Add note on how to switch GCP project (#1539)

* doc

* clarify

* uppercase

* fix mnist example

* todo

docs/source/getting-started/installation.rst

@@ -75,6 +75,10 @@ If running :code:`conda install -c conda-forge google-cloud-sdk` produces the er
 
 Note: if you encounter *Authorization Error (Error 400: invalid_request)* with the url generated by :code:`gcloud auth login`, try installing the latest version of the `Google Cloud SDK <https://cloud.google.com/sdk/docs/install>`_ (e.g., with :code:`conda install -c conda-forge google-cloud-sdk`) on **your local machine** (which opened the browser) and rerun the command.
 
+.. note::
+
+  If you are using multiple GCP projects, list all the configs by :code:`gcloud config configurations list` and activate one by :code:`gcloud config configurations activate <CONFIG_NAME>` (See `GCP docs <https://cloud.google.com/sdk/docs/configurations#activating_a_configuration>`_).
+
 **Azure**
 
 .. code-block:: console

docs/source/reference/tpu.rst

@@ -86,6 +86,7 @@ Here is a complete task YAML that runs `MNIST training <https://cloud.google.com
          pip install "jax[tpu]>=0.2.16" -f https://storage.googleapis.com/jax-releases/libtpu_releases.html
          pip install --upgrade clu
          pip install -e flax
+         pip install tensorflow tensorflow-datasets
       fi
 
    run: |

sky/clouds/gcp.py

@@ -23,6 +23,8 @@
     '~/.config/gcloud/'
     'application_default_credentials.json')
 
+# TODO(wei-lin): config_default may not be the config in use.
+# See: https://github.com/skypilot-org/skypilot/pull/1539
 GCP_CONFIG_PATH = '~/.config/gcloud/configurations/config_default'
 # Do not place the backup under the gcloud config directory, as ray
 # autoscaler can overwrite that directory on the remote nodes.