CA2719673A1

CA2719673A1 - Fencing shared cluster resources

Info

Publication number: CA2719673A1
Application number: CA2719673A
Authority: CA
Inventors: Myung M. Bae; Andrew D. Buckler; Yue Fong Y. Chan; Christian M. Garcia-Arellano; Matthew A. Huras; Markus Mueller; Aamer Sachedina; Frank B. Schmuck; Yuri Volobuev
Original assignee: IBM Canada Ltd
Current assignee: IBM Canada Ltd
Priority date: 2010-11-05
Filing date: 2010-11-05
Publication date: 2011-01-18

Abstract

An illustrative embodiment of a computer-implemented process for fencing shared cluster resources in event of a possible split-brain, identifies a failing resource of a node within a set of shared resources to form an identified failing resource, fences a subset of the set of shared resources to form a winning subset of shared resources and prevents the identified failing resource from communicating with the winning subset of shared resources using a white list maintained at each database system of the winning subset of shared resources. The computer-implemented process further determines whether the identified failing resource has been cleared to form a cleared failing resource and responsive to a determination that the identified failing resource has been cleared, rejoins the cleared failing resource with the winning subset of the set of shared resources absent a re-boot of the cleared failing resource.

Description

FENCING SHARED CLUSTER RESOURCES
BACKGROUND
1. Technical Field:
[00011 This disclosure relates generally to cluster compute elements in a data processing system and more specifically to fencing shared cluster resources in event of a possible split-brain in the data processing system.

2. Description of the Related Art:
100021 In a data processing system a cluster refers to a group of compute elements that work cooperatively to achieve a computational goal, such as, providing a distributed database management system (DBMS). Compute elements cooperating in a cluster may comprise hardware in the form of individual computer systems, or host machines, a virtual machine or software components and are commonly referred to as nodes.
[00031 In cluster systems, nodes maintain coherency through the use of inter-node messaging. The communication between nodes typically uses a communication network such as Ethernet or InfiniBand network. In addition to inter-node communication, shared storage, for example, disk storage is often used in the clusters. Nodes within a cluster have access to the shared storage subsystem enabling each node to read and/or write data from/to the persistent storage.
[00041 In an embodiment of such a cluster a shared data DBMS provides a capability in which each node is a member of a DBMS instance, which is distributed. Each member typically runs on a separate host machine. Shared storage access is provided through a particular file system referred to as a cluster file system or network file system having physical storage devices accessible on a storage controller which is accessible using a network known as a storage area network (SAN) that each host machine accesses.
[00051 A shared cluster state, or simply cluster state, typically comprises persistent contents of the shared storage and in-memory contents of each node, for example, data cached in volatile memory. Data integrity of the shared cluster state is a paramount Page 1 of 39 requirement for mission critical applications. Cluster computing systems therefore have unique challenges to deal with in this regard because member nodes are separate, either logical or physical, entities. Unlike non-cluster computing systems, a cluster computing system is subject to partial failures in which a subset of nodes may fail.
[00061 A goal of a cluster computing system is therefore to maintain a high level of availability using surviving nodes in the event of a partial failure. However, the cluster computing system must ensure when a partial failure occurs no corruption of the shared cluster state is possible. A total failure of a cluster computing system results in a loss of availability and may result from many or all nodes in the cluster failing.
Typically a quorum criteria determines conditions that result in a total failure, for example, in one case, a failure of a majority of nodes results in a total failure even when a minority of nodes survive. Since total failure of a cluster computing system is considered equivalent to total failure of a non-cluster computing system, for example, a single host machine, techniques that deal with the consistency of the state of the computing system, such as log based recovery, are considered previous solutions.
[00071 A unique, infrequent case of partial failure of a cluster computing system that creates a data integrity problem to be solved by all cluster computing systems is known as a split-brain problem. To understand the split-brain scenario, it is important to understand how the presence or absence of a node in a typical cluster computing system is determined. In one example, each node is a separate machine. Nodes communicate with each other through one or more network interconnect. A special distributed software component called a cluster manager (CM) executes on all nodes and is responsible for health monitoring of the nodes. The CM causes periodic heartbeat messages to be sent to each node in the cluster. A node's health is ascertained when the node responds to the heartbeat message sent by the CM. Typically all nodes participate in this type of monitoring. A lack of response to heartbeat messages will eventually cause the CM to consider a non-responsive node as OFFLINE. Nodes that respond normally to heartbeat messages are considered ONLINE.
[00081 The split-brain problem arises when a partition or a failure in the network inter-connect, used for the exchange of heart-beat messages, causes one or more nodes to be CA920100015CA l Page 2 of 39 considered OFFLINE even though the nodes are still operating and are simply not in communication with the other nodes in the cluster. The failure just described may cause a cluster computing system to be partitioned into one or more sub-clusters.
Each sub-cluster is unable to determine the status of peer nodes in another sub-cluster because nodes within each sub-cluster cannot communicate with nodes in other sub-cluster(s).
When one or more nodes fails, for example due to a hardware failure/power outage, the symptoms of loss of heart-beat response perceived by the cluster manager is the same as in the case of a partition or failure in the network inter-connect.
Accordingly a cluster computing system cannot differentiate between nodes that are unresponsive to heartbeat messages because they have actually failed, or have been split from the main cluster into a sub-cluster.

100091 Sub-clusters can result from the rare event of a network partition in which the `brain' of the cluster can be thought of as having been `split'. Without additional protocols for coherency, data to resolve such a split-brain scenario corruption is possible as both sub-clusters might otherwise consider that nodes in the other sub-cluster have failed. Split-brain resolution is a term given to a protocol used to ensure only one sub-cluster has the ability to modify the shared cluster state, such as the contents of the shared storage.

