[Tarantool-patches] [PATCH 4/4] fiber: allocate on_stop triggers using mempool
v.shpilevoy at tarantool.org
Thu Aug 26 23:01:04 MSK 2021
On 26.08.2021 08:14, Oleg Babin via Tarantool-patches wrote:
> Thanks for your review. See my answers below.
> On 25.08.2021 23:34, Vladislav Shpilevoy wrote:
>> Thanks for the patch!
>> See 4 comments below.
>> On 11.08.2021 22:33, Oleg Babin via Tarantool-patches wrote:
>>> From: Oleg Babin<babinoleg at mail.ru>
>>> Previously we use malloc for such purpose however we created
>>> on_stop triggers only if fiber storage was touched. Currently we
>>> create on_stop trigger for each lua fiber. So it seems reasonable
>>> to use mempool for such frequent allocations/deallocations.
>> 1. Does it give any perf win? This commit specifically. Seems like an
>> overkill. A trigger allocation might happen only once per fiber's
>> lifetime, and the fiber's creation is supposed to be much longer than
>> the trigger's malloc. The commit probably gives a tiny fraction of
>> percent improvement or does not change anything. Do you have results?
> It's a good question. I thought thank mempool should be faster than malloc.
> We can easily drop this commit. I had some doubts about it.
According to the benchmarks it seems the mempool here is slower.
I suggest to drop this commit.
More information about the Tarantool-patches