> On 27 Jan 2018, at 15:55, Yann Ylavic <[email protected]> wrote:
> 
> On Sat, Jan 27, 2018 at 4:41 PM, Yann Ylavic <[email protected] 
> <mailto:[email protected]>> wrote:
>> On Sat, Jan 27, 2018 at 4:39 PM, Yann Ylavic <[email protected]> wrote:
>>> On Sat, Jan 27, 2018 at 3:14 PM, Yann Ylavic <[email protected]> wrote:
>>>> 
>>>> So I wonder if we'd better:
>>>> 1. have a constant file name (no generation suffix) for all systems,
>>>> 2. not unlink/remove the SHM file on pconf cleanup,
>>>> 2'. eventually unlink the ones unused in ++generation (e.g. retained
>>>> global list),
>>>> 3. unlink all on pglobal cleanup.
>>>> 
>>>> Now we'd have a working shm_attach() not only for persisted slotmems,
>>>> while shm_create() would only be called for new SHMs which we know can
>>>> be pre-removed (to work around crashes leaving them registered
>>>> somewhere).
>>>> Also, if attach succeeds on startup (first generation) but the SHM is
>>>> not persisted (an old crash still), we can possibly pass the sizes
>>>> checks and re-create the SHM regardless.
>>>> 
>>>> WDYT?
>>> 
>>> Something like the attached patch (it talks better sometimes...).
>>> Not even tested of course :)
>> 
>> Oops, more than needed in there, here is the good one.
> 
> OK, v2 already, with correct usage of gpool (i.e. pconf), distinct
> from ap_pglobal.
> <mod_slotmem_shm-v2.patch>

Thanks, do you recommend we test this version in our pre-production 
environments?

Regards,
Mark

Reply via email to