[00101 A split-brain resolution protocol typically involves the processes of quorum determination and fencing of shared resources. To ensure only one (possible) sub-cluster continues as the brain of the cluster computing system, a quorum protocol is run by each sub-cluster when the loss of heartbeats to/from one or more other nodes occurs. The quorum protocol ensures that only one of a multiple of sub-clusters will have quorum and thus be able to continue. In one example, each sub-cluster elects one node as a quorum master. All nodes in the specific sub-cluster have a vote, which is sent to the quorum master. The sub-cluster obtaining more than one half of the total votes in the cluster computing system is considered the winner of the quorum. A sub-cluster (when one exists) having a minority of the votes is the losing sub-cluster.
[00111 The sub-cluster determined to have the quorum should not continue and assume the shared cluster state will be safe from corruption by the losing sub-cluster. A major Page 3 of 39 problem is one of timing for both the winning sub-cluster and the losing sub-cluster. A
losing sub-cluster may not determine that it is the loser until some time after the winning sub-cluster has determined that it is the winner. Before a winning sub-cluster can continue to modify a shared cluster state, and in an example, execute recovery protocols to recover from the loss of other nodes, the winning sub-cluster must ensure that nodes in the losing sub-cluster cannot modify the shared cluster state.
[00121 An additional fencing protocol needs to be invoked. A fencing protocol, when successful, ensures nodes in a sub-cluster that lost quorum cannot modify shared cluster state any longer. Existing algorithms for fencing of shared cluster resources tend to concentrate heavily on the fencing of shared storage. While several different algorithms exist, with advantages and disadvantages for each, one thing that is common to all of is the losing sub-cluster nodes have to reboot or restart.
100131 Typical algorithms for fencing include a self-initiated reboot, a hardware assisted reboot and resource fencing with a self-initiated reboot. A self initiated reboot is a crude algorithm wherein the winning sub-cluster of the quorum waits for some pre-determined amount of time, typically defined as FENCE TIMEOUT, for some number of seconds usually in a duration specified as 30 seconds, or 60 seconds after the loss of heartbeat has been determined. The losing sub-cluster / nodes are responsible for rebooting/restarting themselves once they determine themselves to be the losers.
[00141 After the reboot/restart, the nodes in the losing sub-cluster go through a cluster JOIN protocol that only allows them to join an existing sub-cluster or form a new sub-cluster when a quorum exists in that sub-cluster. Thus the nodes do not have access to the shared cluster state until a time that they join or create a cluster, which has quorum.
[00151 The value of FENCE TIMEOUT is set or computed to be a large enough value to ensure a high probability that a losing sub-cluster in a split-brain scenario will determine it is the loser and will voluntarily reboot or restart associated nodes thus preventing the nodes from modifying the shared cluster state.
[00161 A hardware assisted reboot is often referred to as shoot the other node in the head (STONITH). In the hardware scenario example, each node is a separate machine that has access to a power switch of another node. An intelligent power switch, which is Page 4 of 39 accessible via a private network, can serve this same purpose. A hardware management console can also serve this same purpose.
[00171 Hardware assisted fencing protocol is less crude than a self-initiated reboot because the winning node accesses the power switch of the losing node and power cycles the losing node(s). The hardware-assisted technique ensures the winning side that the losing sub-cluster, in the event of a split-brain, will not be able to corrupt shared cluster state because the nodes have been power cycled.
[00181 Resource fencing allows a winning sub-cluster that retains quorum to fence off specific shared resources from the nodes of a losing sub-cluster. Resource fencing requires specific protocols for shared cluster resources. For example the Small Computer Systems Interface-3 Persistent Reserve (SCSI-3 PR) protocol can be used to fence one or more nodes in a losing sub-cluster from the shared storage of the cluster computing system. When the nodes in the losing sub-cluster determine they no longer have quorum, the losing nodes typically reboot or restart themselves through a self initiated reboot and re-join the cluster through a normal JOIN protocol after successful reboot or restart.
[00191 Resource fencing, as previously practiced, is typically followed by, and dependant on, errant node(s) rebooting once the errant nodes have determined they have been fenced. The self-initiated reboot or restart is performed in previous solutions because fencing shared resources other than shared storage is not practical.
While a mature protocol, such as SCSI-3 PR exists for shared storage fencing, such protocols do not exist to fence off inter-node communication to ensure when the network partition heals after the winning sub-cluster has considered itself the winner and presumed the loser fenced that nodes in the losing sub-cluster cannot modify in-memory state of nodes in the winning sub-cluster by sending messages [0020] When inter-node communication is performed through a remote direct memory access (RDMA) capable transport, a node may have the ability to directly modify the contents of the memory of another node without requiring any messages that need to be responded to by the target node. A reboot or restart of node(s) in a losing sub-cluster is performed to prevent corruption problems that could result from a partitioned network healing. Such problems can occur when processes or threads are still alive on the node(s) Page 5 of 39 that lost quorum, enabling the nodes able to communicate with processes or threads on other nodes thereby corrupting in-memory state of the nodes. The corruption is particularly likely when RDMA is being used for inter-node communication because a node may have write access to memory of another node.
[0021] A reboot or restart is a rudimentary way to ensure any node in a losing sub-cluster is clean, for example, previous processes are killed, before the node rejoins the cluster. The reboot or restart forces evicted nodes to go through a normal JOIN protocol for the cluster and effectively reduces the rejoin case to reintegration of a rebooted or restarted node into a operational cluster.
[0022] Self-initiated reboot does not provide a guarantee nodes in the losing sub-cluster will reboot before a winning sub-cluster considers the nodes to be dead, and thus fenced from the shared cluster resources including shared disks. While unlikely, an operating system or a hardware bug may cause a node to effectively freeze for some amount of time just before issuing an input/output request to disk. During a predetermined time in which the machine is frozen and healthy nodes consider the errant node as dead, the healthy nodes might start recovery on cluster resources, for example shared disk that the errant node was using. The errant node might thaw and corrupt the shared resources by writing out changes to disk, before the errant node discovers it has been evicted from the cluster.
[0023] Hardware assisted reboot typically provides better guarantees of data consistency than self initiated reboot because the winning sub-cluster can confirm power has been cut to nodes in the losing sub-cluster, before assuming shared resources are not in use by the nodes of the losing sub-cluster. Small data consistency holes remain when this algorithm is used. When the errant node has issued a write operation, which is in-flight, the write operation may still be within the storage sub-system. A hardware assisted reboot does not guarantee that such operations may not ultimately arrive at the specific disk after the power cycling of the node has occurred and corrupt data.
[0024] Resource fencing typically does not include fencing off all network connectivity.
When RDMA is being performed between nodes, corruption of shared cluster state (volatile memory) is possible in the event the network partition heals before the errant node can be readmitted to the cluster through the normal JOIN protocols. A
technique Page 6 of 39 such as closing the ports of the errant node on a network switch is not robust because there is no dynamic mechanism to identify which node uses which port on a network switch.

Page 7 of 39 SUMMARY
100251 According to one embodiment, a computer-implemented process for fencing shared cluster resources in an event of a possible split-brain, identifies a failing resource of a node within a set of shared resources to form an identified failing resource, fences a subset of the set of shared resources to form a winning subset of shared resources and prevents the identified failing resource from communicating with the winning subset of shared resources using a white list maintained at each database system of the winning subset of shared resources. The computer-implemented process further determines whether the identified failing resource has been cleared to form a cleared failing resource and responsive to a determination that the identified failing resource has been cleared, rejoins the cleared failing resource with the winning subset of the set of shared resources absent a re-boot of the cleared failing resource.
[00261 According to another embodiment, a computer program product for fencing shared cluster resources in an event of a possible split-brain comprises a computer recordable-type media containing computer executable program code stored thereon.
The computer executable program code comprises computer executable program code for identifying a failing resource of a node within a set of shared resources to form an identified failing resource, computer executable program code for fencing a subset of the set of shared resources to form a winning subset of shared resources, computer executable program code for preventing the identified failing resource from communicating with the winning subset of shared resources using a white list maintained at each database system of the winning subset of shared resources, computer executable program code for determining whether the identified failing resource has been cleared to form a cleared failing resource, and computer executable program code responsive to a determination that the identified failing resource has been cleared, for rejoining the cleared failing resource with the winning subset of the set of shared resources absent a re-boot of the cleared failing resource.

