Neptune on-premises FAQ#
How often do you release new versions? How do we know when they are available?
We release new versions every 3-6 months. Typically, we package our new on-prem versions in a way to provide our on-prem customers with at least one significant improvement with each update.
You can always check the Product portal for what will be included in the upcoming release.
Note that if a critical bug is found, a new version with a fix is released as soon as the fix is ready. Non-critical bugs are fixed and delivered with the next update.
You will receive a heads-up a few weeks before a new version is available, as we recommend that you perform a backup and plan internally with users when the upgrade is the most convenient for you. During the upgrade, users may experience a short (around 30 minutes) downtime.
How do I know which features are available in each version?
For release information, visit the Product portal .
Can Neptune be installed in our cloud account?
Can Neptune be installed on our own infra?
How long does it take to get the installation up and running?
Assuming the prerequisites are met, anything from a few minutes to an hour, depending on the connection speed.
What kind of data are stored by MySQL, Elasticsearch, and object storage? Which storage pressure will be the greatest?
MySQL storage consumption will usually rise in two cases:
- The number of concurrent runs logging lots of metadata increases (series data passes through MySQL before being stored in the object storage).
- The number of runs, fields, etc. increases.
Elasticsearch storage consumption will rise with the number of runs and fields.
Object storage is where files (such as source code, images, model weights), series contents (numeric series, logs) end up. This is the storage that will likely be under the most pressure.
Behavior and usage
How do I connect my Neptune Python client to the on-premises instance?
Once you access your on-prem installation webpage, just do the same as in the SaaS version: get your API token from the user menu → Get your API token, then use it with the client library. For help, see Setting your API token.
Your currently used domain will be encoded into the token itself, so as long as the place where you run your Neptune client library knows how to resolve this domain, everything should work without any issues.
Note that we strongly recommend using HTTPS (SSL certificates) when exposing Neptune. If you want to use HTTP or you have a self-signed certificate, exporting some environmental variables might be required for the client to work properly.
Is there a way to replace the license key without triggering a redeployment?
2.2.x is the last version in which you need to update the license key this way. With the
2.3 release, you can replace the license key through the web application.
What kind of Git access does Neptune have in our repository?
The (open-source) Python client library accesses your local Git repository upon initiating a run to gather some basic information about the code. The Neptune server does not and cannot access any remote repositories.
For details, see What is logged automatically.
If we want to do Neptune monitoring, which components of Neptune should we mainly focus on? Which metrics of these components should we track and observe?
The standard monitoring of CPU and RAM is enough in most cases. The most CPU-heavy pod is the component called "leaderboard" – under enough pressure it may need to be scaled out.
It's possible to scale the deployment automatically using a Kubernetes mechanism called Horizontal Pod Autoscaling .
How can we get overview data of Neptune, like how many experiments are currently running in total?
Using the Python API, you can download the runs table with the
fetch_runs_table method and the
state="active" filter. This returns all the actively running runs at that point in time.
If we want to do data migration, what should we do?
Data migrations are possible, but often non-trivial due to the amount of data. Always consult with the neptune.ai team beforehand.
Is it possible to do a trial in the SaaS version and then move it to private cloud?
No, as those are separate instances.