[OE-core] [PATCH] openssl: fix occasionally install fail for parallel build

richard.purdie at linuxfoundation.org richard.purdie at linuxfoundation.org
Mon Oct 29 16:30:38 UTC 2018


On Mon, 2018-10-29 at 09:18 -0700, Khem Raj wrote:
> On Tue, Oct 23, 2018 at 11:53 PM Changqing Li
> <changqing.li at windriver.com> wrote:
> > 
> > 
> > 
> > On 10/24/2018 02:04 PM, Richard Purdie wrote:
> > > On Wed, 2018-10-24 at 11:30 +0800, changqing.li at windriver.com
> > > wrote:
> > > > From: Changqing Li <changqing.li at windriver.com>
> > > > 
> > > > when do parallel build, like run "make install -j 48", openssl
> > > > will
> > > > do_install
> > > > failed with below error:
> > > > mv: cannot stat 'openssl/1.1.1-
> > > > r0/image/usr/lib/libcrypto.so.1.1.new': No such file or
> > > > directory
> > > > Makefile:318: recipe for target 'install_dev' failed
> > > > 
> > > > problem caused by target install_dev and target
> > > > install_runtime, both
> > > > installed
> > > > libcrypto.so.1.1, result in log:
> > > > *** Installing development files
> > > > ...
> > > > install libcrypto.so.1.1 -> /var/tmp/foo/lib/libcrypto.so.1.1
> > > > link /var/tmp/foo/lib/libcrypto.so ->
> > > > /var/tmp/foo/lib/libcrypto.so.1.1
> > > > ...
> > > > *** Installing runtime files
> > > > install libcrypto.so.1.1 -> /var/tmp/foo/lib/libcrypto.so.1.1
> > > > 
> > > > command in install_dev and install_runtime will have parallel
> > > > competition.
> > > > 
> > > > have create an issue to upstream:
> > > > 
https://github.com/openssl/openssl/issues/7466#issuecomment-432148137
> > > > 
> > > > but upstream won't support this condition, so use this fix as a
> > > > workaround to fix the problem.
> > > 
> > > This means that:
> > > 
> > > a) Upstream don't support parallel make install
> > > b) Reject patches fixing any bugs in parallel make install
> > > 
> > > This means the Upstream-Status should be rejected with the link
> > > above,
> > > not pending. It also means we may need to think about not using
> > > parallel install although since its working, its tempting to
> > > carry
> > > patches and keep using it if it works.
> > > 
> > > It may be worth pointing out to upstream that it is mostly
> > > working for
> > > us? I am a little worried they know something we don't about
> > > problems
> > > this could run into.
> > > 
> > > Cheers,
> > > 
> > > Richard
> > > 
> > > 
> > 
> > I will point out to upstream that it is mostly working for us,  and
> > ask upstream suggestion about potential problem of parallel build.
> > 
> > if upstream think no obvious potential problem can seen,  I will
> > change
> > upstream-status to reject and resend the patch.
> > 
> > if upstream recommend not to do parallel build,  I will add
> > something
> > like below
> > to stop parallel build for make install, and resend the patch, but
> > in
> > this case,  build
> > openssl  will very slow.
> > 
> > ifneq ($(filter install,$(MAKECMDGOALS)),)
> > .NOTPARALLEL:
> > endif
> 
> I think its best to convince or follow upstream here. Making install
> target as serial might be best approach if there are other issues
> with
> parallel install that upstream is uncomfortable with such a patch we
> should try to find more.

If you follow the upstream discussion they originally said no, then
they are possibly reconsidering after I explained more about the
situation. I've not seen a fix land yet but they did say they are
working on one.

If anyone wants to +1 them fixing it, feel free:

https://github.com/openssl/openssl/issues/7466#issuecomment-432938981

Cheers,

Richard






More information about the Openembedded-core mailing list