Page 8 of 39 [00271 According to another embodiment, an apparatus for fencing shared cluster resources in an event of a possible split-brain, comprises a communications fabric, a memory connected to the communications fabric, wherein the memory contains computer executable program code, a communications unit connected to the communications fabric, an input/output unit connected to the communications fabric, a display connected to the communications fabric and a processor unit connected to the communications fabric. The processor unit executes the computer executable program code to direct the apparatus to identify a failing resource of a node within a set of shared resources to form an identified failing resource, fence a subset of the set of shared resources to form a winning subset of shared resources, prevent the identified failing resource from communicating with the winning subset of shared resources using a white list maintained at each database system of the winning subset of shared resources, determine whether the identified failing resource has been cleared to form a cleared failing resource and responsive, to a determination that the identified failing resource has been cleared, rejoin the cleared failing resource with the winning subset of the set of shared resources absent a re-boot of the cleared failing resource.

Page 9 of 39 BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWINGS
[0028] For a more complete understanding of this disclosure, reference is now made to the following brief description, taken in conjunction with the accompanying drawings and detailed description, wherein like reference numerals represent like parts.
[0029] Figure 1 is a block diagram of an exemplary network of data processing systems operable for various embodiments of the disclosure;
[0030] Figure 2; is a block diagram of a data processing for various embodiments of the disclosure;

[0031] Figure 3; is a block diagram of a cluster management system in accordance with various embodiments of the disclosure;

[0032] Figure 4 is a block diagram of an exemplary white list used with the cluster management system of Figure 3, in accordance with one embodiment of the disclosure;
[0033] Figure 5 is a flowchart of a process used in the cluster management system of Figure 3, in accordance with one embodiment of the disclosure;
[0034] Figure 6 is a flowchart of an IGNORE message handling process used in the process of Figure 5, in accordance with one embodiment of the disclosure; and [0035] Figure 7 is a flowchart of a rejoin process used in the cluster management system of Figure 3, in accordance with one embodiment of the disclosure.

Page 10 of 39 DETAILED DESCRIPTION

[00361 Although an illustrative implementation of one or more embodiments is provided below, the disclosed systems and/or methods may be implemented using any number of techniques. This disclosure should in no way be limited to the illustrative implementations, drawings, and techniques illustrated below, including the exemplary designs and implementations illustrated and described herein, but may be modified within the scope of the appended claims along with their full scope of equivalents.
[00371 As will be appreciated by one skilled in the art, aspects of the present disclosure may be embodied as a system, method or computer program product. Accordingly, aspects of the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment (including firmware, resident software, micro-code, etc.) or an embodiment combining software and hardware aspects that may all generally be referred to herein as a "circuit," "module," or "system." Furthermore, aspects of the present invention may take the form of a computer program product embodied in one or more computer readable medium(s) having computer readable program code embodied thereon.

[00381 Any combination of one or more computer-readable medium(s) may be utilized.
The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CDROM), an optical storage device, or a magnetic storage device or any suitable combination of the foregoing. In the context of this document, a computer-readable storage medium may be Page 11 of 39 any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.
[0039] A computer-readable signal medium may include a propagated data signal with the computer-readable program code embodied therein, for example, either in baseband or as part of a carrier wave. Such a propagated signal may take a variety of forms, including but not limited to electro-magnetic, optical or any suitable combination thereof.
A computer readable signal medium may be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.
[0040] Program code embodied on a computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wire line, optical fiber cable, RF, etc. or any suitable combination of the foregoing.
[0041] Computer program code for carrying out operations for aspects of the present disclosure may be written in any combination of one or more programming languages, including an object oriented programming language such as JavaTM, Smalltalk, C++, or the like and conventional procedural programming languages, such as the "C"
programming language or similar programming languages. Java and all Java-based trademarks and logos are trademarks of Sun Microsystems, Inc., in the United States, other countries or both. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server.
In the latter scenario, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider).
[0042] Aspects of the present disclosure are described below with reference to flowchart illustrations and/or block diagrams of methods, apparatus, (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of Page 12 of 39 blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions.
100431 These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks.
[00441 These computer program instructions may also be stored in a computer readable medium that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable medium produce an article of manufacture including instructions which implement the function/act specified in the flowchart and/or block diagram block or blocks.
[00451 The computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks.
[00461 With reference now to the figures and in particular with reference to Figures 1-2, exemplary diagrams of data processing environments are provided in which illustrative embodiments may be implemented. It should be appreciated that Figures 1-2 are only exemplary and are not intended to assert or imply any limitation with regard to the environments in which different embodiments may be implemented. Many modifications to the depicted environments may be made.
[00471 Figure 1 depicts a pictorial representation of a network of data processing systems in which illustrative embodiments may be implemented. Network data processing system 100 is a network of computers in which the illustrative embodiments may be implemented. Network data processing system 100 contains network 102, which is the medium used to provide communications links between various devices and Page 13 of 39 computers connected together within network data processing system 100.
Network 102 may include connections, such as wire, wireless communication links, or fiber optic cables.

[0048] In the depicted example, server 104 and server 106 connect to network 102 along with storage unit 108. In addition, clients 110, 112, and 114 connect to network 102.
Clients 110, 112, and 114 may be, for example, personal computers or network computers. In the depicted example, server 104 provides data, such as boot files, operating system images, and applications to clients 110, 112, and 114.
Clients 110, 112, and 114 are clients to server 104 in this example. Network data processing system 100 may include additional servers, clients, and other devices not shown.
[0049] In the depicted example, network data processing system 100 is the Internet with network 102 representing a worldwide collection of networks and gateways that use the Transmission Control Protocol/Internet Protocol (TCP/IP) suite of protocols to communicate with one another. At the heart of the Internet is a backbone of high-speed data communication lines between major nodes or host computers, consisting of thousands of commercial, governmental, educational and other computer systems that route data and messages. Of course, network data processing system 100 also may be implemented as a number of different types of networks, such as for example, an intranet, a local area network (LAN), or a wide area network (WAN). Figure 1 is intended as an example, and not as an architectural limitation for the different illustrative embodiments.
[0050] Turning now to Figure 2 a block diagram of an exemplary data processing system operable for various embodiments of the disclosure is presented. In this illustrative example, data processing system 200 includes communications fabric 202, which provides communications between processor unit 204, memory 206, persistent storage 208, communications unit 210, input/output (I/O) unit 212, and display 214.
[0051] Processor unit 204 serves to execute instructions for software that may be loaded into memory 206. Processor unit 204 may be a set of one or more processors or may be a multi-processor core, depending on the particular implementation. Further, processor unit 204 may be implemented using one or more heterogeneous processor systems in which a main processor is present with secondary processors on a single chip.
As another Page 14 of 39 illustrative example, processor unit 204 may be a symmetric multi-processor system containing multiple processors of the same type.
100521 Memory 206 and persistent storage 208 are examples of storage devices 216. A
storage device is any piece of hardware that is capable of storing information, such as, for example without limitation, data, program code in functional form, and/or other suitable information either on a temporary basis and/or a permanent basis. Memory 206, in these examples, may be, for example, a random access memory or any other suitable volatile or non-volatile storage device. Persistent storage 208 may take various forms depending on the particular implementation. For example, persistent storage 208 may contain one or more components or devices. For example, persistent storage 208 may be a hard drive, a flash memory, a rewritable optical disk, a rewritable magnetic tape, or some combination of the above. The media used by persistent storage 208 also may be removable.
For example, a removable hard drive may be used for persistent storage 208.
[00531 Communications unit 210, in these examples, provides for communications with other data processing systems or devices. In these examples, communications unit 210 is a network interface card. Communications unit 210 may provide communications through the use of either or both physical and wireless communications links.
[00541 Input/output unit 212 allows for input and output of data with other devices that may be connected to data processing system 200. For example, input/output unit may provide a connection for user input through a keyboard, a mouse, and/or some other suitable input device. Further, input/output unit 212 may send output to a printer.
Display 214 provides a mechanism to display information to a user.
[00551 Instructions for the operating system, applications and/or programs may be located in storage devices 216, which are in communication with processor unit through communications fabric 202. In these illustrative examples the instructions are in a functional form on persistent storage 208. These instructions may be loaded into memory 206 for execution by processor unit 204. The processes of the different embodiments may be performed by processor unit 204 using computer-implemented instructions, which may be located in a memory, such as memory 206.

