Groups | Search | Server Info | Keyboard shortcuts | Login | Register

Groups > comp.os.linux.misc > #85635

Re: Does Linux Have Better Solution Than "inotify" ?

Show all headers | View raw

c186282 <c186282@nnada.net> wrote:
> On 4/7/26 23:58, Rich wrote:
>> c186282 <c186282@nnada.net> wrote:
>>> Say you want to promptly duplicate what ever has been writ to your
>>> disk on another disk, or remote drive or cloud drive or whatever.
>>>
>>> 'inotify' and its kin can tell you what's changed,
>> 
>> Yep, that is its purpose.
>> 
>>> but it's basically a POLLING sort of operation and
>> 
>> Incorrect.  Inotify is an event driven system.  There is no polling
>> (using the standard definiton of "polling" [1]) going on for inotify
>> events.
>> 
>>> including an entire drive would be insanely slow.
>> 
>> Yes, but only because setting up the inotify notifications would take a
>> while to do, and likely need a fair amount of kernel ram to store them.
>> 
>> But once setup, there is no polling of any form going on.
> 
>   Ya SURE about that ???

That's the claim on the tin.

>   I did some experiments a few years back and even if you added 
>   "nothing happening" dirs the performance got worse.

I'd have to try an experiment to see, but even if "performance got 
worse" that does not necessarily equate to "polling".

>>> It's a sort of localized rsync daemon.
>> 
>> Completely wrong, in so many ways.
>> 
>>> Now SOMEWHERE Linux does put together a filename and destination as
>>> it's about to write.  The big question is where, and can users SEE
>>> this ?
>> 
>> Inotify.
>> 
>>> SoftRAID software does seem to tap into something like this, so it's
>>> THERE, somewhere.  I tried looking at softRAID code but it's huge and
>>> only just so well documented.  No luck - I can't find what it's
>>> looking at - kernel, driver, who knows ?
>> 
>> Unless it installs a custom kernel module (unlikely) then it is most
>> likely using the inotify subsystem.
> 
>   Then RAID is essentially 'inotify("/","-r")' yet it's lightning
>   fast and doesn't use up all the kernel RAM doing it.

No, RAID (Linux MD RAID) is a kernel driver that internally handles the 
write to plural disks plus checksumming to the checksum disks, and 
operates at the block level, not 'file' level.

As to SoftRAID, I have no idea what that one is nor how it works.

>> [1] https://en.wikipedia.org/wiki/Polling_(computer_science)
>> But as you are well known for having made up your own non-standard
>> definitions for CS terms, I felt the need to point to the standard
>> definition.
> 
>   "Polling" in THIS case would mean reading all the file names/times
>   in a folder/subfolders and then reading them again X seconds later
>   to see if anything had changed. This could get very tedious if a
>   large folder is involved. I was looking for something a bit more
>   pro-active, only reporting what HAD changed without having to go
>   looking for it on the disk.

And that is supposed to be the inotify subsystem.

Back to comp.os.linux.misc | Previous | Next — Previous in thread | Next in thread | Find similar

Thread

Does Linux Have Better Solution Than "inotify" ? c186282 <c186282@nnada.net> - 2026-04-07 20:14 -0400
  Re: Does Linux Have Better Solution Than "inotify" ? Rich <rich@example.invalid> - 2026-04-08 03:58 +0000
    Re: Does Linux Have Better Solution Than "inotify" ? c186282 <c186282@nnada.net> - 2026-04-08 13:03 -0400
      Re: Does Linux Have Better Solution Than "inotify" ? Rich <rich@example.invalid> - 2026-04-09 21:51 +0000
    Re: Does Linux Have Better Solution Than "inotify" ? Richard Kettlewell <invalid@invalid.invalid> - 2026-04-08 18:10 +0100
      Re: Does Linux Have Better Solution Than "inotify" ? c186282 <c186282@nnada.net> - 2026-04-08 13:21 -0400
      Re: Does Linux Have Better Solution Than "inotify" ? Rich <rich@example.invalid> - 2026-04-09 21:52 +0000

csiph-web