Problem while installation

[anushkai]

I am getting following error while installtion. Any one have a clue?

Traceback (most recent call last):
File "/home/anb5km/.conda/envs/prime/bin/medperf", line 33, in
sys.exit(load_entry_point('medperf==0.1.0', 'console_scripts', 'medperf')())
File "/home/anb5km/.conda/envs/prime/bin/medperf", line 25, in importlib_load_entry_point
return next(matches).load()
File "/home/anb5km/.conda/envs/prime/lib/python3.9/importlib/metadata.py", line 86, in load
module = import_module(match.group('module'))
File "/home/anb5km/.conda/envs/prime/lib/python3.9/importlib/init.py", line 127, in import_module
return _bootstrap._gcd_import(name[level:], package, level)
File "", line 1030, in _gcd_import
File "", line 1007, in _find_and_load
File "", line 986, in _find_and_load_unlocked
File "", line 680, in _load_unlocked
File "", line 850, in exec_module
File "", line 228, in _call_with_frames_removed
File "/home/anb5km/.conda/envs/prime/lib/python3.9/site-packages/medperf/main.py", line 9, in
from medperf.ui.factory import UIFactory
ModuleNotFoundError: No module named 'medperf.ui'

[AlexanderGetka-cbica]

I'm getting the same when trying to verify installation with --version.

[hasan7n]

Thanks @AnushkaIdamekorala @AlexanderGetka-cbica for reporting this!
We will fix these issues soon.

[hasan7n]

@AnushkaIdamekorala @AlexanderGetka-cbica The problem should now be fixed by #439 #440.

Please follow the updated installation instructions again from scratch and let us know if the problem persists!

[anushkai]

@AnushkaIdamekorala @AlexanderGetka-cbica The problem should now be fixed by #439 #440.

Please follow the updated installation instructions again from scratch and let us know if the problem persists!

@hasan7n When executing
python seed.py --cert cert.crt --demo model
I am getting
HTTPSConnectionPool(host='127.0.0.1', port=8000): Max retries exceeded with url: /version (Caused by NewConnectionError('<urllib3.connection.HTTPSConnection object at 0x2b7c434eb0d0>: Failed to establish a new connection: [Errno 111] Connection refused'))
Error seems the problem is with medperf /version endpoint.

[hasan7n]

@AnushkaIdamekorala This error usually happens when the local server is not running. Was it still running when you received this error?

[hasan7n]

Getting a new error, now:

Traceback (most recent call last): File "F:\work\env_medperf\Scripts\medperf-script.py", line 33, in sys.exit(load_entry_point('medperf', 'console_scripts', 'medperf')()) File "F:\work\env_medperf\Scripts\medperf-script.py", line 25, in importlib_load_entry_point return next(matches).load() File "F:\Python39\lib\importlib\metadata.py", line 77, in load module = import_module(match.group('module')) File "F:\Python39\lib\importlib__init__.py", line 127, in import_module return _bootstrap._gcd_import(name[level:], package, level) File "", line 1030, in _gcd_import File "", line 1007, in find_and_load File "", line 986, in find_and_load_unlocked File "", line 680, in load_unlocked File "", line 850, in exec_module File "", line 228, in call_with_frames_removed File "f:\work\medperf\cli\medperf__main.py", line 10, in from medperf.decorators import clean_except, configurable File "f:\work\medperf\cli\medperf\decorators.py", line 7, in from medperf.utils import ( File "f:\work\medperf\cli\medperf\utils.py", line 17, in from pexpect import spawn ImportError: cannot import name 'spawn' from 'pexpect' (F:\work\env_medperf\lib\site-packages\pexpect__init.py)

@AlexanderGetka-cbica This error should be because you are using Windows if I am correct. pexpect is not supported on Windows. We are working on this here: #417

[AlexanderGetka-cbica]

Ah, got it, thanks!

[anushkai]

@hasan7n Yes, For the clarification you are refering to the server starts by
sh setup-dev-server.sh right?

[hasan7n]

@AnushkaIdamekorala Unfortunately I am not able to reproduce the error.

Can you do the following:

while you have your local server running:

1. Navigate to this URL: https://127.0.0.1:8000/ on your browser and show me what are you getting
2. Show me a screenshot of the terminal where you ran the command sh setup-dev-server.sh

[anushkai]

Unfortunately I am running this on a cluster therefore I am unable to provide with GUI.

