supabase / supavisor / 36638313288

84%

Build Type

push

github

Committed by web-flow

Commit Message

feat: add pool rebalancing (#1218)

A pool stays on the node it started on, so pools started while only part
of the cluster was up are concentrated on the first nodes.

Supavisor.rebalance/1 stops, on every node, the pools whose target node
from determine_node/2 is another one. Their clients are drained and
start the pool on its new node when they reconnect. Pools are stopped in
the background, and the pools being moved are returned. It is exposed as
POST /api/rebalance, with dry_run and max_concurrency options.

The target node depends on the tenant's availability zone, which is now
stored in the :syn metadata of the tenant supervisor. Cluster pools
store nil, as their replicas may be in different zones.

Coverage Stats

43 of 47 new or added lines in 8 files covered. (91.49%)

7 existing lines in 2 files now uncovered.

3050 of 3777 relevant lines covered (80.75%)

49876.83 hits per line