From: Brandon Maier via buildroot <buildroot@buildroot.org>
To: yann.morin@orange.com
Cc: Herve Codina <herve.codina@bootlin.com>,
"Yann E. MORIN" <yann.morin.1998@free.fr>,
Thomas Petazzoni <thomas.petazzoni@bootlin.com>,
buildroot@buildroot.org
Subject: Re: [Buildroot] [PATCH RESEND v2] ppd-merge: speed up per-package-rsync
Date: Thu, 13 Jun 2024 12:59:00 +0000 [thread overview]
Message-ID: <ZmrtFMBU1ByWEsqh@bspbox.kirk> (raw)
In-Reply-To: <ZmqDYG4BrrY7ybPI@tl-lnx-nyma7486-2>
Hi Yann,
On 2024.06.13 07:28, yann.morin@orange.com wrote:
> Brandon, All,
>
> On 2024-06-12 18:22 +0000, Brandon Maier via buildroot spake thusly:
> > From: Brandon Maier via buildroot <buildroot@buildroot.org>
> > The per-package-rsync stage can add a significant amount of time to
> > builds. They can also be annoying as the target-finalize and
> > host-finalize targets are the slowest and run on every `make all`, which
> > is used frequently for partial rebuilds.
> >
> > The per-package-rsync is slow because it launches a new rsync for each
> > source tree, and each rsync must rescan the destination directory and
> > potentially overwrite files multiple times. We can instead merge all the
> > rsync calls down into one call, and rsync is smarter about scanning all
> > the source directories and only copying over the files it needs to.
> >
> > We feed the source trees to rsync in reverse-order, as this preserves
> > the original behaviour. I.e. when using multiple rsyncs, the last source
> > tree would overwrite anything in the destination. Now when using a
> > single rsync, we put the last tree first as rsync will select the first
> > file it finds.
> >
> > This only supports the 'copy' mode, which is used in the finalize step.
> > The 'hardlink' mode requires specifying each source tree with the
> > --link-dest flag but enforces a maximum of 20 trees.
>
> I think that last sentence for the hardlink mode no longer applies with
> this new iteration, since the hardlink case is no longer handled.
I was trying to explain why this patch only updates `per-package-rsync
copy`, and does not update `per-package-rsync hardlink`. Because rsync's
hardlink support works differently. But we can drop this paragraph if
it's confusing.
>
> > Below is a benchmark running the host-finalize target for a build with
> > 200 packages.
> >
> > Benchmark 1: before copy
> > Time (mean ± σ): 27.171 s ± 0.777 s [User: 6.170 s, System: 14.830 s]
> > Range (min … max): 26.343 s … 28.566 s 10 runs
> >
> > Benchmark 2: after copy
> > Time (mean ± σ): 6.296 s ± 0.196 s [User: 2.874 s, System: 5.600 s]
> > Range (min … max): 6.094 s … 6.709 s 10 runs
>
> Impressive! I don't get quite the same improvements here, as my build is
> smaller, but I get from ~21.715 to ~9.080 (5 runs each), which is more
> than twice (~2.4x) faster. Good! :-)
>
> To be noted: your numbers as well as mines imply a running host-finalize
> after a previous completed, successful build, of course.
>
> So, I also did a test to see what happens when the host/ directory did
> not previously exist:
> $ make; rm -rf output/host/
> $ time make host-finalize
>
> And even in that case, I get about a 2.2x improvement too, which is also
> interesting.
Yes, a lot of the slowdown has to do with running multiple rsyncs. Each
rsync scans over output/host/ and overwrites any files from its
per-package directory that doesn't match. So even when output/host/
already exists, a lot of files get overwritten anyway.
This will get worse once your host-finalize patch is applied. As it will
rewrite many files in output/host/ so they no longer match the
per-package/ files.
>
> Finally, I compared the content of host/ in both cases:
>
> $ make; mv output/host{,.slow}
> $ make; mv output/host{,.fast}
> $ diff durN output/host{.slow,.fast}
>
> The only results are dangling symlinks that are only valid at runtime
> (e.g. [XXX]/staging/dev/stdout that points to ../proc/self/stdout).
I tested this and it appears `diff` is printing errors because it can't
find the symlink. But the symlinks themselves are identical, so this
shouldn't be a regression.
$ diff -drN -- output/host{.slow,.fast}
diff: output/host.slow/arm-buildroot-linux-gnueabihf/sysroot/dev/fd: No such file or directory
diff: output/host.fast/arm-buildroot-linux-gnueabihf/sysroot/dev/fd: No such file or directory
$ ls -al output/host{.slow,.fast}/arm-buildroot-linux-gnueabihf/sysroot/dev/fd
lrwxrwxrwx 1 blmaier genusers 15 Jun 11 21:47 output/host.fast/arm-buildroot-linux-gnueabihf/sysroot/dev/fd -> ../proc/self/fd
lrwxrwxrwx 1 blmaier genusers 15 Jun 11 21:47 output/host.slow/arm-buildroot-linux-gnueabihf/sysroot/dev/fd -> ../proc/self/fd
>
> Tested-by: Yann E. MORIN <yann.morin@orange.com>
> Reviewed-by: Yann E. MORIN <yann.morin@orange.com>
>
> Regards,
> Yann E. MORIN.
Thank you for testing!