Page 15 of 39 [0056] These instructions are referred to as program code, computer usable program code, or computer readable program code that may be read and executed by a processor in processor unit 204. The program code in the different embodiments may be embodied on different physical or tangible computer readable media, such as memory 206 or persistent storage 208.
[0057] Program code 218 is located in a functional form on computer readable media 220 that is selectively removable and may be loaded onto or transferred to data processing system 200 for execution by processor unit 204. Program code 218 and computer readable media 220 form computer program product 222 in these examples. In one example, computer readable media 220 may be in a tangible form, such as, for example, an optical or magnetic disc that is inserted or placed into a drive or other device that is part of non-transitory, persistent storage 208 for transfer onto a storage device, such as a hard drive that is part of persistent storage 208. In a tangible form, computer readable media 220 also may take the form of a persistent storage, such as a hard drive, a thumb drive, or a flash memory that is connected to data processing system 200. The tangible form of computer readable media 220 is also referred to as computer recordable storage media. In some instances, computer readable media 220 may not be removable.
[0058] Alternatively, program code 218 may be transferred to data processing system 200 from computer readable media 220 through a communications link to communications unit 210 and/or through a connection to input/output unit 212.
The communications link and/or the connection may be physical or wireless in the illustrative examples. The computer readable media also may take the form of non-tangible media, such as communications links or wireless transmissions containing the program code.
[0059] In some illustrative embodiments, program code 218 may be downloaded over a network to persistent storage 208 from another device or data processing system for use within data processing system 200. For instance, program code stored in a computer readable storage medium in a server data processing system may be downloaded over a network from the server to data processing system 200. The data processing system providing program code 218 may be a server computer, a client computer, or some other device capable of storing and transmitting program code 218.

Page 16 of 39 100601 According to an illustrative embodiment using data processing system 200 of Figure 2 as an example of server 104 of network of data processing systems 100 of Figure 1, processor unit 204 executes a computer-implemented process for fencing shared cluster resources in an event of a possible split-brain. Processor unit identifies a failing resource of a node within a set of shared resources to form an identified failing resource, and fences a subset of the set of shared resources to form a winning subset of shared resources. Processor unit 204 prevents the identified failing resource from communicating through communications unit 210 and network 102 of Figure 1 with the winning subset of shared resources using a white list maintained at each database system of the winning subset of shared resources. Processor unit further determines whether the identified failing resource has been cleared to form a cleared failing resource and responsive to a determination that the identified failing resource has been cleared, rejoins the cleared failing resource with the winning subset of the set of shared resources absent a re-boot of the cleared failing resource.
[00611 In an alternative embodiment, program code 218 containing the computer-implemented process may be stored within computer readable media 220 as computer program product 222. In another illustrative embodiment, the process for fencing shared cluster resources in an event of a possible split-brain may be implemented in an apparatus comprising a communications fabric, a memory connected to the communications fabric, wherein the memory contains computer executable program code, a communications unit connected to the communications fabric, an input/output unit connected to the communications fabric, a display connected to the communications fabric, and a processor unit connected to the communications fabric. The processor unit of the apparatus executes the computer executable program code to direct the apparatus to perform the process.

[00621 With reference to Figure 3 a block diagram of a cluster management system in accordance with various embodiments of the disclosure is presented. Cluster management system 300 is an example of a cluster management system in accordance with various embodiments of the disclosure.

Page 17 of 39 100631 Cluster management system 300 comprises a number of components including cluster manager 302, database systems 304, cluster file system 306, event monitor 308, and a set of nodes, such as node 310. Cluster management system 300 also comprises typical components of a data processing system of sufficient operational complexity to support a management system, such as that of data processing system 200 of Figure 2.
[00641 Cluster manager 302 provides a management capability using commands including start, stop, monitor and cleanup commands to control activity in the managed cluster system. Cluster manager uses event monitor 308 to maintain an awareness of the state of activity within cluster management system 300. Events such as the starting and stopping of nodes are of interest to cluster manager 302.
[00651 Cluster manager 302 also contains an event notifier providing a capability to send messages to database systems 304, and cluster file system 306 regarding specific events. Additionally a sequence number generator for generating a sequence number associated with a message to a database system 304 that typically causes a whitelist state change is included. A sequence number is used as additional verification when determining whether to accept a message at a node. Cluster manager 302 is also responsible for providing a cleaning capability to ensure when a node is recovered the node is in condition to rejoin the cluster.
100661 Database systems 304 represents one or more databases typically provided in a distributed network of database management systems. Each database management system interconnected within database systems 304 manages a set of nodes such as node 310 forming a set of shared resources that comprise data of interest within cluster management system 300. Database systems 304 provide a capability to distribute event information among nodes associated with the system and to interface with cluster manager 302 on behalf of the nodes. Each database system in database systems 304, maintains a copy of white list 312. A "node," such as node-1 310, represents a host, a machine, or an operating system (OS) image on which a database system operates. The database systems, such as database systems 304, track and update whitelists in response to messages from cluster manager 302.

