NetBSD Problem Report #57799

From martin@aprisoft.de  Thu Dec 28 09:52:18 2023
Return-Path: <martin@aprisoft.de>
Received: from mail.netbsd.org (mail.netbsd.org [199.233.217.200])
	(using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits))
	(Client CN "mail.NetBSD.org", Issuer "mail.NetBSD.org CA" (not verified))
	by mollari.NetBSD.org (Postfix) with ESMTPS id 1CF0C1A9238
	for <gnats-bugs@gnats.NetBSD.org>; Thu, 28 Dec 2023 09:52:18 +0000 (UTC)
Message-Id: <20231228095208.8ABD05CC795@emmas.aprisoft.de>
Date: Thu, 28 Dec 2023 10:52:08 +0100 (CET)
From: martin@NetBSD.org
Reply-To: martin@NetBSD.org
To: gnats-bugs@NetBSD.org
Subject: llvm amd64 builds are unbearbly slow and hit the build cluster time limit
X-Send-Pr-Version: 3.95

>Number:         57799
>Category:       toolchain
>Synopsis:       llvm amd64 builds are unbearbly slow and hit the build cluster time limit
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    toolchain-manager
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Thu Dec 28 09:55:00 +0000 2023
>Last-Modified:  Mon May 13 05:30:01 +0000 2024
>Originator:     Martin Husemann
>Release:        NetBSD 10.99.10
>Organization:
The NetBSD Foundation, Inc.
>Environment:
System: NetBSD big-apple.aprisoft.de 10.99.10 NetBSD 10.99.10 (POWERMAC_G5.MP) #5: Fri Dec 15 21:02:28 CET 2023 martin@martins.aprisoft.de:/usr/src/sys/arch/macppc/compile/POWERMAC_G5.MP macppc
Architecture: powerpc
Machine: macppc
>Description:

Build for amd64 with llvm fail both on HEAD and netbsd-10 on the build cluster,
because they hit the 4h-per-single-arch-build time limit.

They are rund like this:

build.sh -m amd64 -B 202312241230Z -V NETBSD_OFFICIAL_RELEASE=no -D /home/builds/ab/HEAD-llvm/amd64/202312241230Z-dest -M /home/builds/ab/HEAD-llvm/amd64/202312241230Z-obj -R /home/builds/ab/HEAD-llvm/amd64/202312241230Z-rlse -T /home/builds/ab/HEAD-llvm/amd64/202312241230Z-tools -U -X /home/source/ab/HEAD-llvm/xsrc -x -N0 -V TMPDIR=/tmp -V BUILD=yes -V MKLLVM=yes -V HAVE_LLVM=yes -V MKGCC=no -V MAKECONF=/home/builds/etc/make.conf -V MKDEBUG=yes -j16 release iso-image install-image live-image


>How-To-Repeat:
s/a

>Fix:
n/a

>Audit-Trail:
From: Martin Husemann <martin@duskware.de>
To: gnats-bugs@NetBSD.org
Cc: 
Subject: Re: toolchain/57799: amd64 builds are unbearbly slow
Date: Sat, 30 Dec 2023 14:03:35 +0100

 Two observations:

  - nothing has "seriously" changed recently to get into this situation,
    but the builds have always been very close to the limit. Last working
    HEAD build was:

 	HEAD-llvm:202312111450Z amd64 b41 13711 passed

    (that is 13711 seconds, with 4h = 14400s being the current limit)

  - the gcc builds are a bit faster, after mrg reworked the amount/detail
    of debug info that is generated for the C++ parts of the compiler
    some time ago - but they are not very much faster, example:

 	HEAD:202312291920Z amd64 b44 12255 passed

 So the real problem is: amd64 builds in general are extremely slow.
 Other builds are way faster, e.g. alpha (which has no 32bit compat):

 	HEAD:202312290600Z alpha b43 5408 passed

 but also sparc64, which has 32bit compat (but no llvm compiled GPU stuff):

 	HEAD:202312291920Z sparc64 localhost 5845 passed

 but here the difference to an llvm build is even more significant:

 	HEAD-llvm:202312291450Z sparc64 b48 8697 passed



 For now I have raised the time limit to 4.5h

 Martin

From: Martin Husemann <martin@duskware.de>
To: gnats-bugs@NetBSD.org
Cc: 
Subject: Re: toolchain/57799: amd64 builds are unbearbly slow
Date: Mon, 13 May 2024 07:26:00 +0200

 We had to increase the timeout to 6h now - 5.5h was not enough.

 This is not acceptable, we should drop the regular LLVM builds from the
 daily builds and only fix fallout when it is reported.

 Martin

Home	PR Database Search