A Jupyter Server Session Manager that persists and rehydrates kernel sessions beyond the lifetime of a server.
This is particularly useful for Jupyter Servers running remote kernels and contents.
Install and enable the extension:
pip install jupyter_server_synchronizer
jupyter server extension enable jupyter_server_synchronizer
When you start a Jupyter Server, it synchronize all managers before the Web application is started.
jupyter server --ServerApp.session_manager_class=jupyter_server_synchronizer.SynchronizerSessionManager
To synchronize periodically, enable the auto-synchronizing feature using the autosync
config option. For example,
jupyter server --ServerApp.session_manager_class=jupyter_server_synchronizer.SynchronizerSessionManager --SynchronizerSessionManager.autosync=True
Otherwise, you can trigger the synchronization making a POST
request to the /api/sync
endpoint.
Below is a example of running the synchronizer with Jupyter Server talking to a Jupyter Kernel Gateway as its "remote" kernel service provider.
First, start the Kernel Gateway. You'll need to enable the list_kernels
method. In the example, we are assuming the Kernel Gateway is not multi-tenant; i.e. there is a single KG for a single Jupyter Server. We'll set the port to 9999
to free up 8888
for our Jupyter Server.
jupyter kernelgateway \
--port 9999 \
--JupyterWebsocketPersonality.list_kernels=True
Second, start the Jupyter Server and point it at the Kernel Gateway. Note that we set a database_filepath
trait in both the SessionManager
and SynchronizerExtension
(these paths don't need to be the same). The Synchronize relies on saving/storing of information about Jupyter kernels and sessions in a persistent database. This information is necessary to rehydrate and synchronize.
We'll enable the "autosync" feature to periodically synchronize the server.
jupyter lab \
--gateway-url=http://127.0.0.1:9999 \
--ServerApp.session_manager_class=jupyter_server_synchronizer.SynchronizerSessionManager
--SynchronizerSessionManager.database_filepath=jupyter-database.db \
--SynchronizerSessionManager.autosync=True \
--SynchronizerSessionManager.log_level=DEBUG
Now, let's kill that server:
kill -9 $(echo $(pgrep -lf jupyter-lab) | awk '{print $1;}')
And restart it to see if the kernels rehydrate and begin synchronizing again.
jupyter lab \
--gateway-url=http://127.0.0.1:9999 \
--ServerApp.session_manager_class=jupyter_server_synchronizer.SynchronizerSessionManager
--SessionManager.database_filepath=jupyter-database.db \
--SynchronizerSessionManager.autosync=True \
--SynchronizerSessionManager.log_level=DEBUG