Stateful connection reset in a storage cluster with a stateless load balancer

• US 10,425,473 B1
• Filed: 07/03/2017
• Issued: 09/24/2019
• Est. Priority Date: 07/03/2017
• Status: Active Grant

• Current Assignee: Pure Storage, Inc.
• Original Assignee: Pure Storage, Inc.
• Inventors: Patel, Jayesh, Sajeepa, Prabhath, Talayco, Daniel
• Primary Examiner(s): Nowlin, Eric
• Application Number:
  US15/641,016
• Time in Patent Office: 813 Days
• CPC Class Codes: H04L 67/10 in which an application is distributed across nodes in the network software deployment G06F8/60; multiprogramming arrangements G06F9/46
• CPC Class Codes: H04L 67/1001 for accessing one among a plurality of replicated servers
• CPC Class Codes: H04L 67/1031 Controlling of the operation of servers by a load balancer, e.g. adding or removing servers that serve requests
• CPC Class Codes: H04L 67/1097 for distributed storage of data in networks, e.g. transport arrangements for network file system [NFS], storage area networks [SAN] or network attached storage [NAS]
• CPC Class Codes: H04L 69/16 Implementation or adaptation of Internet protocol [IP], of transmission control protocol [TCP] or of user datagram protocol [UDP]
• CPC Class Codes: H04L 69/40 for recovering from a failure of a protocol instance or entity, e.g. service redundancy protocols, protocol state redundancy or protocol service redirection management of faults, events, alarms or notifications in data switching networks H04L41/06

Abstract

A method of resetting a stateful connection in a storage cluster with stateless load-balancing is provided. The method includes receiving a packet associated with a stateful connection between a client and a first storage node of the storage cluster. The method includes sending a reset packet from a second storage node to the client, through a stateless load balancer, responsive to the second storage node failing to recognize the packet associated with the stateful connection between the client and the first storage node, wherein the client resets the stateful connection responsive to receiving the reset packet. The method includes sending further reset packets from the load balancer to a plurality of storage nodes of the storage cluster, responsive to receiving, at the load balancer, the reset packet from the second storage node, wherein the plurality of storage nodes reset stateful connection information responsive to receiving the further reset packets.

Claims

1. A method of resetting a stateful connection in a storage cluster with stateless load-balancing, comprising:

receiving a packet associated with a stateful connection between a client and a first storage node of the storage cluster;

sending a reset packet from a second storage node to the client, through a stateless load balancer, responsive to the second storage node failing to recognize the packet associated with the stateful connection between the client and the first storage node, wherein the client resets the stateful connection responsive to receiving the reset packet;

sending further reset packets from the stateless load balancer to a plurality of storage nodes of the storage cluster, responsive to receiving, at the stateless load balancer, the reset packet from the second storage node, wherein the plurality of storage nodes reset stateful connection information responsive to receiving the further reset packets, the sending the further reset packets comprising;

sending a first plurality of reset packets from a first stateless load balancer to a plurality of second load balancers each associated with one of a plurality of chassis, wherein the first stateless load balancer is configurable to load balance across the plurality of chassis, and wherein each chassis has a subset of storage nodes of the storage cluster; and

sending a second plurality of reset packets from each of the plurality of second stateless load balancers to the subset of storage nodes associated with the second load balancer, responsive to receiving one of the first plurality of reset packets at each second stateless load balancer, wherein the second stateless load balancer is configurable to load balance across the subset of storage nodes.

2. The method of claim 1, further comprising:

trapping the reset packet from the second storage node at the stateless load balancer, wherein the sending the further reset packets is responsive to the trapping the reset packet.

3. The method of claim 1, wherein the sending the further reset packets from the stateless load balancer to the plurality of storage nodes comprises sending the further reset packets from the stateless load balancer to all of the storage nodes of the storage cluster except the second storage node.

4. The method of claim 1, further comprising:

establishing the stateful connection between the client and the first storage node, through packets sent through the stateless load balancer; and

updating a table in the stateless load balancer, responsive to adding a further storage node to, or removing one of the plurality of storage nodes from, the storage cluster, wherein the packet associated with the stateful connection between the client and the first storage node is received at the second storage node as a result of the updating the table.

5. The method of claim 1, wherein the sending the further reset packets comprises sending the further reset packets as out of band control packets.

6. The method of claim 1, further comprising:

injecting a packet to a kernel, in each of the plurality of storage nodes, responsive to receiving the further reset packet.

