Self-hosted Neptune
Neptune can be self-hosted. This means deploying and hosting your own instance of the Neptune web application, instead of using the SaaS offering at https://scale.neptune.ai.
For detailed requirements, see Prerequisites for self-hosting Neptune.
Architecture
Neptune runs on a Kubernetes cluster.
The deployment architecture is decomposable. Components can run within a single Kubernetes cluster or integrate with external infrastructure and managed services, such as Kafka.
Neptune uses ClickHouse as the main storage engine. A ClickHouse instance is mandatory for all Neptune deployments. If you don't have one yourself, Neptune can provide it.
You must provide MySQL and Redis instances for the deployment. Neptune can reuse other existing services that you have, or handle the setup and management of required services on the cluster.
High availability and resilience
Neptune handles automated failover and recovery mechanisms via Kubernetes.
Multi-zone deployment is supported.
Full multi-region deployment is not supported.
High scalability
Horizontal and vertical scaling is available on-demand. Storage can be scaled independently, ensuring flexibility in managing compute and storage growth.
Auto-scaling is not supported.
Tiered data storage
To optimize cost and performance, Neptune provides separate storage tiers: SSD and HDD.
For the possibility of using extensible storage, reach out to us at support@neptune.ai.
Monitoring and observability
Neptune provides comprehensive monitoring and observability through open-source tools like Prometheus and Grafana. The monitoring stack can be deployed together with Neptune, or you can integrate it with existing tools, such as OpenTelemetry.
You can also send data to your own monitoring service.
You can configure various alerts to notify teams of performance issues and potential failures. The monitoring covers both the platform and product, including infrastructure, services, and application performance.
Upgrading and maintenance
Neptune releases typically have zero downtime, but major architectural changes may require planned downtime.
Neptune enables dedicated testing instances for validating upgrades before production deployment. In some cases, blue-green deployment is possible.
To discuss or perform any deployment or upgrade, reach out to us at support@neptune.ai.
Backup and recovery
Neptune has a backup and restore mechanism. The backup runs once per day by default, but you can customize the frequency.
The backup pauses ingestion for 40-60 seconds. During the pause, Neptune users can't retrieve or display new experiment data.
Offsite backup storage is supported within the same cloud provider.
Operational readiness and support
Neptune offers 24/7 support, including platform administration assistance.
SRE support with quick response times is available to enterprise customers.