Default of max_stack_depth and getrlimit

"Heikki Linnakangas" <heikki@enterprisedb.com> writes:

In 8.2, we started to use getrlimit(RLIMIT_STACK, ...) to set
max_stack_depth automatically, to a max of 2MB:

http://archives.postgresql.org/pgsql-hackers/2006-10/msg00389.php

However, it's not behaving as I expected when the stack limit is set to
"unlimited".

Is there really any such thing as "unlimited" stack depth? I think that
treating that as "unknown" is a good conservative thing to do, because
it is surely a lie.

regards, tom lane

Tom Lane wrote:

"Heikki Linnakangas" <heikki@enterprisedb.com> writes:

In 8.2, we started to use getrlimit(RLIMIT_STACK, ...) to set
max_stack_depth automatically, to a max of 2MB:

http://archives.postgresql.org/pgsql-hackers/2006-10/msg00389.php

However, it's not behaving as I expected when the stack limit is set to
"unlimited".

Is there really any such thing as "unlimited" stack depth?

No, but I would think it's safe to assume that "unlimited" is greater
than 2MB.

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com

Le Monday 21 July 2008, Heikki Linnakangas a écrit :

In 8.2, we started to use getrlimit(RLIMIT_STACK, ...) to set
max_stack_depth automatically, to a max of 2MB:

http://archives.postgresql.org/pgsql-hackers/2006-10/msg00389.php

However, it's not behaving as I expected when the stack limit is set to
"unlimited". I would expect max_stack_depth to be set to the max of 2MB,
but instead it gets set to 100kB.

I don't normally run without a limit, but it looks like the regression
tests run like that with "make check", at least on my platform. I bumped
into this while running a custom regression test with very deep nesting.

I think we should differentiate between "infinite" and "unknown" in the
return value of get_stack_depth_limit(), and use max_stack_depth of 2MB
in case of infinite, and fall back to the 100kB only in the unknown case.

Why 2MB ? I believed that 3.5MB is the effective good maximum , is that too
much ?

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com

--
Cédric Villemain
Administrateur de Base de Données
Cel: +33 (0)6 74 15 56 53
http://dalibo.com - http://dalibo.org

C�dric Villemain wrote:

Le Monday 21 July 2008, Heikki Linnakangas a �crit :

I think we should differentiate between "infinite" and "unknown" in the
return value of get_stack_depth_limit(), and use max_stack_depth of 2MB
in case of infinite, and fall back to the 100kB only in the unknown case.

Why 2MB ? I believed that 3.5MB is the effective good maximum , is that too
much ?

2MB is the value we set max_stack_depth to, unless getrlimit() says that
the actual stack limit is lower than that.

I believe the 2MB figure is just an arbitrary value, thought to be quite
safe, but also high enough that most people won't need to raise it.
Before we started to use getrlimit(), we used to just default
max_stack_depth=2MB.

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com

Le Monday 21 July 2008, Heikki Linnakangas a écrit :

Cédric Villemain wrote:

Le Monday 21 July 2008, Heikki Linnakangas a écrit :

I think we should differentiate between "infinite" and "unknown" in the
return value of get_stack_depth_limit(), and use max_stack_depth of 2MB
in case of infinite, and fall back to the 100kB only in the unknown
case.

Why 2MB ? I believed that 3.5MB is the effective good maximum , is that
too much ?

2MB is the value we set max_stack_depth to, unless getrlimit() says that
the actual stack limit is lower than that.

I believe the 2MB figure is just an arbitrary value, thought to be quite
safe, but also high enough that most people won't need to raise it.
Before we started to use getrlimit(), we used to just default
max_stack_depth=2MB.

Ok, thank you for the explanation.

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com

--
Cédric Villemain
Administrateur de Base de Données
Cel: +33 (0)6 74 15 56 53
http://dalibo.com - http://dalibo.org