Brandon Maier
>
> > Summary
> > after copy ran
> > 4.32 ± 0.18 times faster than before copy
> >
> > Cc: Herve Codina <herve.codina@bootlin.com>
> > Cc: Yann E. MORIN <yann.morin.1998@free.fr>
> > Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
> > Reviewed-by: Herve Codina <herve.codina@bootlin.com>
> > Signed-off-by: Brandon Maier <brandon.maier@collins.com>
> > ---
> > v1: https://patchwork.ozlabs.org/project/buildroot/patch/20231127224139.35969-1-brandon.maier@collins.com/
> >
> > v2:
> > - Simplify logic for 'copy' mode
> > - Drop support for 'hardlink' mode
> > ---
> > package/pkg-utils.mk | 18 ++++++------------
> > 1 file changed, 6 insertions(+), 12 deletions(-)
> >
> > diff --git a/package/pkg-utils.mk b/package/pkg-utils.mk
> > index d1964299af..3ab9e9fe52 100644
> > --- a/package/pkg-utils.mk
> > +++ b/package/pkg-utils.mk
> > @@ -221,18 +221,12 @@ ifeq ($(BR2_PER_PACKAGE_DIRECTORIES),y)
> > # $4: literal "copy" or "hardlink" to copy or hardlink files from src to dest
> > define per-package-rsync
> > mkdir -p $(3)
> > - $(foreach pkg,$(1),\
> > - rsync -a \
> > - --hard-links \
> > - $(if $(filter hardlink,$(4)), \
> > - --link-dest=$(PER_PACKAGE_DIR)/$(pkg)/$(2)/, \
> > - $(if $(filter copy,$(4)), \
> > - $(empty), \
> > - $(error per-package-rsync can only "copy" or "hardlink", not "$(4)") \
> > - ) \
> > - ) \
> > - $(PER_PACKAGE_DIR)/$(pkg)/$(2)/ \
> > - $(3)$(sep))
> > + $(if $(filter hardlink,$(4)), \
> > + $(foreach pkg,$(1),\
> > + rsync -a --hard-links --link-dest=$(PER_PACKAGE_DIR)/$(pkg)/$(2)/ \
> > + $(PER_PACKAGE_DIR)/$(pkg)/$(2)/ $(3)$(sep)), \
> > + printf "%s/$(2)/\n" $(1) | tac \
> > + | rsync -a --hard-links --files-from=- --no-R -r $(PER_PACKAGE_DIR) $(3))
> > endef
> >
> > # prepares the per-package HOST_DIR and TARGET_DIR of the current
> >
> > ---
> > base-commit: 12b2860d7562d5cf1b7492b58f4f10f0405444c4
> > change-id: 20240612-ppd-merge-b4-8c623ba7ab90
> >
> > Best regards,
> > --
> > Brandon Maier <brandon.maier@collins.com>
> >
> > _______________________________________________
> > buildroot mailing list
> > buildroot@buildroot.org
> > https://lists.buildroot.org/mailman/listinfo/buildroot
>
> --
> ____________
> .-----------------.--------------------: _ :------------------.
> | Yann E. MORIN | Real-Time Embedded | __/ ) | /"\ ASCII RIBBON |
> | | Software Designer | _/ - /' | \ / CAMPAIGN |
> | +33 638.411.245 '--------------------: (_ `--, | X AGAINST |
> | yann.morin (at) orange.com |_=" ,--' | / \ HTML MAIL |
> '--------------------------------------:______/_____:------------------'
>
> ____________________________________________________________________________________________________________
> Ce message et ses pieces jointes peuvent contenir des informations confidentielles ou privilegiees et ne doivent donc
> pas etre diffuses, exploites ou copies sans autorisation. Si vous avez recu ce message par erreur, veuillez le signaler
> a l'expediteur et le detruire ainsi que les pieces jointes. Les messages electroniques etant susceptibles d'alteration,
> Orange decline toute responsabilite si ce message a ete altere, deforme ou falsifie. Merci.
>
> This message and its attachments may contain confidential or privileged information that may be protected by law;
> they should not be distributed, used or copied without authorisation.
> If you have received this email in error, please notify the sender and delete this message and its attachments.
> As emails may be altered, Orange is not liable for messages that have been modified, changed or falsified.
> Thank you.
> _______________________________________________
> buildroot mailing list
> buildroot@buildroot.org
> https://lists.buildroot.org/mailman/listinfo/buildroot
_______________________________________________
buildroot mailing list
buildroot@buildroot.org
https://lists.buildroot.org/mailman/listinfo/buildroot
next prev parent reply other threads:[~2024-06-13 12:59 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-06-12 18:22 [Buildroot] [PATCH RESEND v2] ppd-merge: speed up per-package-rsync Brandon Maier via buildroot
2024-06-13 5:28 ` yann.morin
2024-06-13 12:59 ` Brandon Maier via buildroot [this message]
2024-06-13 13:09 ` yann.morin
2024-06-15 18:56 ` Yann E. MORIN
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=ZmrtFMBU1ByWEsqh@bspbox.kirk \
--to=buildroot@buildroot.org \
--cc=brandon.maier@collins.com \
--cc=herve.codina@bootlin.com \
--cc=thomas.petazzoni@bootlin.com \
--cc=yann.morin.1998@free.fr \
--cc=yann.morin@orange.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox