Re: MemoryContextSwitchTo (Re: [GENERAL] Autovacuum daemon terminated by signal 11)

Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> writes:

> Simon Riggs <simon(at)2ndQuadrant(dot)com> writes:
>> Can we add something to memory contexts to make this fail every time?
>
> No, not really. AFAICS the reason for Alvaro not seeing it must be that
> on his machine the new transaction happens to allocate its
> TopTransactionContext control block right in the same place where the
> old one was.
>
> We could have a debugging mode in which pfree'd space is never recycled
> for reuse (just filled with 0xdeadbeef or whatever and left to sit).

Hm, I wonder how much more practical it would be if we recycled it but offset
the pointer by maxalign. We would waste 4/8 bytes per palloc/free cycle
instead of the whole chunk.

(Whether palloc could actually do this would be another question, there are
lots of allocator algorithms that wouldn't be able to, I think.)

If we had a more formalized save_memory_context()/restore_memory_context()
which gave you more than just a pointer we could do better for this particular
case. save_memory_context() could hand you a struct with a pointer as well as
some sanity check values about the context you're saving.

--
Gregory Stark
EnterpriseDB http://www.enterprisedb.com
Ask me about EnterpriseDB's On-Demand Production Tuning