Re: Is custom MemoryContext prohibited?

2020年1月28日(火) 23:09 Tomas Vondra <tomas(dot)vondra(at)2ndquadrant(dot)com>:
>
> On Tue, Jan 28, 2020 at 10:55:11AM +0900, Kohei KaiGai wrote:
> >Hello,
> >
> >I noticed MemoryContextIsValid() called by various kinds of memory context
> >routines checks its node-tag as follows:
> >
> >#define MemoryContextIsValid(context) \
> > ((context) != NULL && \
> > (IsA((context), AllocSetContext) || \
> > IsA((context), SlabContext) || \
> > IsA((context), GenerationContext)))
> >
> >It allows only "known" memory context methods, even though the memory context
> >mechanism enables to implement custom memory allocator by extensions.
> >Here is a node tag nobody used: T_MemoryContext.
> >It looks to me T_MemoryContext is a neutral naming for custom memory context,
> >and here is no reason why memory context functions prevents custom methods.
> >
>
> Good question. I don't think there's an explicit reason not to allow
> extensions to define custom memory contexts, and using T_MemoryContext
> seems like a possible solution. It's a bit weird though, because all the
> actual contexts are kinda "subclasses" of MemoryContext. So maybe adding
> T_CustomMemoryContext would be a better choice, but that only works in
> master, of course.
>
From the standpoint of extension author, reuse of T_MemoryContext and
back patching the change on MemoryContextIsValid() makes us happy. :)
However, even if we add a new node-tag here, the custom memory-context
can leave to use fake node-tag a few years later. It's better than nothing.

> Also, it won't work if we need to add memory contexts to equalfuncs.c
> etc. but maybe won't need that - it's more a theoretical issue.
>
Right now, none of nodes/XXXfuncs.c support these class of nodes related to
MemoryContext. It shall not be a matter.

> >https://github.com/heterodb/pg-strom/blob/master/src/shmbuf.c#L1243
> >I recently implemented a custom memory context for shared memory allocation
> >with portable pointers. It shall be used for cache of pre-built gpu
> >binary code and
> >metadata cache of apache arrow files.
> >However, the assertion check above requires extension to set a fake node-tag
> >to avoid backend crash. Right now, it is harmless to set T_AllocSetContext, but
> >feel a bit bad.
> >
>
> Interesting. Does that mean the hared memory contexts are part of the
> same hierarchy as "normal" contexts? That would be a bit confusing, I
> think.
>
If this shared memory context is a child of normal context, likely, it should be
reset or deleted as usual. However, if this shared memory context performs
as a parent of normal context, it makes a problem when different process tries
to delete this context, because its child (normal context) exists at the creator
process only. So, it needs to be used carefully.

Thanks,
--
HeteroDB, Inc / The PG-Strom Project
KaiGai Kohei <kaigai(at)heterodb(dot)com>