Page 18 of 39 100671 Cluster file system 306 physically manages various storage devices comprising cluster management system 300. For example, cluster file system 306 provides a capability to use SCSI-3 PR storage management protocol to manage shared devices of cluster management system 300. Cluster file system 306 responds to requests from cluster manager 302 to isolate or fence devices as required.
100681 White list 312 is used to aid on controlling communication among database systems 304. Each database system in the set of database systems, such as database systems 304, maintains a respective copy of white list 312. White list 312 contains state information describing a communication capability for each other connected database system within the set of database systems. White list 312 is further explained by example using Figure 4.
100691 In an example embodiment, cluster manager 302 performs a health check using predefined criteria before adding a previously evicted node, such as node 310, back into the cluster rather than simply allowing the previously evicted node to re-join once network connectivity is restored. For example, predefined criteria may comprise establishing a determination that the previously evicted node is not paging, or has processor utilization below a threshold. Performing the health check proactively ensures a node, determined to be unhealthy, is not re-introduced into a cluster after having once been evicted.
[00701 Additionally, a process that includes a heavier weight eviction requiring a reboot before reintegration may be performed when a node is declared unhealthy due to the predefined criteria or when a node is integrated back into the cluster but evicted again within a predefined duration. For example, when a node fails N times within T
seconds, then the node is marked as unhealthy and no attempt to restart is performed until an administrator performs diagnostics and clears the alert.
100711 An embodiment of cluster management system 300 provides a capability to fence shared cluster resources in the event of a possible split-brain condition.
Both shared disk and a memory state of the cluster are resources that are fenced using the disclosed process of the embodiment. Typically the process is completed in a minimal amount of time without requiring rebooting or restarting of the errant node of a losing sub-cluster, Page 19 of 39 either for the purpose of fencing the node or to re-integrate the node into the cluster computing system once the network partition that caused the split-brain to occur has healed. The disclosed process typically reduces the total time required for fencing of errant nodes and re-integration of the errant nodes into the cluster.
[0072] When cluster manager 302 has determined a loss of heartbeat from one or more nodes in the network, shared disk resources are fenced off by a winning sub-cluster to prevent nodes in the losing sub-cluster, which have stopped sending heartbeat messages, from updating the shared disks. Disk fencing is achieved using the SCSI-3 Persistent Reserve protocol in one embodiment. Other embodiments could restrict access to the fiber channel Storage Area Network by closing ports on one or more fiber channel switches providing node connectivity to the shared disks.
[0073] In addition to disk fencing, the described process of cluster management system 300 fences communication from the database systems on errant nodes to the database systems on nodes in the winning sub-cluster by using a cluster membership protocol to revoke the ability of the errant node to communicate with database systems on nodes of the winning sub-cluster.
[0074] With reference to Figure 4, a block diagram of an exemplary white list used with the cluster management system of Figure 3, in accordance with one embodiment of the disclosure is presented. White list 402, representing the database system on node A, white list 404 representing the database system on node B, and white list 406 representing the database system on node C, are further examples of white list 312 of Figure 3.
[0075] A basic structure for cluster membership protocol is a collection of state information called a white list as shown in white lists 400. Each database system has a capability to track a respective white list containing an entry for every other database system with which the instant node will need to communicate. For example, a white list entry, represented by white list 402 on the database system of node A contains a white list state that governs handling of communications with a database system of the entry.
Header 408 defines a set of columns including an identifier for a database system, a state and a sequence number of each entry. In the example of white list 402, row 410 indicates Page 20 of 39 a database system value of B, a specific state of ACCEPT and a sequence number value of 3. In a similar manner row 412 indicates a database system value of C, a specific state of ACCEPT and a sequence number value of 4.
100761 In the example of white list 404, representing the database system on node B, row 414 indicates a database system value of A, a specific state of ACCEPT and a sequence number value of 5. In a similar manner row 416 indicates a database system value of C, a specific state of ACCEPT and a sequence number value of 4. In the example of white list 406, representing the database system on node C, row 418 indicates a database system value of A, a specific state of ACCEPT and a sequence number value of 5. In a similar manner row 420 indicates a database system value of B, a specific state of ACCEPT and a sequence number value of 3.
[0077] A white list state of REJECT indicates that no communication is allowed between the respective database systems for which the entry applies. The white list entry also contains a sequence number used to guard against incorrect changes in state due to outdated notifications. When a notification, received by a database system, has a lower sequence number than a respective sequence number in the white list entry for a respective database system the incoming notification is discarded. Discarding is necessary because notifications can be arbitrarily reordered by network effects, or scheduling latencies related to input output operations or the operating system.
[0078] White lists 400, logically represent a set of tuples including a (databaseSystemlD, whitelistState, and a sequenceNumber). Each database system tracks a single entry representing a latest state, as determined by a sequence number, for each database system in the cluster. Arrival and departure notification protocols are required to change the white list state in all database systems on nodes that have not been fenced, excluding nodes that are currently shut down or failed. The state information of the white list indicates communication is allowed with a database system (arrival) or disallowed with a failed or bad database system (departure). In the example of white lists 400, all database systems currently list each other with a state of ACCEPT, indicating a normal runtime operation. Also in the example, white list 404 representing the database system on node B and white list 406 representing the database system on node C list the database Page 21 of 39 system on node A with sequence number 5, indicating that only a notification carrying a sequence number value of 6 or greater would be able to change the ACCEPT state for the database system on node A.
[0079] Basic states in a white list include ACCEPT, indicating normal runtime communications, REJECT, indicating that all communications are blocked and an intermediate state of IGNORE, used to avoid potential dangerous intermediate states.
[0080] The IGNORE state, indicates communication going to a database system should hang, and communication from the database system should be ignored. The IGNORE
state effectively freezes any interactions with a respective database system until a REJECT notification is received, at which point the operation driving the interaction can progress through normal failure handling. A REJECT notification will typically not be issued until all database systems in a cluster have acknowledged an IGNORE
state, to prevent an occurrence of any potential intermediate problem states.
[0081] With reference to Figure 5 a flowchart of a process used in the cluster management system of Figure 3, in accordance with one embodiment of the disclosure is presented. Process 500 is an example of a cluster management process implemented within cluster management system 300 of Figure 3.
[0082] Process 500 starts (step 502) and monitors resources of the cluster management system (step 504). Resources being monitored include nodes of the distributed database management system comprising the cluster file system of the example. Process determines whether a resource failure has occurred (step 506). A resource failure is not limited to a hard failure of a node but may also include a network interconnect outage that prevents a node from communicating with another node in the cluster. When a determination is made that a resource failure has occurred, a "yes" result is obtained.
When a determination is made that a resource failure has not occurred, a "no"
result is obtained.

