Re: TensorFlow 2.13 and 2.14 Support Inquiry regarding S3 IO bugs after v0.33.0

[Yong Tang]

Hi Sai,

Thanks for reaching out. As was discussed in our meeting today, we will try to see if a fix/solution can be done soon. In the meantime, if only the file system support is needed then pinning to an earlier version can be an alternative for now, as file system APIs are based on C and are very much compatible.

Thanks

Yong

---

From: Miduthuri, Sai Parthasarathy <sai...@amazon.com>
Sent: Tuesday, February 6, 2024 5:24 PM
To: Yong Tang <yong.tan...@outlook.com>; i...@tensorflow.org <i...@tensorflow.org>
Cc: Katz, Ohad <ohad...@amazon.com>; Chumbalkar, Tejas <chum...@amazon.com>; Sampath, Kirubakaran <kiru...@amazon.com>
Subject: TensorFlow 2.13 and 2.14 Support Inquiry regarding S3 IO bugs after v0.33.0

 

Hello Yong,

 

I am a member of the AWS Deep Learning Frameworks team. Thank you for the delivery of tensorflow-io 0.36.0 this week.

 

We have noticed that our integration tests that use tensorflow-io have started to fail when using tensorflow-io>=0.34.0 on TensorFlow 2.13 and 2.14 docker images. Downgrading to 0.33.0 seemingly prevents the error, but we do not want to go against the compatibility table as defined by tensorflow-io when delivering DLC images.

 

There is a GitHub issue that was opened earlier in January for the same error - https://github.com/tensorflow/io/issues/1912 - which doesn’t seem to have received further engagement. I’m reaching out in advance of the monthly meeting this week to provide some additional context on this issue before bringing it up on the meeting, in case you have the opportunity to gather more details about it.

 

Thank you for the support.

 

Best regards,

Sai Parthasarathy Miduthuri

Software Development Engineer II

Deep Learning Frameworks