You are viewing a plain text version of this content. The canonical link for it is here.
Posted to builds@fluss.apache.org by GitBox <gi...@apache.org> on 2026/08/10 03:22:24 UTC
[GH] (fluss/feature/20260808-table-leader-distribution): Workflow run "Client Integration" failed!
The GitHub Actions job "Client Integration" on fluss.git/feature/20260808-table-leader-distribution has failed.
Run started by GitHub user fxbing (triggered by fxbing).
Head commit for run:
44083d63a04e04d144ef635677a6cc0dff102c96 / 方晓兵 <bi...@qq.com>
[server] Balance replicas and leaders within each table
The existing REPLICA_DISTRIBUTION and LEADER_DISTRIBUTION goals only look
at cluster-wide counts, so a single table can still be badly skewed while
the cluster looks balanced, which turns a hot table into a CPU and request
hot spot on a few TabletServers.
- add TABLE_REPLICA_DISTRIBUTION and TABLE_LEADER_DISTRIBUTION soft goals
that keep the replica and leader count of every table inside a per-table
balance window, preferring leadership transfer between existing replicas
and moving leader replicas only when placement requires it
- track per-table replicas and leader counts in ServerModel and add
per-table standard deviation snapshots to ClusterModelStats so the new
goals can detect a per-table regression
- cover the goals with per-table, partitioned-table, offline-server and
boundary tests, plus goal-combination, convergence and stable numeric
mapping tests that guard against later-goal and wire-compatibility regressions
Report URL: https://github.com/apache/fluss/actions/runs/31352341070
With regards,
GitHub Actions via GitBox