[0083] When a "no" result is obtained in step 506 process 500 loops back to step 504 to continue to monitor as before. When a "yes" is obtained in step 506, process identifies a failing resource to form identified failing resources (step 508).
The identified Page 22 of 39 failing resources may be one or more failing resources, also referred to as errant nodes or resources.
[00841 Process 500 fences the shared resources to identified failing resources from updating the shared resources (step 510). Shared resources include data contained on shared devices such as nodes of the database management systems as well as in-memory data contained at node locations. Process 500 further prevents the identified failing resources from communicating with the shared resources (step 512).
[00851 As previously described, each database system maintains a white list, such as white list 312 of Figure 3, based on event notification messages the database system receives from the cluster manager, such as cluster manager 302 of Figure 3.
The cluster manager generates an ARRIVE event notification when a database system arrives in the cluster. Database systems track and update whitelists in response to messages from a cluster manager. When a node fails messages are sent for each instance of the database system that was running on the node at the time of failure.
100861 For example using the scenario of white lists 400 of Figure 4, a target database system is not allowed to communicate with sending database systems in the cluster until the target database system has received an ARRIVE notification for the specific sending database system. When a database system receives a valid ARRIVE notification for the sending database system , the receiving database system marks the arriving database system as ACCEPT in a white list. For example, the white list entry for an arriving database system of node A is marked by a database system of node B as ACCEPT
in the white list of database system of node B, signaling database system of node B
will accept connections from database system of node A. New connections from the database system of node A are however not accepted by any other database system in the cluster unless node A is listed as ACCEPT in a white list of a receiving database system.
[0087) When a node fails, or is perceived to have failed because of a loss of a detected heartbeat, the cluster manager begins a two-phase departure protocol to inform other database systems on other nodes to fence communications to and from the departed database system. Process 500 sends IGNORE signal or notifications to non-failing resources of all other database systems to update each respective database system white Page 23 of 39 list (step 514). Upon receipt of the IGNORE signal notification the database systems modify the status of the departed node in respective white list entries to IGNORE and accordingly ignore any communications from the departed node.
[00881 Process 500 determines whether all non-failing resources have updated status of identified failing resources (step 516). When a determination is made that all non-failing database systems have updated status of identified failing resources, a "yes"
result is obtained. When a determination is made that not all non-failing database systems have updated status of identified failing resources, a "no" result is obtained.
When a "no"
result is obtained in step 516, process 500 loops back to perform step 516 over again.
When a "yes" result is obtained in step 516, the IGNORE notification is successful on all surviving database systems, process 500, through the cluster manager, sends REJECT
signal notifications to all surviving database systems of the non-failing nodes, to update each respective database system white list (step 518). Upon receipt of a REJECT signal notification each surviving database system modifies the status of the departed node in the white list entries of each respective database system to REJECT. Each surviving node is thus enabled to reject any communications from the errant or departed node.
[00891 Before acknowledging successful processing of REJECT, each recipient database system takes local actions to ensure no new communication can be initiated to the errant node on existing connections and no new connections will be allowed from the errant node. In one embodiment, as part of REJECT processing, a receiving database system can synchronously close all existing communication connections between the errant node and the receiving node. Unfortunately this requires all in-flight communications to the errant node to be completed and drained and 'can thereby result in REJECT processing taking an arbitrary amount of time on the database system processing the REJECT. The process delay is undesirable because recovery of an errant database system, for example, freeing locks that were held, and reversing uncommitted changes processed by the errant node to the shared cluster state, cannot be started until confirmation that the errant node has been fenced which includes marking REJECT in all surviving database systems.

Page 24 of 39 [00901 In another embodiment, the receiving database system of a REJECT
notification may simply ensure that a most recent white list state is checked before any communication is being initiated to any node. Any in-flight communications will be detected upon completion, even when the communications completed successfully, and the results of the communication will be discarded because the target database system was in REJECT state while the communication was in flight. In such an embodiment, the most current copy of the white list state is verified to ensure the target node is in ACCEPT state before the communication is initiated and remains in ACCEPT state after the communication is received.
[00911 In another embodiment, to allow a REJECT to complete before in-flight communications are completed to the errant node, each database system must ensure that any buffers or memory used for communication between a departed node and the database system, will not be re-used or acted upon in a way to corrupt the shared cluster state until it is known that communication connections to the errant node have been closed.
[00921 Process 500 determines whether failing resources, such as failing or errant nodes, have been cleared to form cleared failing nodes (step 520). When a determination is made that processes or threads running from a prior instance of the departed node, which participated in the clustered application, are cleaned, a "yes" result is obtained. When a determination is made that processes or threads running from the prior instance of the departed node, which participated in the clustered application, are not cleaned, a "no"
result is obtained.
[00931 The cleaning procedure ensures the departed node can start in a clean state and re-join the cluster but does not require a full reboot of the departed node because the cleaning process simply targets the specific process or threads. When a "no"
result is obtained in step 520, cleanup has not performed successfully, the departed node is still fenced from the shared disk, and the departed node is still registered as REJECT in the respective white lists of other surviving nodes. Process 500 loops back to perform step 520 again.

Page 25 of 39 [00941 When a "yes" result is obtained in step 520, the cleanup has been successfully performed; process 500 rejoins the cleared failing resources, the departed node, without a full re-boot (step 522). The departed node is unfenced from the shared disks and the cluster manager restarts the software processes or threads that are part of the clustered application. Process 500 returns to monitor resources in step 504 as before.
[00951 With reference to Figure 6 a flowchart of an ignore message handling process used in the process of Figure 5, in accordance with one embodiment of the disclosure is presented. Process 600 is an example of an ignore message handling process including sequence number verification using a sequence number of white lists 400 of Figure 4.
100961 Process 600 starts (step 602) and receives a message, or notification at a target node from a sending node (step 604). Process 600 determines whether the message is an IGNORE message (step 608). When a determination is made that the message is an IGNORE message, a "yes" result is obtained. When a determination is made that the message is not an IGNORE message, a "no" result is obtained.
[00971 When a "yes" result is obtained in step 606, process 600 determines whether the sending node has an ACCEPT status at the target node (step 608). A thread performing an in-flight communication might have not received a processor context during a time in which a state of an errant node was modified to REJECT and the errant node restarted, perhaps for recovery purposes by the cluster manager, or rejoined the cluster because the network partition healed.
[00981 When a determination is made that the sending node has an ACCEPT status at the target node, a "yes" result is obtained. When a determination is made that the sending node does not have an ACCEPT status at the target node, a "no" result is obtained. When a "no" result is obtained in step 606, process 600 discards the message or notification (step 614) with process 600 terminating thereafter (step 616). When a "yes"
result is obtained in step 608, process 600 determines whether a sequence number associated with the sending node in the white list at the target node greater than or equal to a sequence number in the received message (step 610). A check for ACCEPT once the communication has been completed is not sufficient and must be combined with a confirmation the sequence number for the target node on the white list has not changed.

