[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Getting file extensions [linux fs]

Have you looked into eBPF? They have mature Python bindings. It makes
interacting with the kernel as efficient as possible - you can run it in
production at high resolutions without putting things at risk. One of the
benefits - any averaging / aggregation / histograms / etc can be done by
the kernel within the eBPF runtime, then passed back to python using eBPF
"maps" - the link between your userspace program and the eBPF kernel code.
>From python this "map" is just a dict.

You'll need to be running at least kernel version 4.4 to get the basic
functionality, but ideally 4.9 or higher for all the newer and stable
features. No dependencies, its baked into the kernel. You will need clang
support to compile stuff, if you want to build modules on your own.

*Pablo Lucena*

On Sat, Mar 30, 2019 at 8:30 PM Paulo da Silva <
p_s_d_a_s_i_l_v_a_ns at netcabo.pt> wrote:

> ?s 22:18 de 28/03/19, Cameron Simpson escreveu:
> > On 28Mar2019 01:12, Paulo da Silva <p_s_d_a_s_i_l_v_a_ns at netcabo.pt>
> wrote:
> >> ?s 23:09 de 27/03/19, Cameron Simpson escreveu:
> ...
> >
> > Oh, just tangential to this.
> >
> > If you were doing this ad hoc, yes calling the filefrag executable is
> > very expensive. But if you are always doing a large batch of filenames
> > invoking:
> >
> >  filefrag lots of filenames here ...>
> > and reading from its output can be very effective, because the expense
> > of the executable is amortized over all the files - the per file cost is
> > much reduced. And it saves you working out how to use the ioctls from
> > Python :-)
> That's not the case.
> I need to do it on some files basis which I don't know in advance.
> Using IOCTL, I don't need to parse or unpack the output. Only compare
> the output arrays. Besides I need to store many of the outputs. Doing
> that from filefrag text output would be unpractical. I needed, at least,
> to compress the data. Although may be I might have to compress the ioctl
> arrays ... Let's see how big in average is the storage needed.
> I have to go with ioctl. I have to open the files anyway, so there is no
> overhead for that when calling the ioctl.
> Anyway, thank you for the suggestion.
> Regards.
> Paulo
> --
> https://mail.python.org/mailman/listinfo/python-list