wipacrepo / cvmfs Goto Github PK

The python tables package installs a custom Blosc library, compiling with native optimizations. This generates illegal instructions on other (older) cluster nodes. Probably the easiest solution is to install Blosc ourself and have it pick up on that.

/var/condor/execute/.../dir_22934/condor_exec.exe: line 27: setenv: command not found

Note that this is RHEL6, so /bin/sh should be bash. It should then be using export instead of setenv.

We currently search in libdirlist=${LD_LIBRARY_PATH}:/usr/lib:/usr/lib64:/lib:/lib64:/usr/lib/x86_64-linux-gnu. However, all those extra directories may not be in the LD_LIBRARY_PATH, so should not be considered for shared library linking.

Instead, maybe modify the LD_LIBRARY_PATH and then only search on it.

VOMS_USERCONF is set incorrectly. should be:

export VOMS_USERCONF=/cvmfs/icecube.opensciencegrid.org/py3-v4.1.1/RHEL_7_x86_64/etc/vomses

Right now we have a long list of packages and version numbers. While this works sort-of OK, the dependencies from these packages are not versioned and grab whatever is newest from pypi. The correct way to enforce a frozen versionset is to use a requirements file.

Procedure:

• initial build of new repo version uses unversioned packages (or versions as needed to get stability)
• take snapshot of all package versions with pip freeze, saving to requirements file
• use that requirements file for all future builds of repo version

boost::python::numpy is officially supported as of 1.63. We should stop installing our own version.

Note that affects py2-v3, which should be cleaned up.

(imported from trac: #1595. requester: @briedel)

Have RHEL7 in cvmfs....

add the /data/sim/sim-new/downloads/icevariations/ to cvmfs data

Consider deleting easy_install.pth in a future minor release, as it's deprecated and nothing should be using it anymore.

If python is missing setup.sh gets angry:

[riedel1@dt-login01 build]$ eval `/cvmfs/icecube.opensciencegrid.org/py3-v4.1.1/setup.sh`
/cvmfs/icecube.opensciencegrid.org/py3-v4.1.1/setup.sh: line 118: python: command not found
/cvmfs/icecube.opensciencegrid.org/py3-v4.1.1/setup.sh: line 118: python: command not found
/cvmfs/icecube.opensciencegrid.org/py3-v4.1.1/setup.sh: line 125: python: command not found

check needs to look for python2 and python3 as well rather than just python

Spack is makiing a distinction between SL, RHEL, and CentOS. We will need to either run our own fork, or patch it when downloading. thoughts?

to compile nuflux on cvmfs you need to set the env vars CMAKE_PREFIX_PATH=${SROOT} BOOST_ROOT=${SROOT}. It would cut down on user frustration if these were set by setup.sh. See icecube/nuflux#24 (comment)

Somewhat related to #47, the tcsh bug has returned at a few OSG locations. It appears that the default shell is tcsh but jobs are using bash to execute a shell script. This currently breaks the setup.sh script in the following way:

1. setup.sh detects shell as tcsh
2. uses setenv
3. calling script is bash, so setenv doesn't work

Since we've removed root X11 support, we should go all-in on not supporting GUI in the main repo. But because people like their steamshovel, we should probably have an alternate repo with full graphics support.

The only thing I'm concerned about is people trying to use the GUI repo in production on clusters. Maybe we need a smart detection as to which repo to pick, based on if X is available?

Set up a "build master" from which we can build on all platforms and publish changes.

Maybe a custom buildbot setup?

Is there a reason for the repo to remain private? Otherwise, I'm all about sharing stuff.

Maybe @gonzalomerino has an opinion.

ImportError: No module named zmq

required by ipython notebook

Some updates were missing from py2-v1 setup.sh, which seem important. They've been added now, but the other setup.sh files should be updated.

The ancient version of gdb (7.6.1) on the cobalts and NPX isn't able to read the debug info in the py3-v4.3 preventing debugging. This is at least partially because gcc 11 switched the default DWARF version from 4 to 5, which is only supported in gdb 8 and later.

In my case, recompiling the python extension with -gdwarf-4 did NOT fix the issues. Including a gdb version >=8 alongside gcc 13 should work.

Pick one of these and change the other, changing the configs along the way.

