diff --git a/include/xgboost/global_config.h b/include/xgboost/global_config.h
index b8a2ceb5438e..70eca5c23108 100644
--- a/include/xgboost/global_config.h
+++ b/include/xgboost/global_config.h
@@ -15,7 +15,7 @@ namespace xgboost {
 class Json;
 
 struct GlobalConfiguration : public XGBoostParameter<GlobalConfiguration> {
-  int verbosity;
+  int verbosity { 1 };
   DMLC_DECLARE_PARAMETER(GlobalConfiguration) {
     DMLC_DECLARE_FIELD(verbosity)
         .set_range(0, 3)
diff --git a/python-package/xgboost/dask.py b/python-package/xgboost/dask.py
index 67c4d6332c26..c449c0e75665 100644
--- a/python-package/xgboost/dask.py
+++ b/python-package/xgboost/dask.py
@@ -626,6 +626,7 @@ def _get_workers_from_data(dtrain: DaskDMatrix, evals=()):
 
 
 async def _train_async(client,
+                       global_config,
                        params,
                        dtrain: DaskDMatrix,
                        *args,
@@ -639,7 +640,6 @@ async def _train_async(client,
 
     workers = list(_get_workers_from_data(dtrain, evals))
     _rabit_args = await _get_rabit_args(len(workers), client)
-    _global_config = config.get_config()
 
     def dispatched_train(worker_addr, rabit_args, dtrain_ref, dtrain_idt, evals_ref):
         '''Perform training on a single worker.  A local function prevents pickling.
@@ -647,7 +647,7 @@ def dispatched_train(worker_addr, rabit_args, dtrain_ref, dtrain_idt, evals_ref)
         '''
         LOGGER.info('Training on %s', str(worker_addr))
         worker = distributed.get_worker()
-        with RabitContext(rabit_args), config.config_context(**_global_config):
+        with RabitContext(rabit_args), config.config_context(**global_config):
             local_dtrain = _dmatrix_from_list_of_parts(**dtrain_ref)
             local_evals = []
             if evals_ref:
@@ -735,8 +735,11 @@ def train(client, params, dtrain, *args, evals=(), early_stopping_rounds=None,
     '''
     _assert_dask_support()
     client = _xgb_get_client(client)
+    # Get global configuration before transferring computation to another thread or
+    # process.
+    global_config = config.get_config()
     return client.sync(
-        _train_async, client, params, dtrain=dtrain, *args, evals=evals,
+        _train_async, client, global_config, params, dtrain=dtrain, *args, evals=evals,
         early_stopping_rounds=early_stopping_rounds, **kwargs)
 
 
@@ -760,7 +763,7 @@ async def _direct_predict_impl(client, data, predict_fn):
 
 
 # pylint: disable=too-many-statements
-async def _predict_async(client, model, data, missing=numpy.nan, **kwargs):
+async def _predict_async(client, global_config, model, data, missing=numpy.nan, **kwargs):
     if isinstance(model, Booster):
         booster = model
     elif isinstance(model, dict):
@@ -771,11 +774,9 @@ async def _predict_async(client, model, data, missing=numpy.nan, **kwargs):
         raise TypeError(_expect([DaskDMatrix, da.Array, dd.DataFrame],
                                 type(data)))
 
-    _global_config = config.get_config()
-
     def mapped_predict(partition, is_df):
         worker = distributed.get_worker()
-        with config.config_context(**_global_config):
+        with config.config_context(**global_config):
             booster.set_param({'nthread': worker.nthreads})
             m = DMatrix(partition, missing=missing, nthread=worker.nthreads)
             predt = booster.predict(m, validate_features=False, **kwargs)
@@ -801,7 +802,7 @@ def mapped_predict(partition, is_df):
     def dispatched_predict(worker_id, list_of_orders, list_of_parts):
         '''Perform prediction on each worker.'''
         LOGGER.info('Predicting on %d', worker_id)
-        with config.config_context(**_global_config):
+        with config.config_context(**global_config):
             worker = distributed.get_worker()
             list_of_parts = _get_worker_parts_ordered(meta_names, list_of_parts)
             predictions = []
@@ -907,11 +908,12 @@ def predict(client, model, data, missing=numpy.nan, **kwargs):
     '''
     _assert_dask_support()
     client = _xgb_get_client(client)
-    return client.sync(_predict_async, client, model, data,
+    global_config = config.get_config()
+    return client.sync(_predict_async, client, global_config, model, data,
                        missing=missing, **kwargs)
 
 
-async def _inplace_predict_async(client, model, data,
+async def _inplace_predict_async(client, global_config, model, data,
                                  iteration_range=(0, 0),
                                  predict_type='value',
                                  missing=numpy.nan):
@@ -927,6 +929,7 @@ async def _inplace_predict_async(client, model, data,
 
     def mapped_predict(data, is_df):
         worker = distributed.get_worker()
+        config.set_config(**global_config)
         booster.set_param({'nthread': worker.nthreads})
         prediction = booster.inplace_predict(
             data,
@@ -976,7 +979,9 @@ def inplace_predict(client, model, data,
     '''
     _assert_dask_support()
     client = _xgb_get_client(client)
-    return client.sync(_inplace_predict_async, client, model=model, data=data,
+    global_config = config.get_config()
+    return client.sync(_inplace_predict_async, client, global_config, model=model,
+                       data=data,
                        iteration_range=iteration_range,
                        predict_type=predict_type,
                        missing=missing)