7. A tangible, non-transitory, computer-readable media having instructions thereupon which, when executed by a processor, cause the processor to perform a method comprising:

receiving a packet belonging associated with a stateful connection between a client and a first storage node of the storage cluster;

sending a reset packet from a second storage node to the client, through a stateless load balancer, responsive to the second storage node failing to recognize the packet belonging to the stateful connection between the client and the first storage node, wherein the client resets the stateful connection responsive to receiving the reset packet;

sending further reset packets from the stateless load balancer to a plurality of storage nodes of the storage cluster, responsive to trapping the reset packet from the second storage node at the stateless load balancer, wherein the plurality of storage nodes reset stateful connection information responsive to receiving the further reset packets, the sending the further reset packets comprising;

sending a first plurality of reset packets from a first stateless load balancer configurable to load balance across a plurality of chassis each having a subset of storage nodes of the storage cluster, to each of the plurality of chassis; and

sending a second plurality of reset packets from a second stateless load balancer of each of the plurality of chassis, responsive to receiving one of the first plurality of reset packets at the second stateless load balancer, the second stateless load balancer configurable to load balance across the subset of storage nodes.

8. The computer-readable media of claim 7, wherein the sending the further reset packets from the stateless load balancer to the plurality of storage nodes comprises sending a further reset packet from the stateless load balancer to each of the storage nodes of the storage cluster but not to the second storage node.

9. The computer-readable media of claim 7, wherein the method further comprises:

establishing the stateful connection between the client and the first storage node, through packets sent through the stateless load balancer; and

updating a table in the stateless load balancer, responsive to adding a further storage node to, or removing a storage node one of the plurality of storage nodes from, the storage cluster, wherein the table records paths to storage nodes present in the storage cluster for pseudorandom distribution of packets by the stateless load balancer, and wherein the packet belonging to the stateful connection between the client and the first storage node is received at the second storage node as a result of the updating the table.

10. The computer-readable media of claim 7, wherein the sending the further reset packets comprises sending out of band reset packets.

11. The computer-readable media of claim 7, wherein the method further comprises:

injecting a packet to a kernel, by an agent in each of the plurality of storage nodes, responsive to receiving the further reset packet, wherein the injected packet is sized to a window in accordance with packet monitoring by the agent, and wherein the stateful connection is a transmission control protocol (TCP) connection and the reset packet is a TCP reset packet.

12. A storage cluster with stateless load-balancing, comprising:

a plurality of storage nodes, each having one or more processors;

storage memory, distributed across the storage cluster;

a stateless load balancer, configurable to load balance packets across the plurality of storage nodes without tracking connection states;

each of the plurality of storage nodes configurable to send a reset packet to a client, through the stateless load balancer, responsive to receiving a packet associated with a stateful connection between the client and another one of the plurality of storage nodes and failing to recognize the packet, wherein the client resets the stateful connection responsive to receiving the reset packet;

the stateless load balancer configurable to send further reset packets to a subset of the plurality of storage nodes, responsive to receiving the reset packet from one of the plurality of storage nodes, the stateless load balancer comprising;

a first stateless load balancer and a plurality of second stateless load balancers;

the first stateless load balancer configurable to load balance across the plurality of second stateless load balancers;

each of the plurality of second stateless load balancers associated with a chassis having a subset of storage nodes of the storage cluster and configurable to load balance across the subset of storage nodes;

the first stateless load balancer configurable to send a first plurality of reset packets to the plurality of second stateless load balancers, responsive to trapping the reset packet from the one of the plurality of storage nodes; and

each of the plurality of second stateless load balancers configurable to send a second plurality of reset packets to the subset of storage nodes in the chassis, responsive to receiving one of the first plurality of reset packets; and

each of the plurality of storage nodes configurable to reset stateful connection information responsive to receiving one of the further reset packets.

13. The storage cluster of claim 12, further comprising:

the stateless load balancer configurable to trap the reset packet from the one of the plurality of storage nodes, triggering the further reset packets.

14. The storage cluster of claim 12, wherein the subset of the plurality of storage nodes comprises all of the plurality of storage nodes except the one of the plurality of storage nodes from which the reset packet is received by the stateless load balancer.

15. The storage cluster of claim 12, further comprising:

the stateless load balancer configurable to maintain a table having paths to the plurality of storage nodes, update the table responsive to adding a further storage node or removing one of the plurality of storage nodes from the cluster, and distribute packets in a pseudorandom manner in accordance with the table for load-balancing, wherein the failing to recognize the packet is a consequence of such a table update.

16. The storage cluster of claim 12, wherein the reset packet and each of the further reset packets is an out of band control packet, and wherein the stateful connection is a transmission control protocol (TCP) connection and the reset packet is a TCP reset packet.

17. The storage cluster of claim 12, further comprising:

each of the plurality of storage nodes configurable to have an agent inject a packet to a kernel, responsive to receiving one of the further reset packets.