HBASE-30303: Fix balancing for regions assigned through RSGroup fallback

[Sigma-Ma]

JIRA: https://issues.apache.org/jira/browse/HBASE-30303

What changes were proposed in this pull request?

This pull request updates RSGroupBasedLoadBalancer#correctAssignments to use the same candidate-server selection logic as region assignment when RSGroup fallback is enabled.

When the target RSGroup has no online RegionServers, correctAssignments now validates region locations against the effective fallback candidates.

The existing TestRSGroupsFallback test is also extended to verify that no region moves are calculated when regions are correctly hosted on fallback servers.

Why are the changes needed?

When RSGroup fallback is enabled and a table's target RSGroup has no online RegionServers, its regions are assigned to the default RSGroup, or to any available RegionServer if the default group is also unavailable.

Previously, correctAssignments only validated these regions against the table's original RSGroup. Therefore, regions already hosted on valid fallback servers were incorrectly classified as misplaced during every balance run, resulting in continuous and unnecessary region movement.

How was this patch tested?

The following tests were executed:

mvn -pl hbase-server \
  -am \
  -Dtest=org.apache.hadoop.hbase.rsgroup.TestRSGroupsFallback \
  -Dsurefire.failIfNoSpecifiedTests=false \
  test

mvn -pl hbase-server \
  -am \
  -Dtest=org.apache.hadoop.hbase.master.balancer.TestRSGroupBasedLoadBalancer \
  -Dsurefire.failIfNoSpecifiedTests=false \
  test

TestRSGroupsFallback verifies that balancing calculates no moves after regions fall back to the default RSGroup and after they fall back to another available RSGroup.

[chaijunjie0101]

maybe we need trigger balancer once before we check the balance plan，and make sure balancer is doing to generate plan, because the default balance is scheduled

[Sigma-Ma]

Thanks, good catch. movesCalculated is also zero when the balancer is skipped, so the previous assertion could pass without balanceCluster actually running.

I updated the test to capture the BalanceResponse from the same balance invocation, assert isBalancerRan() first, and then assert that getMovesCalculated() is zero for both fallback cases.

[chaijunjie0101]

please confirm all UT test passed @Sigma-Ma

[Sigma-Ma]

please confirm all UT test passed @Sigma-Ma

I ran the unit test suite with:

mvn \
  -pl hbase-server \
  -am \
  -Dtest=org.apache.hadoop.hbase.rsgroup.TestRSGroupsFallback \
  -Dsurefire.failIfNoSpecifiedTests=false \
  test

mvn \
  -pl hbase-server \
  -am \
  -Dtest=org.apache.hadoop.hbase.master.balancer.TestRSGroupBasedLoadBalancer \
  -Dsurefire.failIfNoSpecifiedTests=false \
  test

mvn \
  -pl hbase-server \
  -am \
  -P runAllTests \
  test

The HBASE-30303 related tests passed, but the full run encountered one unrelated failure: TestExecutorService hit an existing synchronization race under high system load. I also reproduced the TestExecutorService failure in isolation.