Refactor codebase to use the loggers property instead of logger

CHANGELOG.md

@@ -83,6 +83,13 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 - Added a `MisconfigurationException` if user provided `opt_idx` in scheduler config doesn't match with actual optimizer index of its respective optimizer ([#11247](https://github.com/PyTorchLightning/pytorch-lightning/pull/11247))
 
+
+- Added a `loggers` property to `Trainer` which returns a list of loggers provided by the user ([#11683](https://github.com/PyTorchLightning/pytorch-lightning/pull/11683))
+
+
+- Added a `loggers` property to `LightningModule` which retrieves the `loggers` property from `Trainer` ([#11683](https://github.com/PyTorchLightning/pytorch-lightning/pull/11683))
+
+
 - Added support for DDP when using a `CombinedLoader` for the training data ([#11648](https://github.com/PyTorchLightning/pytorch-lightning/pull/11648))
 
 

docs/source/common/lightning_module.rst

@@ -952,7 +952,7 @@ The current step (does not reset each epoch)
 .. code-block:: python
 
     def training_step(self, batch, batch_idx):
-        self.logger.experiment.log_image(..., step=self.global_step)
+        self.loggers[logger_index].experiment.log_image(..., step=self.global_step)
 
 hparams
 ~~~~~~~
@@ -983,6 +983,17 @@ The current logger being used (tensorboard or other supported logger)
         # the particular logger
         tensorboard_logger = self.logger.experiment
 
+loggers
+~~~~~~~
+
+The list of loggers currently being used.
+
+.. code-block:: python
+
+    def training_step(self, batch, batch_idx):
+        # List of LightningLoggerBase objects
+        self.loggers
+
 local_rank
 ~~~~~~~~~~~
 

docs/source/common/loggers.rst

@@ -64,7 +64,7 @@ The :class:`~pytorch_lightning.loggers.CometLogger` is available anywhere except
     class MyModule(LightningModule):
         def any_lightning_module_function_or_hook(self):
             some_img = fake_image()
-            self.logger.experiment.add_image("generated_images", some_img, 0)
+            self.loggers[logger_index].experiment.add_image("generated_images", some_img, 0)
 
 .. seealso::
     :class:`~pytorch_lightning.loggers.CometLogger` docs.
@@ -135,9 +135,9 @@ The :class:`~pytorch_lightning.loggers.NeptuneLogger` is available anywhere exce
         def any_lightning_module_function_or_hook(self):
             # generic recipe for logging custom metadata (neptune specific)
             metadata = ...
-            self.logger.experiment["your/metadata/structure"].log(metadata)
+            self.loggers[logger_index].experiment["your/metadata/structure"].log(metadata)
 
-Note that syntax: ``self.logger.experiment["your/metadata/structure"].log(metadata)``
+Note that syntax: ``self.loggers[logger_index].experiment["your/metadata/structure"].log(metadata)``
 is specific to Neptune and it extends logger capabilities.
 Specifically, it allows you to log various types of metadata like scores, files,
 images, interactive visuals, CSVs, etc. Refer to the
@@ -173,7 +173,7 @@ The :class:`~pytorch_lightning.loggers.TensorBoardLogger` is available anywhere
     class MyModule(LightningModule):
         def any_lightning_module_function_or_hook(self):
             some_img = fake_image()
-            self.logger.experiment.add_image("generated_images", some_img, 0)
+            self.loggers[logger_index].experiment.add_image("generated_images", some_img, 0)
 
 .. seealso::
     :class:`~pytorch_lightning.loggers.TensorBoardLogger` docs.
@@ -213,9 +213,9 @@ The :class:`~pytorch_lightning.loggers.WandbLogger` is available anywhere except
         def any_lightning_module_function_or_hook(self):
             some_img = fake_image()
             # Option 1
-            self.logger.experiment.log({"generated_images": [wandb.Image(some_img, caption="...")]})
+            self.loggers[logger_index].experiment.log({"generated_images": [wandb.Image(some_img, caption="...")]})
             # Option 2 for specifically logging images
-            self.logger.log_image(key="generated_images", images=[some_img])
+            self.loggers[logger_index].log_image(key="generated_images", images=[some_img])
 
 .. seealso::
     - :class:`~pytorch_lightning.loggers.WandbLogger` docs.
@@ -246,7 +246,4 @@ The loggers are available as a list anywhere except ``__init__`` in your
     class MyModule(LightningModule):
         def any_lightning_module_function_or_hook(self):
             some_img = fake_image()
-            # Option 1
-            self.logger.experiment[0].add_image("generated_images", some_img, 0)
-            # Option 2
-            self.logger[0].experiment.add_image("generated_images", some_img, 0)
+            self.loggers[logger_index].experiment.add_image("generated_images", some_img, 0)

docs/source/common/trainer.rst

@@ -1741,9 +1741,21 @@ The current logger being used. Here's an example using tensorboard
 
