Eureka
Eureka copied to clipboard
kevin218
[Bug]: segmentation fault MC3 package after running light curves
FAQ check
- [X] Yes, I checked the FAQ and my question has not been addressed.
Instrument
Light curve fitting (Stages 4-6)
What happened?
I get a segmentation fault after running light curve fits (same for dynesty and emcee, and not on all light curves) and I traced down to MC3 package.
It's the function from MC3 from mc3.stats import time_avg that causes it, previous Allan plot looks fine so I'm not sure what the issue is that MC3 crashes - is it a Eureka! or a MC3 problem? Either way it's flagged this way.
I created a new conda environment and new installation and it still pops up.
Error traceback output
When running with gdb:
EMCEE RESULTS: rp: 0.09314383900504214 (+0.00023746832716464716, -0.00024390231091873937) t0: 60336.16629344476 (+0.0001366072174278088, -0.00014108874165685847) inc: 86.14041037419365 (+0.03564806637537288, -0.03959720206140105) a: 7.5178380816512576 (+0.012216578350765062, -0.012060384627009313) u2: 0.0596876616903001 (+0.01720078126260373, -0.016858445839115672) c0: 1.0037328901325697 (+3.041406496073762e-05, -2.8784564057193762e-05) c1: -0.00018125230315689576 (+0.00021138404946226995, -0.00022132731635158144) scatter_mult: 4.604809762772209 (+0.1430173800715835, -0.12412840317554785); 489.802873463214 (+15.212425121351847, -13.203248708624503) ppm
Program received signal SIGSEGV, Segmentation fault. 0x00007ffe987d7c1e in invgamma () from /home/ahrer/.conda/envs/eureka-dev/lib/python3.9/site-packages/mc3/lib/_time_averaging.cpython-39-x86_64-linux-gnu.so
What operating system are you using?
Linux
What version of Python are you running?
Python 3.9.7
What Python packages do you have installed?
No response
Code of Conduct
- [X] I agree to follow this project's Code of Conduct
@evamariaa, can you see if Kevin's tweak in PR #609 for the file src/eureka/S5_lightcurve_fitting/plots_s5.py fixes this issue for you?
Ok, just installed from PR https://github.com/kevin218/Eureka/pull/609 and running through Stage 5 now, so far I didn't get a segmentation fault yet so I'm cautiously optimistic and will let you know if it stays like this! Thanks!
Can one or both of you please try the code in PR #609 and confirm that the problem is resolved for you if you use that version of the code? We'll hopefully have this bug patched ASAP
Nothing changed since my previous comment, as in it seems to run fine with the PR https://github.com/kevin218/Eureka/pull/609 you mentioned, no segmentation faults so far. But I haven't run Stage 5 much since then because of maintenance on our server. I should be able to run more by the end of the week and let you know!
I just merged #609 into the Main branch. Try updating your version of Eureka! and let me know if you still have problems.
Reinstalled Eureka! from the main branch and getting an error now when importing Eureka:
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/ahrer/.conda/envs/eureka/lib/python3.9/site-packages/eureka/__init__.py", line 28, in <module>
from . import S5_lightcurve_fitting
File "/home/ahrer/.conda/envs/eureka/lib/python3.9/site-packages/eureka/S5_lightcurve_fitting/__init__.py", line 12, in <module>
from . import fitters
File "/home/ahrer/.conda/envs/eureka/lib/python3.9/site-packages/eureka/S5_lightcurve_fitting/fitters.py", line 21, in <module>
from . import plots_s5 as plots
File "/home/ahrer/.conda/envs/eureka/lib/python3.9/site-packages/eureka/S5_lightcurve_fitting/plots_s5.py", line 5, in <module>
from mc3.stats import time_avg
File "/home/ahrer/.conda/envs/eureka/lib/python3.9/site-packages/mc3/__init__.py", line 4, in <module>
from .sampler_driver import *
File "/home/ahrer/.conda/envs/eureka/lib/python3.9/site-packages/mc3/sampler_driver.py", line 20, in <module>
from .fit_driver import fit
File "/home/ahrer/.conda/envs/eureka/lib/python3.9/site-packages/mc3/fit_driver.py", line 11, in <module>
from . import stats as ms
File "/home/ahrer/.conda/envs/eureka/lib/python3.9/site-packages/mc3/stats/__init__.py", line 5, in <module>
from .stats import *
File "/home/ahrer/.conda/envs/eureka/lib/python3.9/site-packages/mc3/stats/stats.py", line 31, in <module>
import _binarray as ba
ImportError: /usr/local/autofs.home/home/ahrer/.conda/envs/eureka/lib/python3.9/site-packages/mc3/lib/_binarray.cpython-39-x86_64-linux-gnu.so: undefined symbol: __pow_finite
Problem is our cluster was under maintenance for a couple of days and they did lots of updates and there are still some bugs so I'm not sure if it's even a Eureka! issue or not. I will keep you posted.
