To expand a bit... On Sat, Apr 14, 2018 at 1:46 AM, Yann Ylavic <ylavic....@gmail.com> wrote: > On Sat, Apr 14, 2018 at 12:54 AM, Nick Kew <n...@apache.org> wrote: >> >>> On 13 Apr 2018, at 21:46, Yann Ylavic <ylavic....@gmail.com> wrote: >>> >>> On Fri, Apr 13, 2018 at 7:30 PM, William A Rowe Jr <wr...@rowe-clan.net> >>> wrote: >>>> I'm still unclear why you believe we need APR to purge the elts when the >>>> user chooses to flip the fifo/lifo switch. That seems very wrong to me, the >>>> consumer can do so if that is what they desire. >>> >>> To be clear (with myself): >>> >>> /* What we have */ >>> APU_DECLARE(apr_status_t) apr_reslist_acquire(apr_reslist_t *reslist, >>> void **resource); >>> /* What we want */ >>> #define apr_reslist_acquire_lifo apr_reslist_acquire >>> APU_DECLARE(apr_status_t) apr_reslist_acquire_fifo(apr_reslist_t *reslist, >>> void **resource); >> >> You have a plan to implement that with the existing structs? > > I'm almost there yes, see attached patch (no tested yet..).
The #if 0s there should always be 1s IMHO, and when the ttl is to be checked against the resource we should always peek it as LIFO (i.e. s/fifo/1/ in the first peek_resource() of reslist_acquire() in my patch). This would prevent starvation, and we should possibly do that in 1.6.x too (where apr_reslist_acquire_fifo() can't land). If we do that unconditionnaly, this patch is moot. After all, apr_reslist_maintain() and/hence apr_reslist_release() are already LIFO for recycling resources. (Note that there is a bug in apr_reslist_maintain() which doesn't check if ttl != 0 before expiring resources, hence kills everything between smax and hmax. I have a unit test and plan to commit a fix later). > >> How does it work if an application mixes the two forms of >> acquire randomly, as seems likely if (for example) a reslist is >> shared between applications/components of differing provenances? If finally we need this patch and apr_reslist_acquire_fifo() gets used arbitrarily/concurrently with the non-fifo version, we'd still recycle resources appropriately in order, so I see no issue. Anything besides the performance concern (which isn't one I think)? Regards, Yann.
