Early Stopping With No Validation Set

Early stopping is requested but no eval_set is provided to monitor it.

import xgboost as xgb

model = xgb.XGBClassifier(
    n_estimators=1000,
    learning_rate=0.05,
    early_stopping_rounds=20
)
model.fit(X_train, y_train)

Early stopping never triggers (or the fit errors). What is the bug?