Does Arena support tensorflow's tf.contrib.distribute.CollectiveAllReduceStrategy with tf.estimator?

[Juntai]

I can run ps mode with tf.estimator normally. But when I try some synchronous training strategy without ps, I found only chief is working, other workers is waiting for variables initialization.
So, my questions are:
1.Does Arena support tensorflow's tf.contrib.distribute.CollectiveAllReduceStrategy with tf.estimator?
2.If "yes", any example?
3.If "no", is this on arena's plan list? Or, why not?

Thanks.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.