-
Notifications
You must be signed in to change notification settings - Fork 168
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
SIGSEGV while compressing on 1.5.0-4 version #327
Comments
No idea, I don't think this is mixing up libzstd and libzstd-jni. Can you try replacing Spark's Zstd-jni with more recent? It is API/ABI compatible. |
I was under the impression that it jumps around, considering that addresses for
and code stack goes
Similar core dump happens for |
Hmm, that's interesting. I don't think https://github.com/facebook/zstd will be able to help if that's the case. We will need some linker magic to fix that. What happens if you match the zstd and zstd-jni versions? |
Just pushed 0c2051b - it may help here. And you should not have problems in the other direction as libzstd-jni does not export all symbols as global. |
I will push it to maven later this week. |
It's quite complex to reproduce the fault, the probability of the task failing is too low and I'm not sure which conditions lead to it to create reproducible test. I've tried simple compression/decompression with both libraries loaded with no success. However, we would most likely align zstd-jni version with libzstd version on Ubuntu Noble (1.5.5) and reach back to confirm if it helped with the issue. |
Hi, sorry it too me so long. I have just pushed 1.5.6-7 to Maven |
👋 @luben!
We've been occasionally seeing segmentation faults while compressing parquet files using zstd-jni. In our environment we load
zstd-jni
version1.5.0-4
which comes from Spark'sZStdCompressionCodec
and loadzstd
of version1.4.4
which comes fromZStandardCodec
in Hadoop.Segfault happens very rarely and usually Spark task retry helps to resolve the situation, but I'm still pretty curious on why this happens.
It's also interesting that execution jumps from
zstd-jni
tozstd
library:Do you know if this is expected if we have different versions of zstd loaded? Should we keep them aligned?
The text was updated successfully, but these errors were encountered: