So looking at that output, I have these observations:
* we are dealing with a fairly tight window; in the successful run, the timestamp of the symlink is 06:34:12.981040800 and we run the `ls` command to show us that at 06:34:12,991; that's about 10ms
* based on the timestamp of the next log message in each case, `udevadm settle` on success appears to take substantially longer (06:34:12,907 -> 06:34:12,991; 84ms) than on failure (06:25:43,624 -> 06:25:43,633; 9ms)
* on success, the timestamp of the symlink is between the `udevadm settle` and the next log message, which suggests that it was created as a result of the settle
So none of this points to something _obvious_ (or, at least, not something obvious to me). Given the disparity in the apparent `udevadm settle` execution times, I wonder if the udev events from the resize haven't made their way in to the queue before we execute the settle, meaning it just exits without waiting at all?
So looking at that output, I have these observations:
* we are dealing with a fairly tight window; in the successful run, the timestamp of the symlink is 06:34:12.981040800 and we run the `ls` command to show us that at 06:34:12,991; that's about 10ms
* based on the timestamp of the next log message in each case, `udevadm settle` on success appears to take substantially longer (06:34:12,907 -> 06:34:12,991; 84ms) than on failure (06:25:43,624 -> 06:25:43,633; 9ms)
* on success, the timestamp of the symlink is between the `udevadm settle` and the next log message, which suggests that it was created as a result of the settle
So none of this points to something _obvious_ (or, at least, not something obvious to me). Given the disparity in the apparent `udevadm settle` execution times, I wonder if the udev events from the resize haven't made their way in to the queue before we execute the settle, meaning it just exits without waiting at all?