"Restartless" compute scaling randomly causes restarts and/or proxy connection hangs

[piercypixel]

Related: https://github.com/neondatabase/cloud/issues/14457

Observed in preview.

In an attempt to scale compute within vm cpu bounds, without a restart, sometimes we still end up restarting due to compute live configuration errors or smth else.

Sometimes even without restarts, updating endponint cu results in active psql connections hang for a long time (slow live migration?).

ref https://neondb.slack.com/archives/C03TN5G758R/p1739812183023339

[sharnoff]

@piercypixel can you provide an example where this occurred? (i.e., environment + timestamp + compute ID)