Did your installation of mc3 succeed? It looks to me like you might've been missing a C compiler or something which meant mc3 didn't finish installing or something
I wish, but no I can't confirm either way. Our cluster here still has issues, particularly with the file and storage system and is completely unusable right now. An external company has been brought in this week to fix it so hopefully I can give you an update sometime next week.
I just updated to the latest version of Eureka! on my Linux machine and I'm getting a similar error.
ImportError: /opt/anaconda3/envs/eureka/lib/python3.9/site-packages/mc3/lib/_binarray.cpython-39-x86_64-linux-gnu.so: undefined symbol: _ZGVbN2v_exp
I can confirm that I have a valid C compiler, so maybe this is a Linux issue.
Interesting, I'm running Ubuntu on Windows Subsystem for Linux which should be sufficiently similar to have encountered such an issue on my end if it were a Linux-based issue. I can try reproducing this with a Docker build though to reduce the number of unknowns.
@taylorbell57 I'm also getting this now, on Ubuntu/WSL, identically to Kevin's error message.
ImportError: /home/jbrande/miniconda3/envs/eureka/lib/python3.9/site-packages/mc3/lib/_binarray.cpython-39-x86_64-linux-gnu.so: undefined symbol: _ZGVbN2v_exp
I pushed a workaround to Main, so updating Eureka! should "fix" the problem until we find a more permanent solution.
@kevin218 and @jbrande, did you get those ImportError messages when importing Eureka! or when it came to finally making the plot? Just need to know where it was happening so I can figure out if/when I've reproduced the issue
The error is on import. If it fails, you'll get a warning message that states MC3 failed to import.
I just created a new conda environment and performed a fresh install using Python 3.10 and had no issues importing MC3. Perhaps MC3 has compatibility issues with Python 3.9.
I know that jwst>=1.14.0 requires Python>=3.10, but I just made a test environment with Python=3.9.7 where I just manually installed mc3 and ipython and I was able to install and import mc3 just fine on Windows Subsystem for Linux. And trying to install the main branch of Eureka! with Python=3.9.7 gives the following error (as it should): ERROR: Package 'eureka' requires a different Python: 3.9.7 not in '>=3.10'
@jbrande and @evamariaa, could you please see if this is still an issue for you when starting from a fresh install of the main branch version (using pip install 'eureka[jwst]@git+https://github.com/kevin218/Eureka.git@main') in a fresh conda environment with Python=3.10? As a result of Kevin's recent emergency patch, I seem not to be getting any corner plots even from environments where mc3 used to work fine, so instead of testing the issue with a Stage 5 run, for now can you just try doing from mc3.stats import time_avg from an ipython or jupyter instance to see if you still get the ImportError issue?
Sorry for the delay on this, this got a little bit lost in my to-do list.
I did what you said Taylor, did a fresh install of Eureka with a new conda environment and python 3.10 and ran from mc3.stats import time_avg and that ran without an error!
Alright, since two of you who previously had issues with this no longer do after creating a new conda environment, I'm going to close this issue as it seems not to be reproducible anymore. If anyone re-encounters this issue, feel free to leave a comment and we can re-open this issue