Saw this on Centos 7:

Linking CXX executable ../bin/ipdf-tutorial
/usr/bin/ld: warning: libquadmath.so.0, needed by /cvmfs/icecube.opensciencegrid.org/py2-v2/RHEL_7_x86_64/tools/gfortran/libgfortran.so.3, not found (try using -rpath or -rpath-link)
/cvmfs/icecube.opensciencegrid.org/py2-v2/RHEL_7_x86_64/tools/gfortran/libgfortran.so.3: undefined reference to `fmodq@QUADMATH_1.0'

I guess we need libquadmath.so.0 in cvmfs as well? Does this degenerate into needing all of gcc's libraries?

Mind, this only happened when building on a machine that did not have gfortran installed.

(imported from trac: #1456)

Get rid of the last remnants of ports:

• root
• GEANT4
• Genie

Also think about moving to c++11.

Rob asks for wipac-rest-tools to be installed in the next cvmfs python env, to make it easier to run filtering (to talk to iceprod).

Add untarred copies of pyglidein tarballs.

Probably some similar organization as iceprod.

math.wisc.edu is running Ubuntu 15.10 on their nodes. Check if we can run off the 14.04 version, or if we need to build another version for this.

For py2-v2 and higher, ignore any environment overrides. While this offers less flexibility, it saves us from the DESY problem, where OS_ARCH is defined as something else by default.

from http://code.icecube.wisc.edu/projects/icecube/ticket/1862:

icetray-start requests to be executed using plain sh, but uses syntax which only exists in bash to define functions. This prevents it from running on systems (Ubuntu, FreeBSD) where /bin/sh is either really the Bourne shell or something closer to being conformant with the Bourne shell specification:

/cvmfs/icecube.opensciencegrid.org/py2-v2/icetray-start: 52: /cvmfs/icecube.opensciencegrid.org/py2-v2/icetray-start: Syntax error: "(" unexpected

It appears that the correct solution is to remove all uses of the non-existent function keyword.

Pass2 is using zstd.
It would be useful to have the CLI tools for this format available, as it it the case for gz, bz2 and xz.

Might need portland compiler set up on build machines.

Globus fails to build currently:

configure: error: Missing required header ltdl.h

Probably need to wait for the new cvmfs build system.

There have been several instances of bad publishes in the last few days (python modules mysteriously disappearing).

Develop a test suite to run basic checks and catch these sorts of problems.

pip 8.0 gives us the --prefix option, which is probably the exact thing we want for iceprod building. let's upgrade to that for py2-v2.

/cvmfs/icecube.opensciencegrid.org/py2-v1/icetray-start: line 41: /etc/osg/wn-client/setup.sh: No such file or directory

Probably not checking first if $OSG_GRID/setup.sh is a valid file.

Photospline 2.0 can peacefully coëxist with photospline the I3 project, and having it in py2-v3 would significantly ease the transition.

https://github.com/cnweaver/photospline/releases/tag/v2.0.0

Add nuSQuIDS and SQuIDS.

Probably for py2_v3 or py2_v4.

https://github.com/icecube/icetray/issues/3050

(imported from trac: #1690)

Have Ubuntu 16.04 in cvmfs

the Geant4 version currently installed in cvmfs shows strong signs of flaws. Including a more recent version (in addition to old one) was suggested.

nlopt is a new minimizer that should be getting added to our supported list soon. This is designed to replace minuit2.

Now that I3Ports is going away with #2, we'll only have a single PYTHONPATH entry to make, the default one. Let's see if python can figure this out internally instead of us having to force it.

Package : hdf5
CVE ID : CVE-2016-4330 CVE-2016-4331 CVE-2016-4332 CVE-2016-4333
Debian Bug : 845301

Cisco Talos discovered that hdf5, a file format and library for
storing scientific data, contained several vulnerabilities that could
lead to arbitrary code execution when handling untrusted data.

The followup machines are apparently missing freetype completely (headers + library). This indicates that maybe we should install our own freetype.

This is needed for the new GCD generation.

Probably add this to py2-v2 as well.

New GCD files:

/data/sim/sim-new/downloads/GCD/GeoCalibDetectorStatus_2014.56784_V0_NovSnow.i3.gz
/data/sim/sim-new/downloads/GCD/GeoCalibDetectorStatus_2015.57161_V0_OctSnow.i3.gz
/data/sim/sim-new/downloads/GCD/GeoCalibDetectorStatus_2016.57531_V0_OctSnow.i3.gz

Should think of making the GCD file sync automated one of these days.

IceProd2 needs uberftp installed. This can happen in either py2-v2 or iceprod variants.

We should encourage usage of the HTCondor python bindings. In order to do that, install them for the python in cvmfs.

Can do this for py2-v3 and future.

We should make some ARM builds for CVMFS.