-
Notifications
You must be signed in to change notification settings - Fork 5.7k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Segfault when importing TensorFlow after Ray. #2159
Comments
Note that the segfault happens if we replace cc @pcmoritz |
Could it be related to the change in symbols.map here: apache/arrow@2093f6e? |
If not, doing a search on pyarrow commits and seeing where it was introduced will be helpful |
Unfortunately when pyarrow is compiled from source the problem disappears, so it has something to do with the way wheels are built. |
Tracked it down a bit further in https://issues.apache.org/jira/browse/ARROW-2657. |
Note that this seems to make it impossible to use actors which have a dependency on tensorflow. Even if the imports are ordered correctly, when ray unpickles the actor class it will import tensorflow automatically, causing the agent to die. Here's a minimal example:
Running
|
@robertnishihara @DaveyBiggers |
Compiling from source should work. Also importing tensorflow before ray should work (though that isn't possible on ray workers). Or installing Ray from a commit from before 225608e should work as well. |
System information
Python 3.6.1 :: Continuum Analytics, Inc.
TensorFlow version 1.8.0.
Searching over commits, the issue was introduced in 225608e.
gdb shows
The text was updated successfully, but these errors were encountered: