Edward Oakes
Edward Oakes
@iycheng I see, so to be clear what we would expect to happen on master failure is the following: - GCS crashes because the master is dead. - Head node...
@iycheng sounds good. From a somewhat naiive perspective option (2) seems like the right move assuming that redis++ is mature/battle tested. Could you help me understand why it requires us...
@GeneDer no problem. LMK if you want to pair on it.
Please also include an automated test here. You can add the serve config you used (or similar) to `test_config_files` and then write a test that verifies it works as expected...
@zcin this test fell through the cracks, it's been failing consistently for over a month. Could you please take a look and fix it / determine if it should be...
Simplest version is to just avoid state changing to `UPDATING` on autoscaling changes
@Yard1 would you mind adding a short extension to https://docs.ray.io/en/latest/serve/tutorials/batch.html showing how to do this?
As discussed offline, let's go with "magic" instance variable names for now: - `__ray_serve_max_batch_size` - `__ray_serve_batch_wait_timeout_s`