[anushkai]

@hasan7n For
curl -k https://localhost:8000/version
Server is running and got {"version":"v0"} as the response.

[hasan7n]

Thanks for trying that! Let me know then if seeding the server as follows solves the problem:
python seed.py --cert cert.crt --demo model --server https://localhost:8000

[anushkai]

@hasan7n Thank you for the support, above step worked. I proceed along the document and downloaded the files + Logged in successfully. But when executing

medperf test run \
   --benchmark 1 \
   --model "medperf_tutorial/xrv_densenet/mlcube/mlcube.yaml"

I get following output.

MedPerf 0.1.0
Retrieving Model cube
> Model cube download complete
Checking cube MD5 hash...
> 2590a2e0aaab0f59eae251eb939e8a0668f86177 MD5 hash check complete
Retrieving Evaluator cube
> Evaluator cube download complete
Checking cube MD5 hash...
> xrv_metrics MD5 hash check complete
> Preparation cube download complete
> xrv_prep MD5 hash check complete
❌ There was an error while executing the cube.

As there isn't much of a details provided I am unable to troubleshoot that. I am glad if you can look into that as well.

[aristizabal95]

Hi @AnushkaIdamekorala, You can get additional details by looking at the logs, at ~/.medperf/localhost_8000/logs/medperf.log. If you want more verbose details, then retry running the benchmark with the following command:

medperf --loglevel=debug test run \
   --benchmark 1 \
   --model "medperf_tutorial/xrv_densenet/mlcube/mlcube.yaml"

Let us know if that helps. Also feel free to share the traceback from the logs if you feel the need to!

[aristizabal95]

@AnushkaIdamekorala just in case this helps, an usual point of failure for MLCubes is not having Docker running. Can you ensure that's not your case? Try running docker run hello-world to see if it runs.

[anushkai]

@aristizabal95 Thanks for the input. I found that there is a version error in the using singularity and trying to upgrade in the cluster. I will come back if neither did work.

[aristizabal95]

Hi @AnushkaIdamekorala, were you able to fix your issues with singularity? Let us know if you need more help from our side

[anushkai]

hi @aristizabal95 thank you for asking there is a ongoing cluster maintenance therefore had to delay the work. However it will be done by this week. I will reach out when it is done. Thanks again.

[anushkai]

Hi @aristizabal95 , My cloud team informed that they are unable to support with the singularity issues(They are unable to implement the user and group ID remappings) and I have to continue with my local machine. In the process of reinitiating in local machine following issue occured which I haven't met with previously when executing
sh setup-dev-server.sh
I am getting
`setup-dev-server.sh: line 14: realpath: command not found
setup-dev-server.sh: line 15: realpath: command not found

1
1

0
CERT FILE must not be empty` Should I set an additional ENV_VARIABLE?

[aristizabal95]

Hi @anushkai, sorry to hear the singularity installation has been giving you trouble! As for the error, I believe the CERT FILE message is a consequence of the realpath: command not found error message. I believe you have to install that command on your machine before you continue. It should only require some apt call or similar depending on your OS.

@hasan7n I've noticed this happening to me as well, realpath is not necessarily present as a standard binary, so we should look into either providing instructions for installing it, or replacing this command with a standard one

[laszewsk]

Please see #462

There are multiple issues.

• we can not use fakeroot

• we do not have sudo

• we need a way to specify the singularity build image command as it is locked down (see ticket)

• docker is not available only singularity

• it may be good to use a bind mechanism for the dir as they could be located on a file system that may not be mounted in singularity by default. Maybe the codedir should be mounted as --bind $CWD:/medperf and then in the scripts /medperf is used. On other systems the CWD is automatically mounted, but if the system uses /localscratch that is usually not the case as that is only visible to the node itself. So a more general solution may be needed (lower priority as for now we can run on a machine that has CWD mounted)

• where is realpath used? in the image or on machine. typically it is available on the HPC

• we need a slurm script that starts medperf, we should assume we can run without GUI

[laszewsk]

realpath is available on all machines where we run the script.

best woudl actually be a SLURM script that sets the test up in batch mode.

[hasan7n]

Closing this for inactivity and as being tracked by #457. Feel free to re-open if you have any questions.