SRD build issue: Python installation fails #1493
Labels
bug
Problem when deploying a Data Safe Haven.
Milestone
Comments
10 tasks
|
This section of the SRD image building instructions says That wasn't done here, so we'll need to build a new image. Please can you do this and check the build logs before converting the candidate VM into an image? If there's a problem in the image building scripts we can debug that separately. |
I've transformers is on our tier3 allowlist - but I couldn't see a GitHub issue 51450 on the transformers GH if that's what this means? |
|
but I guess you're suggesting I create another candidate VM and consult the |
|
Yes, that's what I'm suggesting. That contains the full error log and you should be able to see where the error is. As you've already created and deployed a broken image, you can also look at |
|
Seems like the error is related to
|
|
Can you put a longer excerpt from the output log (or a link to the full log) here? |
Result of `tail -f -n+1 /var/log/cloud-init-output.log` on candidate VM |
|
^ Looks like it may be related to transformers after all: |
|
It's not that. It's the exception that's thrown that the top of your log file extract Error message: OSError: libcufft.so.10: cannot open shared object file: No such file or directoryThis is related to |
|
Ok, so remove torch from here: https://github.com/alan-turing-institute/data-safe-haven/blob/develop/deployment/secure_research_desktop/packages/packages-python.yaml Is torchvision totally separate or should that be dropped too? |
|
Yes, you'll need to drop both |
5 tasks
Building the candidate VM again |
|
^ this issue suggests we should downgrade poetry:
In our poetry: # also used by installation scripts
"all": [">1.0.0"] # increase solver flexibility |
|
Newest error - looks like we need to set the version of Poetry dependency issue |
|
You could try |
|
Further version/dependency problems Poetry dependency issue |
|
OK, so if poetry and pystan are incompatible then feel free to drop pystan from any of the Python versions where it isn't working. This is only the pre-installed list of packages - users can still install their own packages in anything less than a tier 4. For example |
|
That seems to have solved all Python issues, but there seems to be a problem with "Azure Data Studio" - not sure whether this is something we need in a SRD? Azure Data Studio error |
|
Can you see the first error in that log extract? What does it say? What do you think the solution might be? |
|
Side question: Why do we support 3 different Ubuntu versions for the SRD? |
|
It's a good idea but it probably won't work. What's happening here is that during the The best solution is just to rerun the deployment script and hope that the connection error was temporary (it usually is). |
We started with |
|
after removing pystan from all Python versions, I'm now back to the original 'Python2Supports' error: latest errorOh this is weird, it's trying to install |
|
I'm going to try removing |
|
Latest error: Python2Suppports virtualenv error |
5 tasks
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
✅ Checklist
💻 System information
🌵 Powershell module versions
🚫 Describe the problem
When building an SRE from the
release-v4.0.4branch, the smoke tests fail on all Python versions - inspecting the SRD build log reveals there was a problem with installing Python (see logs below).🌳 Log messages
Failed Python 3.10 installation on candidate VM:
Python safety check failure during SRD build
Failed smoke tests on SRE
♻️ To reproduce
The text was updated successfully, but these errors were encountered: