question

Upvotes
Accepted
3 1 3 4

New RD library for python is leaking memory when opening/closing streams

Hey

We are trying to use the new Refinitiv Data Library for Python, but we are experiencing memory leaks when opening/closing pricing streams.


1682936446719.png


Can you please investigate and fix?


Over 20 iterations of the above stream opening and closing, the memory increases as below:

Loop #1

Memory before open stream = 103.12109375

Memory after open stream = 103.1953125

Memory after get_snapshot = 103.1953125

Memory after close stream = 103.1953125

End loop #1

[...]

Loop #19

Memory before open stream = 103.5234375

Memory after open stream = 103.5234375

Memory after get_snapshot = 103.5234375

Memory after close stream = 103.5234375

End loop #19


I am attaching a small python file where this can be tested and debugged.testRDP.txt


Best regards

Sabina

python#technologystreaming-pricesmemory-leaks
1682936446719.png (54.3 KiB)
testrdp.txt (1.5 KiB)
icon clock
10 |1500

Up to 2 attachments (including images) can be used with a maximum of 512.0 KiB each and 1.0 MiB total.

Upvotes
Accepted
4.2k 11 2 7

Hi @Sabina Tanase ,


I have just learned from the Dev team, that the issue is now fixed with version 1.4.0. Please see below the before and after memory graphs:

screenshot-2023-10-13-at-093553.png

screenshot-2023-10-13-at-093516.png

The current fix includes the Optimisation in working with listeners of streams, optimisation library imports will be delivered in 1.5.0 and in the future releases the optimisation of internal callback mechanisms will be enhanced as well.


Best regards,

Haykaz


icon clock
10 |1500

Up to 2 attachments (including images) can be used with a maximum of 512.0 KiB each and 1.0 MiB total.

Upvotes
4.2k 11 2 7

Hi @Sabina Tanase ,


My colleague @nick.zincone raised this to the Dev team. The issue is under investigation, will update accordingly.


Best regards,

Haykaz


icon clock
10 |1500

Up to 2 attachments (including images) can be used with a maximum of 512.0 KiB each and 1.0 MiB total.

Hey Haykaz,

Any updates on this?

Best regards

Sabina.

Hi Sabina, no updates unfortunately, followed up with the dev team.
Upvotes
3 1 3 4

Hey Haykaz, Nick


Thank you - looking fwd to your answer.


Best regards

Sabina.

icon clock
10 |1500

Up to 2 attachments (including images) can be used with a maximum of 512.0 KiB each and 1.0 MiB total.

Upvotes
3 1 3 4

Hey

Any updates?

Best regards

Sabina.

icon clock
10 |1500

Up to 2 attachments (including images) can be used with a maximum of 512.0 KiB each and 1.0 MiB total.

Upvotes
4.2k 11 2 7

Hi @Sabina Tanase ,


The dev team come back to me with a following observation:

-----

If we will have current particular case, yes, we will see some memory consumption growing on 20 iterations.


If we will spread use case to get 500 or 1000 iterations I see that memory consumption starts from 138.42Mb has some peak at 139.04Mb, then it falls to 123.03Mb, then peak at 123.85Mb, falls to 108Mb. To the end of 1000 iteration it fell down to 64.95Mb. We can assume that garbage collector works after some conditions came and process release memory. Tested with python3.10 on MacOS Ventura 13.4, refinitiv-data 1.3.0


Our library is a really thin layer, from the current case in example, we just storing the last updates callback in stream, that’s it.


If the customer have practical, living streaming use case with resulting in growing memory consumption, please, share it.

-----

please let me know what you think and if you can provide a practical use case where the team can investigate further?


Best regards,

Haykaz

icon clock
10 |1500

Up to 2 attachments (including images) can be used with a maximum of 512.0 KiB each and 1.0 MiB total.

Upvotes
3 1 3 4

Hey Haykaz ( @h.aramyan01 )

Thank you for the reply.

I have increased the loop to 5000 iterations, and tested with forcing the garbage collector to tend to memory (outside and inside the loop). The result is unfortuantely the same for us - memory is increasing.

I have attached the program I have been using (testrdp.txt), which has run on Windows10 using python lib "refinitiv-data==1.3.0".

This is how the memory consumtion is looking for 5000 iterations using testrdp.py:mem.png

I would have assumed that stream.close() would release the memory for me, as well as release the handles used by the respective stream.

Please let me know if the developers are testing using my testrdp.py or can you please share their code?

Best regards

Sabina.


mem.png (40.9 KiB)
icon clock
10 |1500

Up to 2 attachments (including images) can be used with a maximum of 512.0 KiB each and 1.0 MiB total.

Hello @Sabina Tanase ,


Thank you very much for your patience, I have just got a follow up from the Dev team after I sent your notes above. The team told that there are in process of making some adjustments for streams. Currently, the change is under review and testing. They already have some intermediate results of improvements, but will release lately when task will be done completely.

I will inform you separately to upgrade the lib for the change.


Best regards,

Haykaz

Upvotes
3 1 3 4

Hey Haykaz

Super nice!

We will start testing in the next 2 weeks, and we will let you know how it look on our side.

Thank you for the efforts,

Sabina.

icon clock
10 |1500

Up to 2 attachments (including images) can be used with a maximum of 512.0 KiB each and 1.0 MiB total.

Sounds good, looking forward to the update from your side
Hello @Sabina Tanase

Do you still encounter the issue?

Hello @Sabina Tanase

Does the issue still exist?

Write an Answer

Hint: Notify or tag a user in this post by typing @username.

Up to 2 attachments (including images) can be used with a maximum of 512.0 KiB each and 1.0 MiB total.