dblink versus long connection strings

Started by Tom Laneabout 15 years ago11 messages
#1Tom Lane
tgl@sss.pgh.pa.us

This bug report:
http://archives.postgresql.org/pgsql-bugs/2010-11/msg00139.php
shows that this patch was ill-considered:
http://archives.postgresql.org/pgsql-committers/2010-06/msg00013.php
and this later attempt didn't fix it, because it still misbehaves in
HEAD:
http://archives.postgresql.org/pgsql-committers/2010-06/msg00070.php
not to mention that that second patch didn't even touch pre-8.4
branches.

I'm inclined to think that we should just change all the
truncate_identifier calls to warn=false, and forget about providing
identifier-truncated warnings here. It's too difficult to tell whether
a string is really meant as an identifier.

regards, tom lane

#2Itagaki Takahiro
itagaki.takahiro@gmail.com
In reply to: Tom Lane (#1)
Re: dblink versus long connection strings

On Tue, Nov 23, 2010 at 01:27, Tom Lane <tgl@sss.pgh.pa.us> wrote:

This bug report:
http://archives.postgresql.org/pgsql-bugs/2010-11/msg00139.php
shows that this patch was ill-considered:
http://archives.postgresql.org/pgsql-committers/2010-06/msg00013.php
and this later attempt didn't fix it, because it still misbehaves in
HEAD:
http://archives.postgresql.org/pgsql-committers/2010-06/msg00070.php
not to mention that that second patch didn't even touch pre-8.4
branches.

I'm inclined to think that we should just change all the
truncate_identifier calls to warn=false, and forget about providing
identifier-truncated warnings here.  It's too difficult to tell whether
a string is really meant as an identifier.

It is not a truncated identifier, but I think the truncation is still
worth warning because we cannot distinguish two connections that
differ only >63 bytes.

Do we need another logic to name non-named connections?
For example, md5 hash of the connection string.

--
Itagaki Takahiro

#3Tom Lane
tgl@sss.pgh.pa.us
In reply to: Itagaki Takahiro (#2)
Re: dblink versus long connection strings

Itagaki Takahiro <itagaki.takahiro@gmail.com> writes:

On Tue, Nov 23, 2010 at 01:27, Tom Lane <tgl@sss.pgh.pa.us> wrote:

I'm inclined to think that we should just change all the
truncate_identifier calls to warn=false, and forget about providing
identifier-truncated warnings here.  It's too difficult to tell whether
a string is really meant as an identifier.

It is not a truncated identifier, but I think the truncation is still
worth warning because we cannot distinguish two connections that
differ only >63 bytes.

The problem is to not give a warning when the string isn't meant as a
connection name at all, but as a libpq conninfo string (which can
perfectly reasonably run to more than 63 characters). Most if not all
of the dblink functions will accept either.

Perhaps a reasonable compromise is to issue the truncation warnings when
an overlength name is being *entered* into the connection table, but not
for simple lookups.

regards, tom lane

#4Andrew Dunstan
andrew@dunslane.net
In reply to: Tom Lane (#3)
Re: dblink versus long connection strings

On 11/22/2010 11:51 AM, Tom Lane wrote:

Itagaki Takahiro<itagaki.takahiro@gmail.com> writes:

On Tue, Nov 23, 2010 at 01:27, Tom Lane<tgl@sss.pgh.pa.us> wrote:

I'm inclined to think that we should just change all the
truncate_identifier calls to warn=false, and forget about providing
identifier-truncated warnings here. � It's too difficult to tell whether
a string is really meant as an identifier.

It is not a truncated identifier, but I think the truncation is still
worth warning because we cannot distinguish two connections that
differ only>63 bytes.

The problem is to not give a warning when the string isn't meant as a
connection name at all, but as a libpq conninfo string (which can
perfectly reasonably run to more than 63 characters). Most if not all
of the dblink functions will accept either.

Perhaps a reasonable compromise is to issue the truncation warnings when
an overlength name is being *entered* into the connection table, but not
for simple lookups.

Can't we distinguish a name from a conninfo string by the presence of an
= sign?

cheers

andrew

#5Tom Lane
tgl@sss.pgh.pa.us
In reply to: Andrew Dunstan (#4)
Re: dblink versus long connection strings

Andrew Dunstan <andrew@dunslane.net> writes:

On 11/22/2010 11:51 AM, Tom Lane wrote:

Perhaps a reasonable compromise is to issue the truncation warnings when
an overlength name is being *entered* into the connection table, but not
for simple lookups.

