Create loggers property for Trainer and LightningModule

pytorch_lightning/trainer/connectors/logger_connector/logger_connector.py

@@ -78,17 +78,21 @@ def should_update_logs(self) -> bool:
     def configure_logger(self, logger: Union[bool, LightningLoggerBase, Iterable[LightningLoggerBase]]) -> None:
         if isinstance(logger, bool):
             # default logger
-            self.trainer.logger = (
-                TensorBoardLogger(
+            if logger:
+                default_logger = TensorBoardLogger(
                     save_dir=self.trainer.default_root_dir, version=SLURMEnvironment.job_id(), name="lightning_logs"
                 )
-                if logger
-                else None
-            )
+                self.trainer.logger = default_logger
+                self.trainer.loggers = [default_logger]
+            else:
+                self.trainer.logger = None
+                self.trainer.loggers = []
         elif isinstance(logger, Iterable):
             self.trainer.logger = LoggerCollection(logger)
+            self.trainer.loggers = list(logger)
         else:
             self.trainer.logger = logger
+            self.trainer.loggers = [logger]
 
     def log_metrics(self, metrics: _OUT_DICT, step: Optional[int] = None) -> None:
         """Logs the metric dict passed in. If `step` parameter is None and `step` key is presented is metrics, uses

pytorch_lightning/trainer/trainer.py

@@ -565,6 +565,7 @@ def __init__(
 
         # init logger flags
         self.logger: Optional[LightningLoggerBase]
+        self.loggers: List[LightningLoggerBase]
         self.logger_connector.on_trainer_init(logger, flush_logs_every_n_steps, log_every_n_steps, move_metrics_to_cpu)
 
         # init debugging flags
@@ -613,7 +614,9 @@ def _init_debugging_flags(
             self.fit_loop.max_epochs = 1
             val_check_interval = 1.0
             self.check_val_every_n_epoch = 1
-            self.logger = DummyLogger() if self.logger is not None else None
+            if self.logger is not None:
+                self.logger = DummyLogger()
+                self.loggers = [self.logger]
 
             rank_zero_info(
                 "Running in fast_dev_run mode: will run a full train,"

tests/loggers/test_base.py

@@ -176,6 +176,32 @@ def training_step(self, batch, batch_idx):
     assert logger2.finalized_status == "success"
 
 
+def test_loggers_property(tmpdir):
+    logger1 = CustomLogger()
+    logger2 = CustomLogger()
+
+    # trainer.loggers should be a copy of the input list
+    trainer = Trainer(max_steps=2, log_every_n_steps=1, logger=[logger1, logger2], default_root_dir=tmpdir)
+
+    assert trainer.loggers == [logger1, logger2]
+
+    # trainer.loggers should create a list of size 1
+    trainer = Trainer(max_steps=2, log_every_n_steps=1, logger=logger1, default_root_dir=tmpdir)
+
+    assert trainer.loggers == [logger1]
+
+    # trainer.loggers should be None
+    trainer = Trainer(max_steps=2, log_every_n_steps=1, logger=False, default_root_dir=tmpdir)
+
+    assert trainer.loggers is []
+
+    # trainer.loggers should be a list of size 1 holding the default logger
+    trainer = Trainer(max_steps=2, log_every_n_steps=1, logger=True, default_root_dir=tmpdir)
+
+    assert trainer.loggers == [trainer.logger]
+    assert type(trainer.loggers[0]) == TensorBoardLogger
+
+
 def test_multiple_loggers_pickle(tmpdir):
     """Verify that pickling trainer with multiple loggers works."""