 .. code-block:: python
 
-    def training_step(self, batch, batch_idx):
-        logger = self.trainer.logger
-        tensorboard = logger.experiment
+    logger = trainer.logger
+    tensorboard = logger.experiment
+
+
+loggers (p)
+***********
+
+The list of loggers currently being used.
+
+.. code-block:: python
+
+    # List of LightningLoggerBase objects
+    loggers = trainer.loggers
+    for logger in loggers:
+        logger.log_metrics({"foo": 1.0})
 
 
 logged_metrics

docs/source/extensions/logging.rst

@@ -192,7 +192,7 @@ If you want to log anything that is not a scalar, like histograms, text, images,
     def training_step(self):
         ...
         # the logger you used (in this case tensorboard)
-        tensorboard = self.logger.experiment
+        tensorboard = self.loggers[logger_index].experiment
         tensorboard.add_image()
         tensorboard.add_histogram(...)
         tensorboard.add_figure(...)
@@ -378,7 +378,8 @@ in the `hparams tab <https://pytorch.org/docs/stable/tensorboard.html#torch.util
 
         # Using custom or multiple metrics (default_hp_metric=False)
         def on_train_start(self):
-            self.logger.log_hyperparams(self.hparams, {"hp/metric_1": 0, "hp/metric_2": 0})
+            for logger in self.loggers:
+                logger.log_hyperparams(self.hparams, {"hp/metric_1": 0, "hp/metric_2": 0})
 
 
         def validation_step(self, batch, batch_idx):

docs/source/starter/introduction_guide.rst

@@ -492,7 +492,7 @@ You can also use any method of your logger directly:
 .. code-block:: python
 
     def training_step(self, batch, batch_idx):
-        tensorboard = self.logger.experiment
+        tensorboard = self.loggers[logger_index].experiment
         tensorboard.any_summary_writer_method_you_want()
 
 Once your training starts, you can view the logs by using your favorite logger or booting up the Tensorboard logs:

pl_examples/domain_templates/generative_adversarial_net.py

@@ -206,7 +206,8 @@ def on_train_epoch_end(self):
         # log sampled images
         sample_imgs = self(z)
         grid = torchvision.utils.make_grid(sample_imgs)
-        self.logger.experiment.add_image("generated_images", grid, self.current_epoch)
+        for logger in self.loggers:
+            logger.experiment.add_image("generated_images", grid, self.current_epoch)
 
 
 def main(args: Namespace) -> None:

pytorch_lightning/callbacks/device_stats_monitor.py

@@ -44,7 +44,7 @@ class DeviceStatsMonitor(Callback):
     """
 
     def setup(self, trainer: "pl.Trainer", pl_module: "pl.LightningModule", stage: Optional[str] = None) -> None:
-        if not trainer.logger:
+        if not trainer.loggers:
             raise MisconfigurationException("Cannot use DeviceStatsMonitor callback with Trainer that has no logger.")
 
     def on_train_batch_start(
@@ -55,17 +55,18 @@ def on_train_batch_start(
         batch_idx: int,
         unused: Optional[int] = 0,
     ) -> None:
-        if not trainer.logger:
+        if not trainer.loggers:
             raise MisconfigurationException("Cannot use `DeviceStatsMonitor` callback with `Trainer(logger=False)`.")
 
         if not trainer.logger_connector.should_update_logs:
             return
 
         device = trainer.strategy.root_device
         device_stats = trainer.accelerator.get_device_stats(device)
-        separator = trainer.logger.group_separator
-        prefixed_device_stats = _prefix_metric_keys(device_stats, "on_train_batch_start", separator)
-        trainer.logger.log_metrics(prefixed_device_stats, step=trainer.global_step)
+        for logger in trainer.loggers:
+            separator = logger.group_separator
+            prefixed_device_stats = _prefix_metric_keys(device_stats, "on_train_batch_start", separator)
+            logger.log_metrics(prefixed_device_stats, step=trainer.global_step)
 
     def on_train_batch_end(
         self,
@@ -76,17 +77,18 @@ def on_train_batch_end(
         batch_idx: int,
         unused: Optional[int] = 0,
     ) -> None:
-        if not trainer.logger:
+        if not trainer.loggers:
             raise MisconfigurationException("Cannot use `DeviceStatsMonitor` callback with `Trainer(logger=False)`.")
 
