2024-09-18T14:12:13.919Z

<avasiliev> Hi, I am trying to find the root cause and fix a problem in radosgw. I need some help to challenge what I have already found and determine the best approach. For now, I have a somewhat reliable way to reproduce this problem and a probable solution.

I have included everything I have in the ticket.
<https://tracker.ceph.com/issues/63935#note-6>

2024-09-18T15:25:51.876Z

<Casey Bodley> weekly rgw meeting starting shortly in [ <https://pad.ceph.com/p/rgw-weekly](https://meet.google.com/mmj-uzzv-qce> ) (empty agenda so far)

2024-09-18T16:10:18.093Z

<Samaksh Dhingra> Hi team, I was trying to setup ceph on macbook m3 pro using Fedora 39 aarch64 VM on UTM. However, while executing dependency installation, I encountered error when grpcio and grpcio-tools were building. I tried a lot to find the cause on internet - trying out with python3.10 in a venv, etc. but no help. Anyone who faced similar issue, please let me know the fix that worked?

Previous to this I tried Ubuntu 22.04 aarch64 VM, there dependency intallations worked fine, build also was successful with some errors in between though. However, when I ran unit tests in order to check if everything was setup properly or not, many test case gave "(did not run)" message due to some missing files. Hence, I tried with fedora as I heard many people are using that.

Thanks.

2024-09-18T21:10:57.697Z

<Matthews Jose> Thank you very much for your reply it is very helpful

2024-09-18T21:17:29.301Z

<Matthews Jose> I have a follow up question if there is a write and a read happening simultaneously and if the client randomly sends the read request to a replica (because the flag is enabled) doesn't this mean that there is some inconsistency since the write could potentially update the same pg/object that is being read and the mosprep message has not reached the replica yet from the primary

2024-09-18T21:20:05.845Z

<Ilya Dryomov> The replica OSD would return a distinctive error instructing the client to resend the request to the primary OSD

2024-09-18T21:42:28.691Z

<Matthews Jose> This is my doubt how does the replica know in this case there is a pending write operation. Since it has not received the rep message from the primary. 
Could you also say why this is not enabled by default and what kind of scenarios it is recommended to turn this on.