RE: Many poudriere builds get stuck in lib-depends and run-depends phases and never succeed

From: Mark Millard <marklmi_at_yahoo.com>
Date: Mon, 10 Oct 2022 21:23:19 UTC
Yuri <yuri_at_FreeBSD.org> wrote on
Date: Mon, 10 Oct 2022 17:56:33 UTC :

> Every once in a while when many builds happen to be all in lib-depends 
> and run-depends phases in poudriere - they get stuck and later all fail 
> together.
> 
> All failures look like in the log below. It's looks like many 'pkg 
> install llvmNN' processes for different builds compete with each other 
> and don't progress.
> 
> 
> Does anybody else have such problem? What could be a solution?
> 
> 
> 
> Thanks,
> 
> Yuri
> 
> 
> 
> 
> 
> ---begin log---
> 
> ===>   qcsxcad-0.6.2.9_2 depends on shared library: libQt5Gui.so - not found
> ===>   Installing existing package /packages/All/qt5-gui-5.15.5p165.pkg
> [13amd64-local-workstation-job-03] Installing qt5-gui-5.15.5p165...
> [13amd64-local-workstation-job-03] `-- Installing mesa-dri-21.3.8...
> [13amd64-local-workstation-job-03] |   `-- Installing libXv-1.0.11_2,1...
> [13amd64-local-workstation-job-03] |   `-- Extracting libXv-1.0.11_2,1: 
> .......... done
> [13amd64-local-workstation-job-03] |   `-- Installing libXvMC-1.0.12...
> [13amd64-local-workstation-job-03] |   `-- Extracting libXvMC-1.0.12: 
> .......... done
> [13amd64-local-workstation-job-03] |   `-- Installing 
> libunwind-20211201_1...
> [13amd64-local-workstation-job-03] |   `-- Extracting 
> libunwind-20211201_1: .......... done
> [13amd64-local-workstation-job-03] |   `-- Installing libxshmfence-1.3_1...
> [13amd64-local-workstation-job-03] |   `-- Extracting 
> libxshmfence-1.3_1: ......... done
> [13amd64-local-workstation-job-03] |   `-- Installing llvm13-13.0.1_3...
> [13amd64-local-workstation-job-03] |   | `-- Installing lua53-5.3.6...
> [13amd64-local-workstation-job-03] |   | `-- Extracting lua53-5.3.6: 
> .......... done
> [13amd64-local-workstation-job-03] |   `-- Extracting llvm13-13.0.1_3: .....
> Failed to install the following 1 package(s): 
> /packages/All/qt5-gui-5.15.5p165.pkg
> *** Error code 1
> 
> ---end log---

Any interesting console messages? /var/log/messages content?
dmesg -a content?

Can you use poudriere bulk -w reasonably in your context for
reproducing the problem? The captured /wrkdir/ trees might
hold clues, especially if you have cores set to be stored in
the local directory instead of in /tmp (avoiding places -w
does not record).


Note: I've similarly captured clang++ failure data via having
it store the files locally (.) instead of in /tmp . There is a
command line option for clang/clang++ for controlling where
the files are put. That can make bulk -w capture the files.

===
Mark Millard
marklmi at yahoo.com