         if not trainer.logger_connector.should_update_logs:
             return
 
         device = trainer.strategy.root_device
         device_stats = trainer.accelerator.get_device_stats(device)
-        separator = trainer.logger.group_separator
-        prefixed_device_stats = _prefix_metric_keys(device_stats, "on_train_batch_end", separator)
-        trainer.logger.log_metrics(prefixed_device_stats, step=trainer.global_step)
+        for logger in trainer.loggers:
+            separator = logger.group_separator
+            prefixed_device_stats = _prefix_metric_keys(device_stats, "on_train_batch_end", separator)
+            logger.log_metrics(prefixed_device_stats, step=trainer.global_step)
 
 
 def _prefix_metric_keys(metrics_dict: Dict[str, float], prefix: str, separator: str) -> Dict[str, float]:

pytorch_lightning/callbacks/gpu_stats_monitor.py

@@ -124,7 +124,7 @@ def __init__(
         self._gpu_ids: List[str] = []  # will be assigned later in setup()
 
     def setup(self, trainer: "pl.Trainer", pl_module: "pl.LightningModule", stage: Optional[str] = None) -> None:
-        if not trainer.logger:
+        if not trainer.loggers:
             raise MisconfigurationException("Cannot use GPUStatsMonitor callback with Trainer that has no logger.")
 
         if trainer._device_type != _AcceleratorType.GPU:
@@ -162,8 +162,9 @@ def on_train_batch_start(
             # First log at beginning of second step
             logs["batch_time/inter_step (ms)"] = (time.time() - self._snap_inter_step_time) * 1000
 
-        assert trainer.logger is not None
-        trainer.logger.log_metrics(logs, step=trainer.global_step)
+        assert trainer.loggers
+        for logger in trainer.loggers:
+            logger.log_metrics(logs, step=trainer.global_step)
 
     @rank_zero_only
     def on_train_batch_end(
@@ -187,8 +188,9 @@ def on_train_batch_end(
         if self._log_stats.intra_step_time and self._snap_intra_step_time:
             logs["batch_time/intra_step (ms)"] = (time.time() - self._snap_intra_step_time) * 1000
 
-        assert trainer.logger is not None
-        trainer.logger.log_metrics(logs, step=trainer.global_step)
+        assert trainer.loggers
+        for logger in trainer.loggers:
+            logger.log_metrics(logs, step=trainer.global_step)
 
     @staticmethod
     def _get_gpu_ids(device_ids: List[int]) -> List[str]:

pytorch_lightning/callbacks/lr_monitor.py

@@ -104,7 +104,7 @@ def on_train_start(self, trainer: "pl.Trainer", *args: Any, **kwargs: Any) -> No
             MisconfigurationException:
                 If ``Trainer`` has no ``logger``.
         """
-        if not trainer.logger:
+        if not trainer.loggers:
             raise MisconfigurationException(
                 "Cannot use `LearningRateMonitor` callback with `Trainer` that has no logger."
             )
@@ -149,7 +149,7 @@ def _check_no_key(key: str) -> bool:
         self.last_momentum_values = {name + "-momentum": None for name in names_flatten}
 
     def on_train_batch_start(self, trainer: "pl.Trainer", *args: Any, **kwargs: Any) -> None:
-        assert trainer.logger is not None
+        assert trainer.loggers
         if not trainer.logger_connector.should_update_logs:
             return
 
@@ -158,16 +158,18 @@ def on_train_batch_start(self, trainer: "pl.Trainer", *args: Any, **kwargs: Any)
             latest_stat = self._extract_stats(trainer, interval)
 
             if latest_stat:
-                trainer.logger.log_metrics(latest_stat, step=trainer.global_step)
+                for logger in trainer.loggers:
+                    logger.log_metrics(latest_stat, step=trainer.global_step)
 
     def on_train_epoch_start(self, trainer: "pl.Trainer", *args: Any, **kwargs: Any) -> None:
-        assert trainer.logger is not None
+        assert trainer.loggers
         if self.logging_interval != "step":
             interval = "epoch" if self.logging_interval is None else "any"
             latest_stat = self._extract_stats(trainer, interval)
 
             if latest_stat:
-                trainer.logger.log_metrics(latest_stat, step=trainer.global_step)
+                for logger in trainer.loggers:
+                    logger.log_metrics(latest_stat, step=trainer.global_step)
 
     def _extract_stats(self, trainer: "pl.Trainer", interval: str) -> Dict[str, float]:
         latest_stat = {}

pytorch_lightning/callbacks/model_checkpoint.py

@@ -385,8 +385,9 @@ def save_checkpoint(self, trainer: "pl.Trainer") -> None:
         self._save_last_checkpoint(trainer, monitor_candidates)
 
