Re: Excessive PostmasterIsAlive calls slow down WAL redo

On Wed, Apr 11, 2018 at 10:22 PM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:
> On 10/04/18 04:36, Thomas Munro wrote:
>> Just an idea, not tested: what about a reusable WaitEventSet with zero
>> timeout?  Using the kqueue patch, that'd call kevent() which'd return
>> immediately and tell you if any postmaster death notifications had
>> arrive on your queue since last time you asked.  It doesn't even touch
>> the pipe, or any other kernel objects apart from your own queue IIUC.
>
> Hmm. In PostmasterIsAlive(), you'd still need to call kevent() to check if
> postmaster has died. It would just replace the current read() syscall on the
> pipe with the kevent() syscall. Is it faster?

It should be (based on the report of read() being slow here because of
contention on the pipe itself, I guess because of frequent poll() in
WaitLatch() elsewhere?).

But as I said over on another thread[1] (sorry, it got tangled up with
that other conversation about a related topic), maybe testing
getppid() would be simpler and about as fast as possible given you
have to make a syscall (all processes should normally be children of
postmaster, right?).  And only check every nth time through the loop,
as you said, to avoid high frequency syscalls.  I think I might have
been guilty of having a solution looking for a problem, there ;-)

[1] https://www.postgresql.org/message-id/CAEepm%3D298omvRS2C8WO%3DCxp%2BWcM-Vn8V3x4_QhxURhKTRCSfYg%40mail.gmail.com

-- 
Thomas Munro
http://www.enterprisedb.com