> On 30 Mar 2021, at 08:55, Devendra Tewari <devendra.tew...@gmail.com> wrote:
> 
> I understand that parallel tasks can result in such issues, but would expect 
> some kind of dependency graph that would prevent them from happening. I wish 
> I had more time to understand the issue fully.
> 
> Here’s the last version of my patch that still uses os.rename, and on error 
> uses shutil.move, with the reasoning explained in the commit message. There 
> are tons of os.rename in code elsewhere, and I’ve encountered similar issues 
> when doing incremental builds, but I’ll leave fixing them to those with more 
> time at hand. Cheers.

Here’s a patch that catches error with all os.rename in the code, and retries 
with shutil.move when that happens.


> 
> From aeba1f9728f69cdf2ce4ba285be86761d8024f9d Mon Sep 17 00:00:00 2001
> From: Devendra Tewari <devendra.tew...@gmail.com>
> Date: Tue, 30 Mar 2021 08:27:40 -0300
> Subject: [PATCH] Use shutil.move when os.rename fails
> 
> Incremental build in Docker fails with
> 
> OSError: [Errno 18] Invalid cross-device link
> 
> When source and destination are on different overlay filesystems.
> 
> This change handles error with os.rename and retries with shutil.move.
> The reason os.rename is still used is because shutil.move is too slow
> for speed sensitive sections of code.
> ---
> meta/classes/sstate.bbclass | 22 ++++++++++++++++++----
> 1 file changed, 18 insertions(+), 4 deletions(-)
> 
> diff --git a/meta/classes/sstate.bbclass b/meta/classes/sstate.bbclass
> index f579168162..301dfc27db 100644
> --- a/meta/classes/sstate.bbclass
> +++ b/meta/classes/sstate.bbclass
> @@ -384,6 +384,7 @@ def sstate_installpkg(ss, d):
> def sstate_installpkgdir(ss, d):
>     import oe.path
>     import subprocess
> +    import shutil
> 
>     sstateinst = d.getVar("SSTATE_INSTDIR")
>     d.setVar('SSTATE_FIXMEDIR', ss['fixmedir'])
> @@ -401,7 +402,10 @@ def sstate_installpkgdir(ss, d):
> 
>     for state in ss['dirs']:
>         prepdir(state[1])
> -        os.rename(sstateinst + state[0], state[1])
> +        try:
> +            os.rename(sstateinst + state[0], state[1])
> +        except OSError:
> +            shutil.move(sstateinst + state[0], state[1])
>     sstate_install(ss, d)
> 
>     for plain in ss['plaindirs']:
> @@ -413,7 +417,10 @@ def sstate_installpkgdir(ss, d):
>         dest = plain
>         bb.utils.mkdirhier(src)
>         prepdir(dest)
> -        os.rename(src, dest)
> +        try:
> +            os.rename(src, dest)
> +        except OSError:
> +            shutil.move(src, dest)
> 
>     return True
> 
> @@ -638,6 +645,7 @@ python sstate_hardcode_path () {
> 
> def sstate_package(ss, d):
>     import oe.path
> +    import shutil
> 
>     tmpdir = d.getVar('TMPDIR')
> 
> @@ -664,7 +672,10 @@ def sstate_package(ss, d):
>                     continue
>                 bb.error("sstate found an absolute path symlink %s pointing 
> at %s. Please replace this with a relative link." % (srcpath, link))
>         bb.debug(2, "Preparing tree %s for packaging at %s" % (state[1], 
> sstatebuild + state[0]))
> -        os.rename(state[1], sstatebuild + state[0])
> +        try:
> +            os.rename(state[1], sstatebuild + state[0])
> +        except OSError:
> +            shutil.move(state[1], sstatebuild + state[0])
> 
>     workdir = d.getVar('WORKDIR')
>     sharedworkdir = os.path.join(d.getVar('TMPDIR'), "work-shared")
> @@ -674,7 +685,10 @@ def sstate_package(ss, d):
>             pdir = plain.replace(sharedworkdir, sstatebuild)
>         bb.utils.mkdirhier(plain)
>         bb.utils.mkdirhier(pdir)
> -        os.rename(plain, pdir)
> +        try:
> +            os.rename(plain, pdir)
> +        except OSError:
> +            shutil.move(plain, pdir)
> 
>     d.setVar('SSTATE_BUILDDIR', sstatebuild)
>     d.setVar('SSTATE_INSTDIR', sstatebuild)
> -- 
> 2.29.2
> 
> 
>> On 30 Mar 2021, at 08:10, Richard Purdie 
>> <richard.pur...@linuxfoundation.org> wrote:
>> 
>> On Mon, 2021-03-29 at 16:00 -0700, Andre McCurdy wrote:
>>> On Mon, Mar 29, 2021 at 3:45 PM Devendra Tewari
>>> <devendra.tew...@gmail.com> wrote:
>>>> 
>>>> Thanks! My bad. The example I looked up in Python docs had a break and I 
>>>> just realized it was a looping example.
>>>> 
>>>> Here’s the updated patch (or should I submit it again via git send-email?)
>>> 
>>> It would be better to use shutil.move unconditionally in all cases,
>>> rather than have a separate shutil.move code path which only gets
>>> tested by people doing incremental builds in docker.
>> 
>> This is a speed sensitive section of code and shutil has traditionally 
>> proved 
>> to be slow so I disagree with that, rename is very much preferred here where
>> we can.
>> 
>> Cheers,
>> 
>> Richard
>> 
>> 
> 

Attachment: 0001-use-shutil.move-when-os.rename-fails.patch
Description: Binary data

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#150682): 
https://lists.openembedded.org/g/openembedded-core/message/150682
Mute This Topic: https://lists.openembedded.org/mt/81698791/21656
Group Owner: openembedded-core+ow...@lists.openembedded.org
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub 
[arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-

Reply via email to