         # notify loggers
-        if trainer.is_global_zero and trainer.logger:
-            trainer.logger.after_save_checkpoint(proxy(self))
+        if trainer.is_global_zero:
+            for logger in trainer.loggers:
+                logger.after_save_checkpoint(proxy(self))
 
     def _should_skip_saving_checkpoint(self, trainer: "pl.Trainer") -> bool:
         from pytorch_lightning.trainer.states import TrainerFn
@@ -580,18 +581,22 @@ def __resolve_ckpt_dir(self, trainer: "pl.Trainer") -> None:
         if self.dirpath is not None:
             return  # short circuit
 
-        if trainer.logger is not None:
+        if trainer.loggers:
             if trainer.weights_save_path != trainer.default_root_dir:
                 # the user has changed weights_save_path, it overrides anything
                 save_dir = trainer.weights_save_path
             else:
-                save_dir = trainer.logger.save_dir or trainer.default_root_dir
-
+                if len(trainer.loggers) == 1 and trainer.loggers[0].save_dir:
+                    save_dir = trainer.loggers[0].save_dir
+                else:
+                    save_dir = trainer.default_root_dir
+            # TODO: Find out we handle trainer.logger.version without LoggerCollection
             version = (
                 trainer.logger.version
                 if isinstance(trainer.logger.version, str)
                 else f"version_{trainer.logger.version}"
             )
+            # TODO: Find out we handle trainer.logger.name without LoggerCollection
             ckpt_path = os.path.join(save_dir, str(trainer.logger.name), version, "checkpoints")
         else:
             ckpt_path = os.path.join(trainer.weights_save_path, "checkpoints")

pytorch_lightning/callbacks/progress/base.py

@@ -213,6 +213,7 @@ def get_standard_metrics(trainer: "pl.Trainer", pl_module: "pl.LightningModule")
     if pl_module.truncated_bptt_steps > 0:
         items_dict["split_idx"] = trainer.fit_loop.split_idx
 
+    # TODO: Find out we handle trainer.logger.version without LoggerCollection
     if trainer.logger is not None and trainer.logger.version is not None:
         version = trainer.logger.version
         if isinstance(version, str):

pytorch_lightning/callbacks/xla_stats_monitor.py

@@ -69,7 +69,7 @@ def __init__(self, verbose: bool = True) -> None:
         self._verbose = verbose
 
     def on_train_start(self, trainer: "pl.Trainer", pl_module: "pl.LightningModule") -> None:
-        if not trainer.logger:
+        if not trainer.loggers:
             raise MisconfigurationException("Cannot use XLAStatsMonitor callback with Trainer that has no logger.")
 
         if trainer._device_type != _AcceleratorType.TPU:
@@ -87,7 +87,7 @@ def on_train_epoch_start(self, trainer: "pl.Trainer", pl_module: "pl.LightningMo
         self._start_time = time.time()
 
     def on_train_epoch_end(self, trainer: "pl.Trainer", pl_module: "pl.LightningModule") -> None:
-        if not trainer.logger:
+        if not trainer.loggers:
             raise MisconfigurationException("Cannot use XLAStatsMonitor callback with Trainer that has no logger.")
 
         device = trainer.strategy.root_device
@@ -101,10 +101,11 @@ def on_train_epoch_end(self, trainer: "pl.Trainer", pl_module: "pl.LightningModu
         peak_memory = trainer.strategy.reduce(peak_memory) * 0.001
         epoch_time = trainer.strategy.reduce(epoch_time)
 
-        trainer.logger.log_metrics(
-            {"avg. free memory (MB)": float(free_memory), "avg. peak memory (MB)": float(peak_memory)},
-            step=trainer.current_epoch,
-        )
+        for logger in trainer.loggers:
+            logger.log_metrics(
+                {"avg. free memory (MB)": float(free_memory), "avg. peak memory (MB)": float(peak_memory)},
+                step=trainer.current_epoch,
+            )
 
         if self._verbose:
             rank_zero_info(f"Average Epoch time: {epoch_time:.2f} seconds")

pytorch_lightning/core/hooks.py

@@ -285,7 +285,7 @@ def on_before_optimizer_step(self, optimizer, optimizer_idx):
                 # example to inspect gradient information in tensorboard
                 if self.trainer.global_step % 25 == 0:  # don't make the tf file huge
                     for k, v in self.named_parameters():
-                        self.logger.experiment.add_histogram(
+                        self.loggers[logger_index].experiment.add_histogram(
                             tag=k, values=v.grad, global_step=self.trainer.global_step
                         )
         """