Page 26 of 39 [0099] When a determination is made that the sequence number associated with the sending node in the white list at the target node greater than or equal to a sequence number of the received message, a "yes" result is obtained. When a determination is made that the sequence number associated with the sending node in the white list at the target node not greater than or equal to a sequence number of the received message, a "no" result is obtained. When a "no" result is obtained in step 610, process 600 discards the message or notification (step 614) with process 600 terminating thereafter (step 616).
When a "yes" result is obtained in step 610, process 600 accepts the message or notification (step 612) and terminates thereafter (step 616).
[00100] Returning to step 606, when a "no" result is obtained in step 606, process 600 determines whether the message is a REJECT message (step 618). When a determination is made that the message is a REJECT message, a "yes" result is obtained.
When a determination is made that the message is not a REJECT message, a "no"
result is obtained. When a "yes" result is obtained in step 618, process 600 verifies the current state is set to IGNORE (step 620). Process 600 loops back to step 608. When a "no"
result is obtained in step 618, process 600 terminates (step 616).
[00101] With reference to Figure 7 a flowchart of a rejoin process used in the cluster management system of Figure 3, in accordance with one embodiment of the disclosure is presented. Process 700 is an example of a rejoin operation in combination with a clean operation used in the cluster management system of Figure 3.
[00102] In the example of process 700, to allow a REJECT to complete before in-flight communications are completed to the errant node, each node must ensure that any buffers or memory used for communication between a departed node and the node, will not be re-used or acted upon in a way to corrupt the shared cluster state until it is known that communication connections to the errant node have been closed.
[00103] Process 700 starts (step 702) and initiates a clean process for a failed node also referred to as a departed node or errant node (step 704). When the network-split heals, the cluster manager first verifies that any processes or threads running from the prior instance of the departed node, which participate in the clustered application, are cleaned or removed. Process 700 determines whether the clean process for a process or thread at a Page 27 of 39 failed node is complete (step 706). When a determination is made that the clean process for a process or thread at a failed node is complete, a "yes" result is obtained. When a determination is made that the clean process for a process or thread at a failed node is not complete, a "no" result is obtained. Successfully completing the clean process check is a prerequisite for lifting a previously set disk fencing operation, for example that used with SCSI-3 PR protocol.
1001041 When a "no" result is obtained, process 700 loops back to perform step 706.
When a "no" result is obtained in step 706, cleanup has not performed successfully, and the departed node is still fenced from the shared disk, for example, through the SCSI-3 PR protocol that was invoked when no heartbeat was detected for the departed node. The departed node may still be registered as REJECT in the respective white lists of other surviving nodes unless a database system previously running on the departed node has been restarted on another surviving node.
[001051 When a "yes" result is obtained, process 700 allows the previously failed node to rejoin the cluster to form a rejoined node (step 708). The cleaning procedure ensures the departed node can start in a clean state and re-join the cluster but does not require a full reboot of the departed node because the cleaning process simply targets the specific process or threads. The departed node is unfenced from the shared disks and the cluster manager restarts the software processes or threads that are part of the clustered application. Process 700 returns a database system previously running on the node to the rejoined node (step 710). The database system instances that were running on the departed node are stopped on the surviving or winning nodes and restarted on the rejoining node.
1001061 Process 700 issues an ARRIVAL message or notification for the database system on the rejoined node to all non-failing nodes, also referred to as surviving nodes (step 712) and terminates thereafter (step 714). As part of the rejoin process an arrival message is generated for the restarting database system instances on the rejoining node. As part of the normal start protocol, ARRIVAL notifications are generated and sent to surviving nodes in the cluster, which then mark the arriving node as ACCEPT in the respective Page 28 of 39 node white lists. The surviving nodes will accept new communications connections from the newly started node.
[00107] Thus is provided, in one illustrative embodiment, a computer-implemented process for fencing shared cluster resources in event of a possible split-brain, identifies a failing resource of a node within a set of shared resources to form an identified failing resource, fences a subset of the set of shared resources to form a winning subset of shared resources and prevents the identified failing resource from communicating with the winning subset of shared resources using a white list maintained at each database system of the winning subset of shared resources. The computer-implemented process further determines whether the identified failing resource has been cleared to form a cleared failing resource and responsive to a determination that the identified failing resource has been cleared, rejoins the cleared failing resource with the winning subset of the set of shared resources absent a re-boot of the cleared failing resource.
[00108] The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention.
In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing a specified logical function. It should also be noted that, in some alternative implementations, the functions noted in the block might occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems that perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
[00109] The corresponding structures, materials, acts, and equivalents of all means or step plus function elements in the claims below are intended to include any structure, material, or act for performing the function in combination with other claimed elements Page 29 of 39 as specifically claimed. The description of the present invention has been presented for purposes of illustration and description, but is not intended to be exhaustive or limited to the invention in the form disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the invention. The embodiment was chosen and described in order to best explain the principles of the invention and the practical application, and to enable others of ordinary skill in the art to understand the invention for various embodiments with various modifications as are suited to the particular use contemplated.
[00100]The invention can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment containing both hardware and software elements. In a preferred embodiment, the invention is implemented in software, which includes but is not limited to firmware, resident software, microcode, and other software media that may be recognized by one skilled in the art.
1001011It is important to note that while the present invention has been described in the context of a fully functioning data processing system, those of ordinary skill in the art will appreciate that the processes of the present invention are capable of being distributed in the form of a computer readable medium of instructions and a variety of forms and that the present invention applies equally regardless of the particular type of signal bearing media actually used to carry out the distribution. Examples of computer readable media include recordable-type media, such as a floppy disk, a hard disk drive, a RAM, CD-ROMs, DVD-ROMs, and transmission-type media, such as digital and analog communications links, wired or wireless communications links using transmission forms, such as, for example, radio frequency and light wave transmissions. The computer readable media may take the form of coded formats that are decoded for actual use in a particular data processing system.
[001021A data processing system suitable for storing and/or executing program code will include at least one processor coupled directly or indirectly to memory elements through a system bus. The memory elements can include local memory employed during actual execution of the program code, bulk storage, and cache memories which provide Page 30 of 39 temporary storage of at least some program code in order to reduce the number of times code must be retrieved from bulk storage during execution.
[00103] Input/output or I/O devices (including but not limited to keyboards, displays, pointing devices, etc.) can be coupled to the system either directly or through intervening I/O controllers.
[00104]Network adapters may also be coupled to the system to enable the data processing system to become coupled to other data processing systems or remote printers or storage devices through intervening private or public networks. Modems, cable modems, and Ethernet cards are just a few of the currently available types of network adapters.
1001051The description of the present invention has been presented for purposes of illustration and description, and is not intended to be exhaustive or limited to the invention in the form disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiment was chosen and described in order to best explain the principles of the invention, the practical application, and to enable others of ordinary skill in the art to understand the invention for various embodiments with various modifications as are suited to the particular use contemplated.

Page 31 of 39

Claims

CLAIMS:

What is claimed is:

1. A computer-implemented process for fencing shared cluster resources in event of a possible split-brain, the computer-implemented process comprising:
identifying a failing resource of a node within a set of shared resources to form an identified failing resource;
fencing a subset of the set of shared resources to form a winning subset of shared resources;
preventing the identified failing resource from communicating with the winning subset of shared resources using a white list maintained at each database system of the winning subset of shared resources;
determining whether the identified failing resource has been cleared to form a cleared failing resource; and responsive to a determination that the identified failing resource has been cleared, rejoining the cleared failing resource with the winning subset of the set of shared resources absent a re-boot of the cleared failing resource.

2. The computer-implemented process of claim 1 wherein fencing a subset of the set of shared resources to form a winning subset of shared resources further comprises fencing of shared disk resource and a memory state resource of a cluster.

3. The computer-implemented process of claim 1 wherein preventing the identified failing resource from communicating with the winning subset of shared resources further comprises:
sending an IGNORE signal to each database system of the winning subset of shared resources to update a white list at each database system.

4. The computer-implemented process of claim 3 wherein preventing the identified failing resource from communicating with the winning subset of shared resources further comprises:
determining whether each database system of the winning subset of shared resources has updated a respective white list of the each database system; and responsive to a determination that each database system of the winning subset of shared resources has updated a respective white list of the each database system, sending a REJECT signal to each database system of the winning subset of shared resources to update a respective white list of the each database system.

5. The computer-implemented process of claim 1 wherein preventing the identified failing resource from communicating with the winning subset of shared resources further comprises:
receiving a message at a target node from a sending node;
determining whether the message is an IGNORE message;
responsive to a determination that the message is an IGNORE message, determining whether the sending node has an ACCEPT status at the target node;
responsive to a determination that the sending node has an ACCEPT status at the target node, determining whether a sequence number associated with the sending node in the white list at the target node greater than or equal to a sequence number in a message;
and responsive to a determination that the sequence number associated with the sending node in the white list at the target node greater than or equal to a sequence number in a message, accepting the message at the target node.

