Passive Network Connection Migration for Cluster Maintenance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network communication protocols, such as SMB2, lack mechanisms to manage client connections seamlessly during node maintenance in a cluster, leading to disruptions and reduced availability when a cluster node is taken offline for maintenance.
Innovation Solution
Implement a method to non-disruptively migrate client connections from a draining node by categorizing clients into always resolving, cache resolving, and resolve once implementations, using DNS server management and client caching mechanisms to redirect connections to other nodes, ensuring minimal user disruption and maintaining cluster availability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a cluster node is taken offline for maintenance using existing network communication protocols, then the node can be maintained, but client connections are disrupted and availability is reduced
Solution Approach 1:
The system performs preliminary actions by categorizing clients and draining connections before the node is taken offline. The node identifies and categorizes clients into different types (always resolving, cache resolving, resolve once), then systematically drains connections in a controlled manner, allowing clients to be migrated to other nodes before maintenance begins, thus avoiding disruption to client availability
Solution Approach 2:
The system segments clients into different categories based on their connection resolution behavior. By dividing the client base into distinct groups (always resolving, cache resolving, resolve once), the system can apply different migration strategies to each segment, enabling maintenance while preserving connectivity for each client type through appropriate migration timing and methods
2Reliability
If client connections are migrated during node maintenance, then availability is maintained, but the system complexity increases
Solution Approach 1:
The system implements self-service by enabling clients to autonomously reconnect to different nodes after being disconnected. When a node drains connections, clients naturally attempt to reconnect, and the system's DNS infrastructure automatically directs them to available nodes, reducing the need for complex manual connection management while maintaining availability
Solution Approach 2:
The system uses DNS as an intermediary to manage connection migration. Rather than implementing complex direct connection management between clients and nodes, the DNS infrastructure acts as a mediator that automatically redirects client connection requests to available nodes, simplifying the overall connection management complexity while ensuring continuous availability
Data Source
AI summary
Techniques are provided for passive management of network connections. In an example, a cluster node determines to migrate its client connections to other nodes of its cluster. The cluster node can determine whether a client implements an always resolving policy (regarding resolving a domain name of the cluster), a cache resolving policy, or a resolve once policy. Where the node disconnects a client and it does not reconnect (because a domain name service (DNS) server no longer resolves a domain name to the node), the client implements an always resolving policy. Where the node eventually stops reconnecting, the client implements a cache resolving policy. After nodes of those types are migrated, nodes that implement a resolve once policy. The node can migrate these clients by disconnecting them in coordinating with stopping accepting connections. This can lead to non-disruptive migration of all clients.