Can't we distinguish a name from a conninfo string by the presence of an
= sign?

No, because = isn't disallowed in names ...

regards, tom lane

#6Andrew Dunstan
andrew@dunslane.net
In reply to: Tom Lane (#5)
Re: dblink versus long connection strings

On 11/22/2010 12:08 PM, Tom Lane wrote:

Andrew Dunstan<andrew@dunslane.net> writes:

On 11/22/2010 11:51 AM, Tom Lane wrote:

Perhaps a reasonable compromise is to issue the truncation warnings when
an overlength name is being *entered* into the connection table, but not
for simple lookups.

Can't we distinguish a name from a conninfo string by the presence of an
= sign?

No, because = isn't disallowed in names ...

Ok, true, but it still might not be a bad heuristic to use for issuing a
warning on lookup.

cheers

andrew

#7Tom Lane
tgl@sss.pgh.pa.us
In reply to: Andrew Dunstan (#6)
Re: dblink versus long connection strings

Andrew Dunstan <andrew@dunslane.net> writes:

On 11/22/2010 12:08 PM, Tom Lane wrote:

No, because = isn't disallowed in names ...

Ok, true, but it still might not be a bad heuristic to use for issuing a
warning on lookup.

The fine manual says that using "=" in a connection name might be unwise
because of the risk of confusion. It doesn't say that you should expect
to get a NOTICE every single time you use the name. People have
complained of postmaster log bloat for lots less reason than this.

In any case I don't see an argument why warning on connection creation
isn't sufficient.

regards, tom lane

#8Andrew Dunstan
andrew@dunslane.net
In reply to: Tom Lane (#7)
Re: dblink versus long connection strings

On 11/22/2010 12:21 PM, Tom Lane wrote:

Andrew Dunstan<andrew@dunslane.net> writes:

On 11/22/2010 12:08 PM, Tom Lane wrote:

No, because = isn't disallowed in names ...

Ok, true, but it still might not be a bad heuristic to use for issuing a
warning on lookup.

The fine manual says that using "=" in a connection name might be unwise
because of the risk of confusion. It doesn't say that you should expect
to get a NOTICE every single time you use the name. People have
complained of postmaster log bloat for lots less reason than this.

In any case I don't see an argument why warning on connection creation
isn't sufficient.

OK.

cheers

andrew

#9Itagaki Takahiro
itagaki.takahiro@gmail.com
In reply to: Tom Lane (#7)
Re: dblink versus long connection strings

On Tue, Nov 23, 2010 at 02:21, Tom Lane <tgl@sss.pgh.pa.us> wrote:

In any case I don't see an argument why warning on connection creation
isn't sufficient.

I'll check all versions of dblink. truncate_identifier() will be called
with warn=false in all cases except dblink_coneect() -> createNewConnection().

--
Itagaki Takahiro

#10queej
dqj@authentrics.com
In reply to: Itagaki Takahiro (#9)
Re: dblink versus long connection strings

I have views that use the dblink(connStr text, sql text) call. They cannot
use a two-step process. So postgres 9.0 has broken all of those views. Is
there a straightforward solution to this?
--
View this message in context: http://postgresql.1045698.n5.nabble.com/dblink-versus-long-connection-strings-tp3275575p3284620.html
Sent from the PostgreSQL - hackers mailing list archive at Nabble.com.

#11Itagaki Takahiro
itagaki.takahiro@gmail.com
In reply to: queej (#10)
Re: dblink versus long connection strings

On Tue, Nov 30, 2010 at 01:01, queej <dqj@authentrics.com> wrote:

I have views that use the dblink(connStr text, sql text) call.  They cannot
use a two-step process.  So postgres 9.0 has broken all of those views.  Is
there a straightforward solution to this?

Could you explain your views? I cannot get any warnings from
dblink(connStr text, sql text) with long connStr.

Also, I wonder two things:
* dblink(connStr text, sql text) never raises warning logs even without
the recent fix, because they don't register connection names.
* Connection names could be truncated, but connection strings are never
truncated. I'm not sure why connection strings are logged in your log.

--
Itagaki Takahiro