probably not all that much -- 
>> but
>> there can certainly be cases where nearby langaugae also has to be
>> changed to make it work properly. But I have a hard time seeing it as
>> being a *huge* undertaking.
> 
> I find this proposal to be dubious and unsubstantiated.  Do we need to
> get rid of "multimaster", "postmaster"?
> 

IMHO, hat would seem a bad idea.  Let's not take the politicising too 
far.

I would say leave it at abolishing 'slave' (as we have already done).


thanks,

Erik Rijkers

On 10/2/19 7:39 AM, Chris Travers wrote:
>
>
> On Wed, Oct 2, 2019 at 12:57 PM Erikjan Rijkers <er@xs4all.nl
> <mailto:er@xs4all.nl>> wrote:
>
>     On 2019-10-02 12:46, Peter Eisentraut wrote:
>     > On 2019-10-02 10:21, Magnus Hagander wrote:
>     >> Exactly. Both might be accurate, but one comes with a lot less
>     >> baggage.
>     >>
>     >>     I support a search and replace. 
>     >>
>     >> I think it'll take a bit more than just a simple "sed script to
>     >> replace", if that's what you mean. But probably not all that much --
>     >> but
>     >> there can certainly be cases where nearby langaugae also has to be
>     >> changed to make it work properly. But I have a hard time seeing it as
>     >> being a *huge* undertaking.
>     >
>     > I find this proposal to be dubious and unsubstantiated.  Do we need to
>     > get rid of "multimaster", "postmaster"?
>     >
>
>     IMHO, hat would seem a bad idea.  Let's not take the politicising too
>     far.
>
>     I would say leave it at abolishing 'slave' (as we have already done).
>
>
> But that raises an important point, which is that if we remove master
> entirely from the replication lexicon, then I don't see how multi-master
> makes sense.  If consistency is a goal, postmaster still works but there
> is no alternative to multi-master in common usage.

At various events and tradeshows that include representation from other
database systems, the terminology that I hear is "active-active" -- this
is not one-off, but from a lot of people. This is also a common term for
the major proprietary systems as well. I hear it much more commonly than
"multi-master" even.

> Can I make a suggestion here to help ease that problem:
>
> We standardize on "primary" and "replica" but on the first usage of
> "primary" we have a parenthetical note that "primary" is sometimes
> called "master" so that terms like multi-master continue to be
> intuitively intelligible.

I'd +1 s/master/primary/ -- I don't know if it needs parenthetical on
the first usage in places (maybe in sections on replication/clustering,
but not everywhere).

Let's at least consider using "active-active" instead of "multi-master"
given there is already usage of that term in the industry. It would be
good to see what other systems do; matching terminology could have its
advantages. It's not politicizing if we're making the terminology more
inline with the industry.

I don't think postmaster needs to change; this is a title in many
countries[1] and I presume would also require a nontrivial effort and
potentially affect systems.

Jonathan

[1] https://en.wikipedia.org/wiki/Postmaster

Greetings,

* Magnus Hagander (magnus@hagander.net) wrote:
> On Wed, Oct 2, 2019 at 3:10 PM Jonathan S. Katz <jkatz@postgresql.org>
> wrote:
> > On 10/2/19 7:39 AM, Chris Travers wrote:
> > > On Wed, Oct 2, 2019 at 12:57 PM Erikjan Rijkers <er@xs4all.nl
> > > <mailto:er@xs4all.nl>> wrote:
> > >
> > >     On 2019-10-02 12:46, Peter Eisentraut wrote:
> > >     > On 2019-10-02 10:21, Magnus Hagander wrote:
> > >     >> Exactly. Both might be accurate, but one comes with a lot less
> > >     >> baggage.
> > >     >>
> > >     >>     I support a search and replace.
> > >     >>
> > >     >> I think it'll take a bit more than just a simple "sed script to
> > >     >> replace", if that's what you mean. But probably not all that much
> > --
> > >     >> but
> > >     >> there can certainly be cases where nearby langaugae also has to be
> > >     >> changed to make it work properly. But I have a hard time seeing
> > it as
> > >     >> being a *huge* undertaking.
> > >     >
> > >     > I find this proposal to be dubious and unsubstantiated.  Do we
> > need to
> > >     > get rid of "multimaster", "postmaster"?
> > >     >
> > >
> > >     IMHO, hat would seem a bad idea.  Let's not take the politicising too
> > >     far.
> > >
> > >     I would say leave it at abolishing 'slave' (as we have already done).
> > >
> > >
> > > But that raises an important point, which is that if we remove master
> > > entirely from the replication lexicon, then I don't see how multi-master
> > > makes sense.  If consistency is a goal, postmaster still works but there
> > > is no alternative to multi-master in common usage.
> >
> > At various events and tradeshows that include representation from other
> > database systems, the terminology that I hear is "active-active" -- this
> > is not one-off, but from a lot of people. This is also a common term for
> > the major proprietary systems as well. I hear it much more commonly than
> > "multi-master" even.
>
> That has the tiny problem of not being correct though.
>
> A classic primary/standby cluster is *also* active/active. It used to be
> very common to have active/passive clusters -- these were the typical
> shared-disk-mounted-on-one-node-at-a-time style clusters. This indicates
> that the standby node isn't available *at all* until after a
> fail/switchover. So pretty much anything based on our streaming replication
> today is active/active..

