How can I squash the stuck thread warning from XLA?

[C-J-Cundy]

I'm doing a task where I'm pmapping an operation that takes a variable amount of time over different devices.

This means that quite often, one device will finish much faster than the others. When this happens, I get this error message:

022-11-17 16:25:04.454825: E external/org_tensorflow/tensorflow/compiler/xla/service/rendezvous.cc:31] This thread has been waiting for 10 seconds and may be stuck:

2022-11-17 16:25:15.673476: E external/org_tensorflow/tensorflow/compiler/xla/service/rendezvous.cc:36] Thread is unstuck! Warning above was a false-positive. Perhaps the timeout is too short.

Is there a way to increase the timeout to stop this error message being printed, perhaps with an XLA flag?
It really clutters up my stdout when training.
I would prefer to avoid a solution that involves suppressing all error messages, as they can be very useful.

[bjuncek]

ping on this?