Re: ALTER TABLE .. SET SCHEMA lock strength

Simon Riggs <simon(at)2ndQuadrant(dot)com> writes:
> On Sat, 2011-01-01 at 11:06 -0500, Robert Haas wrote:
>> While reviewing the SQL/MED patch, I happened to notice that
>> ExecAlterObjectSchemaStmt calls AlterTableNamespace with a lock mode
>> argument of AccessExclusiveLock. Does anyone see a reason why
>> ShareUpdateExclusiveLock would be insufficient?

> It seemed unsafe to me to do that while an object was being accessed,
> since it effectively changes the search_path, which is dangerous.

ALTER RENAME and ALTER SET SCHEMA are both in the nature of changing the
object's identity. Consider the fairly typical use-case where you are
renaming an "old" instance out of the way and renaming another one into
the same schema/name. Do you really want that to be a low-lock
operation? I find it really hard to envision a use case where it'd be
smart to allow some concurrent operations to continue using the the old
instance while others start using the new one.

regards, tom lane