I don't agree with this claim.  While we could argue about if a hot
standby is considered "active" or not, the vast majority of the world
considers "active/active" to actually be where you can use the two
systems interchangably, including being able to write to both.  As such,
I disagree with this claim- while perhaps you could make an argument
that it's "technically" correct, it's not how the terms are used in
practice and saying active/active instead would be well understood by
the community and industry at large.

Thanks,

Stephen

On Wed, Oct  2, 2019 at 03:04:55PM -0400, Stephen Frost wrote:
> I don't agree with this claim.  While we could argue about if a hot
> standby is considered "active" or not, the vast majority of the world
> considers "active/active" to actually be where you can use the two
> systems interchangably, including being able to write to both.  As such,
> I disagree with this claim- while perhaps you could make an argument
> that it's "technically" correct, it's not how the terms are used in
> practice and saying active/active instead would be well understood by
> the community and industry at large.

With master/standby-replica-slave, it is clear what multi-master is, and
what master/replica is.  If you start using active-active, is it
active/replica?  The full choices are:

    master
    primary
    active

and

    standby
    replica
    slave

Whatever terms we use, it would be nice to use the same term for the
multi-master as for master/replica.  Using active-active and
primary/replica just seems odd.  Multi-primary?  Seems odd since primary
suggests one, though multiple master seems odd too, i.e., more than one
master.  Multi-active seems the most logical, or active-active, but then
active-replica seems odd because it suggests the repica is not active,
i.e. does nothing.  Is no clear logical terminology possible?

-- 
  Bruce Momjian  <bruce@momjian.us>        http://momjian.us
  EnterpriseDB                             http://enterprisedb.com

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

On Sat, 5 Oct 2019 at 10:57, Bruce Momjian <bruce@momjian.us> wrote:
> With master/standby-replica-slave, it is clear what multi-master is, and
> what master/replica is.  If you start using active-active, is it
> active/replica?   The full choices are: ...

There are more choices. Coming from a different corner of computing,
we have changed these computing resource names to other
anthropomorphic titles found around office environments: "manager" and
either "worker" or "agent". With these names, some derived terms are
"multi-manager" and "standby-replica-worker".

On Mon, Oct  7, 2019 at 10:49:54AM +1300, Mike Taves wrote:
> On Sat, 5 Oct 2019 at 10:57, Bruce Momjian <bruce@momjian.us> wrote:
> > With master/standby-replica-slave, it is clear what multi-master is, and
> > what master/replica is.  If you start using active-active, is it
> > active/replica?   The full choices are: ...
> 
> There are more choices. Coming from a different corner of computing,
> we have changed these computing resource names to other
> anthropomorphic titles found around office environments: "manager" and
> either "worker" or "agent". With these names, some derived terms are
> "multi-manager" and "standby-replica-worker".

I think the problem is that "worker" doesn't have the idea that it is a
copy of the primary, which replica and standby kind of do.  On the other
hand, worker and slave seem almost identical, and you are right they
don't have the concept of being a copy either.  :-(  I guess I was
hoping to move to a term that had _copy_ built into the term.

-- 
  Bruce Momjian  <bruce@momjian.us>        http://momjian.us
  EnterpriseDB                             http://enterprisedb.com

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

Greetings,

* Bruce Momjian (bruce@momjian.us) wrote:
> On Wed, Oct  2, 2019 at 03:04:55PM -0400, Stephen Frost wrote:
> > I don't agree with this claim.  While we could argue about if a hot
> > standby is considered "active" or not, the vast majority of the world
> > considers "active/active" to actually be where you can use the two
> > systems interchangably, including being able to write to both.  As such,
> > I disagree with this claim- while perhaps you could make an argument
> > that it's "technically" correct, it's not how the terms are used in
> > practice and saying active/active instead would be well understood by
> > the community and industry at large.
>
> With master/standby-replica-slave, it is clear what multi-master is, and
> what master/replica is.  If you start using active-active, is it
> active/replica?  The full choices are:

I'm a bit confused, as I thought that I explained exactly my thoughts on
this in the paragraph you quoted.

No, active/active isn't "active/replica", it's what we call today
"multi-master".

