I have two almost identical Haswell machines, both running Ubuntu 16.04. One one, I am running QC with no problems:
http://www.gpugrid.net/results.php?hostid=357332

On the other, I am now getting Python errors:
http://www.gpugrid.net/results.php?hostid=334241&offset=0&show_names=0&state=0&appid=30

The errors started a couple of days ago, when I started to run Folding on the GPU, which also uses python. That seems to have triggered it, though it may have been something else. Also, I have just performed an update of everything on Ubuntu on this machine, which included a python update of some sort. It didn't fix it, but may have sealed the nail in the coffin. (I have stopped running Folding on this machine, but the damage may have already been done, whatever it is.)

In principle, our WUs don't use the system's (or other projects') Python, but unfortunately it's not that simple because there are numerous system libraries which are involved.

In the meantime, try resetting the project: this deletes all traces of "our" python environment and downloads it again.

Important questions (for me to understand):

1. Has the "working" machine also been updated?
2. Did you follow those "advanced" FAH install instructions?

Important questions (for me to understand):

1. Has the "working" machine also been updated?
2. Did you follow those "advanced" FAH install instructions?

(1) No, I have left the "working" machine alone. But now that you mention it, the working machine is also running Folding on a GPU, so it is not necessarily bad in all cases, though it seems to be the culprit for the other machine.
(2) No, I did not use the advanced instructions. I just right-clicked on the FAH Client and Control setup files, and used the standard Ubuntu installer.

Also, I typed "python" in a terminal window on both machines, and got the same response:
i7-4790 (problem machine):

Python 2.7.12 (default, Dec  4 2017, 14:50:18) 
[GCC 5.4.0 20160609] on linux2

i7-4770 (working machine):

Python 2.7.12 (default, Dec  4 2017, 14:50:18) 
[GCC 5.4.0 20160609] on linux2

So they seem to be updated identically insofar as that is concerned, but as you point out there must be other factors.

I will try your suggestion when I get a chance. Thanks. I would like to put both machines on QC.

In the meantime, try resetting the project: this deletes all traces of "our" python environment and downloads it again.

Thanks. That fixed it.
http://www.gpugrid.net/workunit.php?wuid=13743106

I expect it had something to do with the order in which I installed Folding on the two machines. I may have upgraded one from the previous version (7.4.16) to the new version (7.5.1) over the top, while on the other I may have removed the previous version first before installing the new version. But I don't remember which was which. It is tricky.

I get python 2.7.13 on both Linux boxes, where on the laptop with no GPU board the
QC 3.21 tasks execute and validate, while they fail on the SUN WS which executes GPUGRID GPU tasks.
Tullio

Thanks. That fixed it.

Thanks. I think it's a more or less random occurrence. The "internal" python installation is left in an inconsistent state by post-start upgrades. Some future update should fix the cause. In the meantime, resetting should be the cure-all.

I get python 2.7.13 on both Linux boxes, where on the laptop with no GPU board the
QC 3.21 tasks execute and validate, while they fail on the SUN WS which executes GPUGRID GPU tasks.
Tullio

It's the location of your installation (deeply nested under BOINC/BOINC/BOINC, I guess by mistake). Try another location. Python version is irrelevant.

BOINC on boinc.berkeley.edu site for Linux is still 7.2.42. I am using BOINC 7.8.3 dpwnloaded from a SETI user, this is why of its strange location. But it works on GPUGRID GPU tasks and on LHC@home Atlas tasks, which need VirtualBox, installed.Next week I shall have new version of SixTrack CPU tasks from CERN.
Tullio

---

BOINC on boinc.berkeley.edu site for Linux is still 7.2.42. I am using BOINC 7.8.3 dpwnloaded from a SETI user, this is why of its strange location. But it works on GPUGRID GPU tasks and on LHC@home Atlas tasks, which need VirtualBox, installed.Next week I shall have new version of SixTrack CPU tasks from CERN.
Tullio

That's odd that its only 7.2.42 at BOINC yet I've downloaded 7.9 from the Ubuntu repository and that's a development version.

It does say to use the package manager:
"These versions are not current. We recommend that instead you install BOINC using a package manager, run it using Docker, or use a bootable ISO image."

Shrug.

SuSE is not following much BOINC, so I have to find newer versions where I can. But 7.8.3 is working well on my two Linux boxes, all with SuSE Leap 42.3 64 bit.
Tullio

The repository versions should all be updated to the recommended new release v7.10.2 by now.

These versions all run in a secure sandbox and support all BOINC features.

Berkeley provided a stand-alone, less secure, version for use on Linux distributions with no repository support for BOINC: this is the version which is stuck at v7.2.42, and is unlikely to be further developed. This version runs in user space.

Some users are prepared to trade off the added security of the repository versions for the extra flexibility of running in user space, and users are welcome to download to download the source code and compile user-mode binaries for their own use. But these private builds are not supported by the BOINC writing community, and you'll need to make your own support arrangements if you need them. There are communities of interest all over the internet - not always conducted in English - where this sort of mutual support is available.

The repository versions should all be updated to the recommended new release v7.10.2 by now.

As of a day or two ago, I did not have any luck trying to update from the Ubuntu repository. I had to go to the LocutusOfBorg repository.

sudo add-apt-repository ppa:costamagnagianfranco/boinc
sudo apt-get update  => Then install BOINC from Ubuntu Software