John Spray 14e05276a3 storcon: fix a case where optimise could get stuck on unschedulable node (#10648) ...

## Problem

When a shard has two secondary locations, but one of them is on a node
with MaySchedule::No, the optimiser would get stuck, because it couldn't
decide which secondary to remove.

This is generally okay if a node is offline, but if a node is in Pause
mode for a long period of time, it's a problem.

Closes: https://github.com/neondatabase/neon/issues/10646

## Summary of changes

- Instead of insisting on finding a node in the wrong AZ to remove, find
an available node in the _right_ AZ, and remove all the others. This
ensures that if there is one live suitable node, then other
offline/paused nodes cannot hold things up.

2025-02-05 16:05:12 +00:00

..

client

storcon: don't assume host:port format in storcon client (#10347)

2025-01-10 18:35:16 +00:00

migrations

Rename "disabled" safekeeper scheduling policy to "pause" (#10410)

2025-01-16 14:30:49 +00:00

src

storcon: fix a case where optimise could get stuck on unschedulable node (#10648)

2025-02-05 16:05:12 +00:00

Cargo.toml

storcon: use diesel-async again, now with tls support (#10614)

2025-02-03 11:53:51 +00:00