6. The computer-implemented process of claim 1 wherein determining whether the identified failing resource has been cleared to form a cleared failing resource further comprises:
initiating a clean process for a failed node;

determining whether the clean process for a process or thread at the failed node is complete; and responsive to a determination that the clean process for a process or thread at the failed node is complete, wherein the failed node now has a heartbeat, allowing the failed node to rejoin the cluster to form a rejoined node.

7. The computer-implemented process of claim 6 wherein determining whether the identified failing resource has been cleared to form a cleared failing resource further comprises:
responsive to a determination that the clean process for a process or thread at the failed node is complete, returning a database system to the rejoined node; and issuing an ARRIVAL message for the returned database system on the rejoined node to all database systems on unfenced nodes.

8. A computer program product for fencing shared cluster resources in event of a possible split-brain, the computer program product comprising:
a computer recordable-type media containing computer executable program code stored thereon, the computer executable program code comprising:
computer executable program code for identifying a failing resource of a node within a set of shared resources to form an identified failing resource;
computer executable program code for fencing a subset of the set of shared resources to form a winning subset of shared resources;
computer executable program code for preventing the identified failing resource from communicating with the winning subset of shared resources using a white list maintained at each database system of the winning subset of shared resources;
computer executable program code for determining whether the identified failing resource has been cleared to form a cleared failing resource; and computer executable program code responsive to a determination that the identified failing resource has been cleared, for rejoining the cleared failing resource with the winning subset of the set of shared resources absent a re-boot of the cleared failing resource.

9. The computer program product of claim 8 wherein computer executable program code for fencing a subset of the set of shared resources to form a subset of shared resources further comprises fencing of shared disk resource and a memory state resource of a cluster.

10. The computer program product of claim 8 wherein computer executable program code for preventing the identified failing resource from communicating with the winning subset of shared resources further comprises:
computer executable program code for sending an IGNORE signal to each database system of the winning subset of shared resources to update a white list at each database system.

11. The computer program product of claim 10 wherein computer executable program code for preventing the identified failing resource from communicating with the winning subset of shared resources further comprises:
computer executable program code for determining whether each database system of the winning subset of shared resources has updated a respective white list of the each database system; and computer executable program code responsive to a determination that each database system of the winning subset of shared resources has updated a respective white list of the each database system, for sending a REJECT signal to each database system of the winning subset of shared resources to update a respective white list of the each database system.

12. The computer program product of claim 8 wherein computer executable program code for preventing the identified failing resource from communicating with the winning subset of shared resources further comprises:

computer executable program code for receiving a message at a target node from a sending node;
computer executable program code for determining whether the message is an IGNORE message;
computer executable program code responsive to a determination that the message is an IGNORE message for determining whether the sending node has an ACCEPT
status at the target node;
computer executable program code responsive to a determination that the sending node has an ACCEPT status at the target node, for determining whether a sequence number associated with the sending node in the white list at the target node greater than or equal to a sequence number in a message; and computer executable program code responsive to a determination that the sequence number associated with the sending node in the white list at the target node greater than or equal to a sequence number in a message, for accepting the message at the target node.

13. The computer program product of claim 8 wherein computer executable program code for determining whether the identified failing resource has been cleared to form a cleared failing resource further comprises:
computer executable program code for initiating a clean process for a failed node;
computer executable program code for determining whether the clean process for a process or thread at the failed node is complete; and computer executable program code responsive to a determination that the clean process for a process or thread at the failed node is complete, wherein the failed node now has a heartbeat, for allowing the failed node to rejoin the cluster to form a rejoined node.

14. The computer program product of claim 13 wherein computer executable program code for determining whether the identified failing resource has been cleared to form a cleared failing resource further comprises:

computer executable program code responsive to a determination that the clean process for a process or thread at the failed node is complete, returning a database system to the rejoined node; and computer executable program code for issuing an ARRIVAL message for the rejoined node to the database systems on all unfenced nodes.

15. An apparatus for fencing shared cluster resources in event of a possible split-brain, the apparatus comprising:
a communications fabric;
a memory connected to the communications fabric, wherein the memory contains computer executable program code;
a communications unit connected to the communications fabric;
an input/output unit connected to the communications fabric;
a display connected to the communications fabric; and a processor unit connected to the communications fabric, wherein the processor unit executes the computer executable program code to direct the apparatus to:
identify a failing resource of a node within a set of shared resources to form an identified failing resource;
fence a subset of the set of shared resources to form a winning subset of shared resources;

prevent the identified failing resource from communicating with the winning subset of shared resources using a white list maintained at each database system of the subset of shared resources;
determine whether the identified failing resource has been cleared to form a cleared failing resource; and responsive to a determination that the identified failing resource has been cleared, rejoin the cleared failing resource with the winning subset of the set of shared resources absent a re-boot of the cleared failing resource.

16. The apparatus of claim 15 wherein the processor unit further executes the computer executable program code to fence a subset of the set of shared resources to form a winning subset of shared resources further directs the apparatus to fence the shared disk resource and the memory state resource of a cluster.

17. The apparatus of claim 15 wherein the processor unit further executes the computer executable program code to prevent the identified failing resource from communicating with the winning subset of shared resources further directs the apparatus to:
send an IGNORE signal to each database system of the winning subset of shared resources to update a white list at each database system.

18. The apparatus of claim 17 wherein the processor unit further executes the computer executable program code to prevent the identified failing resource from communicating with the winning subset of shared resources further directs the apparatus to:
determine whether each database system of the winning subset of shared resources has updated a respective white list of the each database system; and responsive to a determination that each database system of the winning subset of shared resources has updated a respective white list of the each database system, send a REJECT signal to each database system of the winning subset of shared resources to update a respective white list of the each database system.

19. The apparatus of claim 15 wherein the processor unit further executes the computer executable program code to prevent the identified failing resource from communicating with the winning subset of shared resources further directs the apparatus to:
receive a message at a target node from a sending node;
determine whether the message is an IGNORE message;

responsive to a determination that the message is an IGNORE message, determine whether the sending node has an ACCEPT status at the target node;
responsive to a determination that the sending node has an ACCEPT status at the target node, determine whether a sequence number associated with the sending node in the white list at the target node greater than or equal to a sequence number in a message;
and responsive to a determination that the sequence number associated with the sending node in the white list at the target node greater than or equal to a sequence number in a message, accept the message at the target node.

20. The apparatus of claim 15 wherein the processor unit further executes the computer executable program code to determine whether the identified failing resource has been cleared to form a cleared failing resource further directs the apparatus to:
initiate a clean process for a failed node;
determine whether the clean process for a process or thread at the failed node is complete;
responsive to a determination that the clean process for a process or thread at the failed node is complete, wherein the failed node now has a heartbeat, allow the failed node to rejoin the cluster to form a rejoined node; and send an ARRIVAL message for the rejoined node to the database systems on all unfenced nodes.