_set_keyspace_for_all_pools callback receives only the last pool's errors instead of all accumulated errors

[mykaul]

Bug

Session._set_keyspace_for_all_pools() passes host_errors (the last pool to finish) instead of errors (the accumulated dict of all pools) to its callback.

cassandra/cluster.py:3442 — present since the original commit d281b50c9 (2013-10-11).

def pool_finished_setting_keyspace(pool, host_errors):
    remaining_callbacks.remove(pool)
    if host_errors:
        errors[pool.host] = host_errors          # accumulates into `errors`

    if not remaining_callbacks:
        callback(host_errors)                     # BUG: should be callback(errors)

Impact

When the last pool to finish has no errors, the callback receives [] (empty list) instead of the dict of all accumulated errors. Downstream, _set_keyspace_completed interprets this as success:

def _set_keyspace_completed(self, errors):
    if not errors:              # `not []` is True
        self._set_final_result(None)    # reports success
    else:
        self._set_final_exception(...)

This means keyspace changes can silently succeed from the driver's perspective even when some pools failed — partial failures are dropped.

Reproduction

tests/unit/test_cluster.py::SessionTest::test_set_keyspace_for_all_pools_reports_all_errors — this test expects the correct behavior but fails because of the bug. It was likely never run before against a compiled extension that exercised this code path with multiple pools.

Fix

callback(host_errors) → callback(errors) on line 3442.