Обсуждение: PgAdmin 4.2 unable to connect to AWS Postgres read-replica instances

"Bawol, Brian" <brian.bawol@freightverify.com> writes:
> My co-worker and I are both receiving the following error message when
> attempting to connect to AWS Postgres 10.4 (and 10.6) database instances
> using PgAdmin 4.2:

> "ERROR: cannot execute SELECT in a read-only transaction"

That seems pretty strange.  What's the actual underlying query that's
failing?  (The server log should show that, as a appendage to the
occurrence of this error, even if PgAdmin won't tell you.)

            regards, tom lane

On 2019-Feb-07, Bawol, Brian wrote:

> It looks like this is the query that occurs at the same time:
> 
> STATEMENT:  SET DateStyle=ISO; SET client_min_messages=notice;UPDATE
> pg_settings SET setting = 'escape'    WHERE name = 'bytea_output';SET
> client_encoding='UNICODE';
> 
> Does this help?

Ah, so pgadmin 4.2 is trying to set bytea_output, which is disallowed;
the reason it says "cannot execute SELECT" is that the UPDATE for the
view is turned into a "SELECT set_config()" by a view rule.

That makes sense.  This looks like a pgadmin bug ...

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

On 2019-Feb-07, Bawol, Brian wrote:

> It looks like this is the query that occurs at the same time:
> 
> STATEMENT:  SET DateStyle=ISO; SET client_min_messages=notice;UPDATE
> pg_settings SET setting = 'escape'    WHERE name = 'bytea_output';SET
> client_encoding='UNICODE';
> 
> Does this help?

Ah, so pgadmin 4.2 is trying to set bytea_output, which is disallowed;
the reason it says "cannot execute SELECT" is that the UPDATE for the
view is turned into a "SELECT set_config()" by a view rule.

That makes sense.  This looks like a pgadmin bug ...

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> On 2019-Feb-07, Bawol, Brian wrote:
>> It looks like this is the query that occurs at the same time:
>> STATEMENT:  SET DateStyle=ISO; SET client_min_messages=notice;UPDATE
>> pg_settings SET setting = 'escape'    WHERE name = 'bytea_output';SET
>> client_encoding='UNICODE';

> Ah, so pgadmin 4.2 is trying to set bytea_output, which is disallowed;
> the reason it says "cannot execute SELECT" is that the UPDATE for the
> view is turned into a "SELECT set_config()" by a view rule.

Right, and the transformed query still specifies that pg_settings
should be checked for UPDATE privilege, which is what is making
ExecCheckXactReadOnly spit up on it.

In principle you could argue that ExecCheckXactReadOnly could be smarter
about this type of situation.  A probably more significant issue is that a
permissions check is a really strange, and not at all bulletproof-seeming,
way of deciding whether a query is read-only.  But I don't have any
burning desire to redesign that right now.

> This looks like a pgadmin bug ...

Looking at it, I'm betting somebody thought this was a cute way to
set bytea_output if it exists, without getting a failure on servers
too old to have it.  We added that in 9.0, so maybe it's not quite
old enough to make it a moot point; but failing on RO servers is
not nice either.  Anyway, yeah, it's pgadmin's problem to fix.

            regards, tom lane

Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> On 2019-Feb-07, Bawol, Brian wrote:
>> It looks like this is the query that occurs at the same time:
>> STATEMENT:  SET DateStyle=ISO; SET client_min_messages=notice;UPDATE
>> pg_settings SET setting = 'escape'    WHERE name = 'bytea_output';SET
>> client_encoding='UNICODE';

> Ah, so pgadmin 4.2 is trying to set bytea_output, which is disallowed;
> the reason it says "cannot execute SELECT" is that the UPDATE for the
> view is turned into a "SELECT set_config()" by a view rule.

Right, and the transformed query still specifies that pg_settings
should be checked for UPDATE privilege, which is what is making
ExecCheckXactReadOnly spit up on it.

In principle you could argue that ExecCheckXactReadOnly could be smarter
about this type of situation.  A probably more significant issue is that a
permissions check is a really strange, and not at all bulletproof-seeming,
way of deciding whether a query is read-only.  But I don't have any
burning desire to redesign that right now.

> This looks like a pgadmin bug ...

Looking at it, I'm betting somebody thought this was a cute way to
set bytea_output if it exists, without getting a failure on servers
too old to have it.  We added that in 9.0, so maybe it's not quite
old enough to make it a moot point; but failing on RO servers is
not nice either.  Anyway, yeah, it's pgadmin's problem to fix.

            regards, tom lane

I wrote:
>> This looks like a pgadmin bug ...

> Looking at it, I'm betting somebody thought this was a cute way to
> set bytea_output if it exists, without getting a failure on servers
> too old to have it.  We added that in 9.0, so maybe it's not quite
> old enough to make it a moot point; but failing on RO servers is
> not nice either.  Anyway, yeah, it's pgadmin's problem to fix.

Better idea:

select set_config('bytea_output','escape',false) from pg_settings where name = 'bytea_output';

This is depressingly lacking in safe schema-qualification, btw.

            regards, tom lane

I wrote:
>> This looks like a pgadmin bug ...

> Looking at it, I'm betting somebody thought this was a cute way to
> set bytea_output if it exists, without getting a failure on servers
> too old to have it.  We added that in 9.0, so maybe it's not quite
> old enough to make it a moot point; but failing on RO servers is
> not nice either.  Anyway, yeah, it's pgadmin's problem to fix.

Better idea:

select set_config('bytea_output','escape',false) from pg_settings where name = 'bytea_output';

This is depressingly lacking in safe schema-qualification, btw.

            regards, tom lane