bug#39413: 26.2; Emacs gets hung

["ISHIKAWA,chiaki" <ishikawa@yk.rim.or.jp>]

On 2021/08/16 20:35, Eli Zaretskii wrote:
>> Cc: larsi@gnus.org, npostavs@gmail.com, 39413@debbugs.gnu.org,
>>   chiaki.ishikawa@ubin.jp
>> From: "ISHIKAWA,chiaki" <ishikawa@yk.rim.or.jp>
>> Date: Mon, 16 Aug 2021 08:27:01 +0900
>>
>> But if you look slightly above, you will notice CPU core #4 is used 100%
>> (!).
>> That is emacs process. No other process is running that earnestly at
>> that moment.
> If we believe everything these tools show us, maybe.  But those tools
> suffer from the same problem Emacs does on a busy system: the tool
> itself might not get CPU to update its data, so you actually see a
> snapshot of what it saw last time it did get CPU.
>
>> PS: I found profiler-cpu-* functions, but I don't think it is wise to
>> run them during GC since they seem to allocate vector tables. However,
>> taking a snapshot of strack trace every now and then during GC seems
>> attractive for my investigation to figure out WHERE in GC, the excessive
>> time is spent.
> Emacs built-in profiling will not help us here, because it cannot
> profile below Lisp primitives.
>

Very tough.
I will see what I can do and report back if anything interesting shows up.

(The last time I tried to see where the time was spent, GC was busy 
reclaiming |cons| cells.
May not mean much.

I can only recall the old paper and in an 16 GB environment (including 
OS),  we may see some bad behavior.
"


  Address/memory management for a gigantic LISP environment or, GC
  considered harmful

"
by Jon L. White

https://dl.acm.org/doi/abs/10.1145/800087.802797