On 04-10-2022 23:49, Ludovic Courtès wrote:
> Hi,
> 
> Maxime Devos <maximedevos@telenet.be> skribis:
>
>> 'sync' seems relatively inexpensive to me, compared to the time
>> required for building a package and even more inexpensive compared to
>> the cost of debugging store corruption:
> 
> That’s not a fair comparison.  :-)

Possibly, openjdk is a bit of an extreme case.

> Imagine, you run reconfigure/upgrade;
> that downloads tens to hundreds of store items.  Calling sync(2) after
> each item may be hardly noticeably on an SSD, but I bet it’s going to be
> super expensive on an HDD.  (In the syslogd case, each fsync(2) call—not
> even sync(2)—would cause pauses of several 100s of ms.)

If after some testing, this turns out to be a problem, there are some 
options to avoid this (see: the delaying 'fsync' of the previous response).

> Maybe a good test would be to run a daemon on an “average” HDD with
> sync-before-registering = true and to run ‘perf timechart record’ while
> it’s fetching a large number of substitutes.  That way we’d have
> concrete data to talk about.
> 
> Any takers?  :-)

I don't have a HDD to test sync-before-registering=true with.

Greetings,
Maxime.