On Wed, Dec 9, 2015 at 5:31 PM, Amit Kapila <amit.kapila16@gmail.com> wrote:
>
> Another bigger issue I see in the above part of code is that it doesn't
> seem to be safe to call load_hba() at that place in PostgresMain() as
> currently load_hba() is using a context created from PostmasterContext
> to perform the parsing and some other stuff, the PostmasterContext
> won't be available at that time. It is deleted immediately after
> InitPostgres
> is completed. So either we need to make PostmasterContext don't go
> away after InitPostgres() or load_hba shouldn't use it and rather use
> CurrentMemroyContext similar to ProcessConfigFile or may be use
> TopMemoryContext instead of PostmasterContext if possible. I think
> this needs some more thoughts.
>
> Apart from above, this patch doesn't seem to work on Windows or
> for EXEC_BACKEND builds as we are loading the hba file in a
> temporary context (PostmasterContext, refer PerformAuthentication)
> which won't be alive for the backends life. This works on linux because
> of fork/exec mechanism which allows to inherit the parsed file
> (parsed_hba_lines). This is slightly tricky problem to solve and we
> have couple of options (a) use TopMemoryContext instead of Postmaster
> Context to load hba; (b) Use CurrentMemoryContext (c) pass the parsed
> hba file for Windows/Exec_Backend using save_backend_variables/
> restore_backend_variables mechanism or if you have any other idea.
> If you don't have any better idea, then you can evaluate above ideas
> and see which one makes more sense.
Reverting the context release patch is already provided in the first
mail of this
thread [1]. Forgot to mention about the same in further mails.
Here I attached the same patch. This patch needs to be applied first before
pg_hba_lookup patch. I tested it in windows version also.
[1] - http://www.postgresql.org/message-id/CAJrrPGfFyf45mfK7uB+QHWhXn_tTMkNRVhTuDefQzuZZRwEeQg@mail.gmail.com
Regards,
Hari Babu
Fujitsu Australia