I'm also at a loss as to why we're discussing what to call 'master/
master' here since we don't actually use the term 'master/master' in our
docs at all today that I can see.  There's one place where we talk about
'Multimaster' and that's here:

https://www.postgresql.org/docs/current/different-replication-solutions.html

> Whatever terms we use, it would be nice to use the same term for the
> multi-master as for master/replica.  Using active-active and
> primary/replica just seems odd.  Multi-primary?  Seems odd since primary
> suggests one, though multiple master seems odd too, i.e., more than one
> master.  Multi-active seems the most logical, or active-active, but then
> active-replica seems odd because it suggests the repica is not active,
> i.e. does nothing.  Is no clear logical terminology possible?

I'm confused here again.  Why would we want to use the same term for two
primaries that are working together as we would for a primary and a
replica...?  Those are two quite different setups, and, at least in my
experience, 'active-active' only applies to the case where you have two
primaries involved.

In any case, removing the term 'master' from:

https://www.postgresql.org/docs/current/warm-standby.html

doesn't actually run into any of these issues because we don't say
'Multimaster' or 'master/master' or anything like that on that page, we
just keep going back and forth between 'master' and 'primary' without
any real reason for doing so.  There's also some areas which could use
clarification, like:

In standby mode, the server continuously applies WAL received from the
master server.

Which isn't entirely accurate- the server continuously applies WAL
received from the primary, or the upstream standby it is connected to,
though we could possibly just include a reference down to Cacading
Replication to clarify that.

Of course, down in cascading replication we actually say:

which eventually links to a single master/primary server.

Which just re-enforces that we're pretty confused in that page regarding
the language/distinction between those two.

Coming back around to:

https://www.postgresql.org/docs/current/different-replication-solutions.html

and the "master/Multimaster" references there- most of the "master" ones
can be changed directly to "primary" (and most of the "standby" one
should really probably be "replica"...  that's a different task though,
to go through and clean up that, but also a worthwhile one, imv).  The
"Multimaster" references could go to "active-active" without any loss in
readability or understanding, imv, and we further explain each case in
the description of those anyway.

Though I have to say that I take some exception with the general
"Multimaster" or "active-active" concept being applied to offline /
periodically sync'd systems like when you have laptops or such involved,
but that's a different issue.

Thanks,

Stephen

On 2019-10-08 18:39, Stephen Frost wrote:
> I'm also at a loss as to why we're discussing what to call 'master/
> master' here since we don't actually use the term 'master/master' in our
> docs at all today that I can see.

The underlying premise appears to be that the word should be banned.  In
which case that would certainly propagate to the web site, the wiki,
other nearby resources, some of which certainly do use that term
liberally in various configurations.  So we should think that through
regardless.

-- 
Peter Eisentraut              http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

Greetings,

* Peter Eisentraut (peter.eisentraut@2ndquadrant.com) wrote:
> On 2019-10-08 18:39, Stephen Frost wrote:
> > I'm also at a loss as to why we're discussing what to call 'master/
> > master' here since we don't actually use the term 'master/master' in our
> > docs at all today that I can see.
>
> The underlying premise appears to be that the word should be banned.  In
> which case that would certainly propagate to the web site, the wiki,
> other nearby resources, some of which certainly do use that term
> liberally in various configurations.  So we should think that through
> regardless.

Saying 'banned' goes beyond what I think the goal here is- but, yes, I
believe the premise here is that we should avoid using the term.  That's
only part of it though- we should also be trying to actually be consistent
in our use of terms and we're far from that today.  Fixing the consistency
issue would just about entirely remove the term by itself, to the point
where adjusting the remaining one page in the docs is a modest change
that also puts us in a better light.

Regarding taking that avoidance to other places- I believe it's pretty
clear that we have not implemented any kind of ban even for other words
which we now avoid using on postgresql.org and in the docs- a search for
'slave' on the wiki turns up a pretty impressive number of hits, so I
don't know why we're thinking of expanding this to include that as we
haven't before.  Perhaps we should consider that, but that's an entirely
different discussion and not relevant to this discussion.  Additionally,
the wiki is pretty clearly a user resource and nothing on it should be
considered an endorsement from the project- which is part of the reason
we moved things like policies from the wiki over to the main website.

Reviewing the website, beyond the docs, turns up almost zero references
to 'master' or 'multi-master' or even 'remastering'- and the ones that
do exist appear to all be from older press releases which I don't think
anyone would reasonably expect us to change at this point (though I do
wonder if perhaps we should start to clean up "presskit91" and
friends...).  We do have "Mastering PostgreSQL", as a reference to the
book of the same name, but I don't believe anyone has an issue with
that.  Saying that these terms are in use liberally on the main website
just doesn't stand up to any scrutiny or even cursory review, from what
I can see, and if we aren't using them in the docs, it seems extremely
unlikely that they'd end up getting added to the main website.

If there are other resources which you have specific concerns regarding,
I'd suggest you go look and see to what extent they're actually using
the terminology in question and then we can discuss them and how much
they are viewed as representing the project.

Thanks,

Stephen