You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I maintain a library for molecular dynamics trajectory analysis, freud (https://bitbucket.org/glotzer/freud). It's heavily written in Cython and C++. We currently use CMake and Cython to build the package, and frequently run out of memory/time (it fails about 75% of the time). On RTD, we have been building with clang to reduce memory usage (gcc fails every time).
We are trying to make the package easier to maintain, by building more like a "normal" Cython package (using setup.py). Our package has been split from one huge module to several smaller modules, which should also help with memory. We had hoped that this would also make the installation behave better on ReadTheDocs. However, we're still hitting the limits and would like to explore increasing time/memory as has been recently mentioned by other projects (#4432#4251#4403). The documentation itself is pretty simple, but we can't render Cython docstrings without building the package. As other packages have mentioned, the installation of conda dependencies takes a lot of time.
Our latest and stable builds use the CMake build system, and the branch pure_cython_rtd is what we're using to test ReadTheDocs support on the new Cython-only build system.
ReadTheDocs is a fantastic service. I really appreciate your time and support, and recognize that asking for more resources is a significant burden. However, it would be nice if we could get the docs to build once, correctly, instead of manually re-triggering builds several times until they work.
Actual Result
Builds fail frequently due to running out of memory/time.
The text was updated successfully, but these errors were encountered:
Hi, just wanted to give an update -- since the shift to Azure a week ago, it seems like our docs have been building successfully. I'm not sure what changed under the hood of ReadTheDocs, but this issue may be resolved. I'll update this issue again if we continue to have challenges. Thanks!
Thank you @bdice for reporting this back! I really appreciate.
I will close this issue then, but please, let me know if you experience this again in the future.
We are working on different solutions. You can read more about this at #4573.
I'm not sure what changed under the hood of ReadTheDocs, but this issue may be resolved
Our new servers on Azure are bigger (more CPU and more RAM) which gives us less building time for projects and seems the extra RAM gives us more stability to allow builds that consume more resources in parallel.
Details
Expected Result
I maintain a library for molecular dynamics trajectory analysis, freud (https://bitbucket.org/glotzer/freud). It's heavily written in Cython and C++. We currently use CMake and Cython to build the package, and frequently run out of memory/time (it fails about 75% of the time). On RTD, we have been building with
clang
to reduce memory usage (gcc
fails every time).We are trying to make the package easier to maintain, by building more like a "normal" Cython package (using setup.py). Our package has been split from one huge module to several smaller modules, which should also help with memory. We had hoped that this would also make the installation behave better on ReadTheDocs. However, we're still hitting the limits and would like to explore increasing time/memory as has been recently mentioned by other projects (#4432 #4251 #4403). The documentation itself is pretty simple, but we can't render Cython docstrings without building the package. As other packages have mentioned, the installation of conda dependencies takes a lot of time.
Our
latest
andstable
builds use the CMake build system, and the branchpure_cython_rtd
is what we're using to test ReadTheDocs support on the new Cython-only build system.ReadTheDocs is a fantastic service. I really appreciate your time and support, and recognize that asking for more resources is a significant burden. However, it would be nice if we could get the docs to build once, correctly, instead of manually re-triggering builds several times until they work.
Actual Result
Builds fail frequently due to running out of memory/time.
The text was updated successfully, but these errors were encountered: