Functions Immutable but not parallel safe?

Started by David Rowleyover 9 years ago9 messageshackers
Jump to latest
#1David Rowley
dgrowleyml@gmail.com

There's 11 functions which are marked immutable, but are marked as
parallel unsafe.

postgres=# select proname from pg_proc where provolatile = 'i' and
proparallel = 'u';
proname
-----------------------------
_pg_expandarray
_pg_keysequal
_pg_truetypid
_pg_truetypmod
_pg_char_max_length
_pg_char_octet_length
_pg_numeric_precision
_pg_numeric_precision_radix
_pg_numeric_scale
_pg_datetime_precision
_pg_interval_type
(11 rows)

I'm finding hard to imagine a reason why these might be unsafe, but
failed. I do notice they're all only used in information_schema.

Could it just perhaps be that these just missed the verification
process the other functions went through to determine their parallel
safety?

--
David Rowley http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

#2Robert Haas
robertmhaas@gmail.com
In reply to: David Rowley (#1)
Re: Functions Immutable but not parallel safe?

On Thu, Nov 24, 2016 at 5:29 AM, David Rowley
<david.rowley@2ndquadrant.com> wrote:

There's 11 functions which are marked immutable, but are marked as
parallel unsafe.

postgres=# select proname from pg_proc where provolatile = 'i' and
proparallel = 'u';
proname
-----------------------------
_pg_expandarray
_pg_keysequal
_pg_truetypid
_pg_truetypmod
_pg_char_max_length
_pg_char_octet_length
_pg_numeric_precision
_pg_numeric_precision_radix
_pg_numeric_scale
_pg_datetime_precision
_pg_interval_type
(11 rows)

I'm finding hard to imagine a reason why these might be unsafe, but
failed. I do notice they're all only used in information_schema.

Could it just perhaps be that these just missed the verification
process the other functions went through to determine their parallel
safety?

Yes, I think that's it. I went through pg_proc.h, but never looked at
information_schema.sql.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

#3Peter Eisentraut
peter_e@gmx.net
In reply to: Robert Haas (#2)
Re: Functions Immutable but not parallel safe?

On 11/24/16 18:13, Robert Haas wrote:

I'm finding hard to imagine a reason why these might be unsafe, but
failed. I do notice they're all only used in information_schema.

Could it just perhaps be that these just missed the verification
process the other functions went through to determine their parallel
safety?

Yes, I think that's it. I went through pg_proc.h, but never looked at
information_schema.sql.

This hasn't been fixed yet. It's easy to to, but taking a step back,

- Is there any reason an immutable function (that is not lying about it)
should be anything but parallel safe?

- If so, could CREATE FUNCTION default it that way?

- Maybe add a check to opr_sanity to make sure the default set of
functions is configured the way we want?

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

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

#4Robert Haas
robertmhaas@gmail.com
In reply to: Peter Eisentraut (#3)
Re: Functions Immutable but not parallel safe?

On Wed, Apr 5, 2017 at 8:57 AM, Peter Eisentraut
<peter.eisentraut@2ndquadrant.com> wrote:

On 11/24/16 18:13, Robert Haas wrote:

I'm finding hard to imagine a reason why these might be unsafe, but
failed. I do notice they're all only used in information_schema.

Could it just perhaps be that these just missed the verification
process the other functions went through to determine their parallel
safety?

Yes, I think that's it. I went through pg_proc.h, but never looked at
information_schema.sql.

This hasn't been fixed yet. It's easy to to, but taking a step back,

- Is there any reason an immutable function (that is not lying about it)
should be anything but parallel safe?

It certainly isn't very likely. It's not outright impossible. For
example, imagine a function that does a calculation which is
deterministic given the inputs but which creates and uses temporary
tables in the course of performing the calculation. Because the
function performs writes, it's parallel-unsafe.

- If so, could CREATE FUNCTION default it that way?

This could be done but I'm not sure whether it's wise to make the
default value for one parameter depend on another parameter.

- Maybe add a check to opr_sanity to make sure the default set of
functions is configured the way we want?

That seems like a good idea.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

#5Tom Lane
tgl@sss.pgh.pa.us
In reply to: Robert Haas (#4)
Re: Functions Immutable but not parallel safe?

Robert Haas <robertmhaas@gmail.com> writes:

On Wed, Apr 5, 2017 at 8:57 AM, Peter Eisentraut
<peter.eisentraut@2ndquadrant.com> wrote:

- Maybe add a check to opr_sanity to make sure the default set of
functions is configured the way we want?

That seems like a good idea.

+1 for that. We could adopt the strategy already used in opr_sanity of
searching for functions having an unexpected combination of these
attributes. If there are any legitimate exceptions, they could be
embedded in the expected output.

I concur that changing the behavior of CREATE FUNCTION seems a bit too
cute.

regards, tom lane

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

#6Peter Eisentraut
peter_e@gmx.net
In reply to: Robert Haas (#4)
Re: Functions Immutable but not parallel safe?

On 4/5/17 09:58, Robert Haas wrote:

- Maybe add a check to opr_sanity to make sure the default set of
functions is configured the way we want?

That seems like a good idea.

patch

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

Attachments:

0001-Mark-immutable-functions-in-information-schema-as-pa.patchinvalid/octet-stream; name=0001-Mark-immutable-functions-in-information-schema-as-pa.patchDownload+24-3
#7Robert Haas
robertmhaas@gmail.com
In reply to: Peter Eisentraut (#6)
Re: Functions Immutable but not parallel safe?

On Wed, Apr 5, 2017 at 12:18 PM, Peter Eisentraut
<peter.eisentraut@2ndquadrant.com> wrote:

On 4/5/17 09:58, Robert Haas wrote:

- Maybe add a check to opr_sanity to make sure the default set of
functions is configured the way we want?

That seems like a good idea.

patch

LGTM

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

#8Tom Lane
tgl@sss.pgh.pa.us
In reply to: Peter Eisentraut (#6)
Re: Functions Immutable but not parallel safe?

Peter Eisentraut <peter.eisentraut@2ndquadrant.com> writes:

On 4/5/17 09:58, Robert Haas wrote:

- Maybe add a check to opr_sanity to make sure the default set of
functions is configured the way we want?

That seems like a good idea.

patch

Looks sane to me, although I wonder if opr_sanity ought to be looking
for any other combinations.

regards, tom lane

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

#9Peter Eisentraut
peter_e@gmx.net
In reply to: Robert Haas (#7)
Re: Functions Immutable but not parallel safe?

On 4/5/17 12:26, Robert Haas wrote:

On Wed, Apr 5, 2017 at 12:18 PM, Peter Eisentraut
<peter.eisentraut@2ndquadrant.com> wrote:

On 4/5/17 09:58, Robert Haas wrote:

- Maybe add a check to opr_sanity to make sure the default set of
functions is configured the way we want?

That seems like a good idea.

patch

LGTM

committed

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

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers