Cluster membership monitor

[0001] This application claims the benefit of U.S. Provisional Patent Application No. 60/201,210 filed May 2, 2000, and entitled “Cluster Membership Monitor”, and U.S. Provisional Patent Application No. 60/201,099, filed May 2, 2000 and entitled “Carrier Grade High Availability Platform”, which are hereby incorporated by reference.

[0002] 1. Field of the Invention

[0003] This invention relates to cluster management of a carrier-grade high availability platform, and more particularly, to a management system and method for managing, maintaining, and using cluster membership data.

[0004] 2. Discussion of the Related Art

[0005] High availability computer systems provide basic and real-time computing services. In order to provide highly available services, members of the system must be aware or capable of being aware of the viability and accessibility of services and hardware components in real-time.

[0006] Computer networks allow data and services to be distributed among computer systems. A clustered network provides a network with system services, applications and hardware divided into nodes that can join or leave a cluster as is necessary. A clustered high availability computer system must maintain cluster information in order to provide services in real-time. Generally this creates a large overhead and commitment of system resources ultimately slowing system response. System costs are also increased by the need for additional memory and storage space for the applications used to monitor system viability.

[0007] The present invention is directed to providing an innovative system and method for monitoring cluster memberships within a clustered computer network that substantially obviates one or more of the problems due to limitations and disadvantages or the related art. An object of the present invention is to provide an innovative system and method for managing, maintaining, and using cluster membership data within a clustered computer network.

[0008] It is an additional object of the present invention to maintain knowledge of the current cluster members.

[0009] It is also an object of the present invention to provide notification throughout the cluster of node join and leave events.

[0010] It is a further object of the present invention to provide a method to verify viability of the cluster members.

[0011] It is an additional object of the present invention to provide a method for electing master and vice-master nodes for a cluster.

[0012] Additional features and advantages of the invention will be set forth in the description that follows, and in part will be apparent from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

[0013] To achieve these other advantages and in accordance with the purpose of the present invention, as embodied and broadly described, a cluster membership monitor is distributed throughout the cluster as monitor entities on each node of the cluster. The monitor entities provide a mechanism for each peer node to maintain knowledge of current cluster members. This knowledge may be dynamic and may not be persistently stored. Cluster membership change notifications are provided to highly available services and highly available aware applications. These notifications can be used to coordinate any highly available service changes when peer nodes join or leave the cluster.

[0014] In accordance with a further aspect of the invention, a method is provided for joining a node to a cluster and establishing master and vice-master nodes within the cluster.

[0015] In accordance with an additional aspect of the invention, a method is provided for a master node to verify viability of the nodes within a cluster and to take appropriate action if a node is non-responsive.

[0016] It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory and are intended to provide further explanation of the invention as claimed.

[0021] Reference will now be made in detail to the preferred embodiment of the present invention, examples of which are illustrated in the accompanying drawings.

[0022] The present invention, referred to in this embodiment as the cluster membership monitor, provides a distributed service through the collaboration of individual monitor entities, each running on a peer node in a cluster. The role of the cluster membership monitor is to monitor the responsiveness of the configured peer nodes of a cluster, maintain the list of responsive nodes constituting the cluster members and to elect one node as the cluster master. Upon detecting a change in cluster membership the cluster membership monitor logs the event and notifies a cluster coordination function. Upon cluster reformation or the failure of the master node the monitor elects a new master node.

[0023] APIs are also provided to: 1) query the cluster membership and master identity; and 2) request a re-read of the configured cluster node list. The cluster membership monitor also exports an API that provides cluster membership information and status to other common cluster services, to highly available services, to distributed system services (DSS), and to highly available aware applications.

[0024] A cluster consists of some number of peer nodes that cooperate to provide a highly available service. If a node in the cluster fails or is brought offline, the cluster continues to provide the service. The cluster membership monitor considers the interconnected peer nodes of a cluster to be cluster members.

[0025] Within a cluster, there is a set of core highly available services that use a 2-n redundancy model. These services must be operational before other system services can become so. The active instances of the core highly available services generally run on the master node of the cluster. Standby instances of the core highly available services are generally run on the vice-master node of the cluster. The cluster membership monitor is responsible for electing both the master and the vice-master nodes from among all eligible nodes. Eligible nodes are typically diskfull system controller, while nodes that are not eligible are typically diskless satellites.

[0026] The cluster membership monitor entity on the master node is responsible for detecting members that join or leave the cluster. The cluster membership monitor API provides information on currently participating members, i.e. nodes that have joined the cluster. It also provides notifications on members that join and leave the cluster as well as notifications of master and vice-master node elections.

[0027] The cluster membership monitor is also provisioned with information about potential cluster members. This provisioning can take place at cluster startup or when the cluster is active.

[0028] Referring to FIG. 1, a cluster 10 is a grouping of n nodes 20, 22, 24, and 26. Node configuration is maintained in a cluster configuration repository, a copy of which is located on the master and vice-master nodes. Within the repository, a cluster 10 is represented as a table of rows, one row per node. In a preferred embodiment each row contains at least the following information:

[0029]

node name	used for log purposes only
node id	unique identifier associated with a node
node address	communications address of the node (IP address)
node type	non-peer, NSP, HSP
node priority	priority for assuming master role (0 = non master)
timeout	heartbeat timer for this node
retry count	number of missed heartbeats before initiating cluster
	consensus procedure

[0030] On each node 20, 22, 24, and 26, the node id 30, node type 32 and master priority 34 are known at cluster membership monitor initialization. This information can be retrieved from Non-Volatile Random Access Memory (NVRAM) or the boot image. Node ids 30 are unique within the cluster 10 and there exists a partial order on them.

[0031] The cluster membership monitor is able to deduce whether a node 20, 22, 24, or 26, is able to act as a cluster master by examining certain information. Each node 20, 22, 24, and 26, provides a lightweight access high-resolution timer 40 that is both monotonic and linear but not necessarily persistent over reboot. A hold off timer must be configured (boot image or NVRAM) that is accessible by the cluster membership monitor entity running on master capable nodes 20, 22, and 26. This timer is used to configure the time that the cluster membership monitor entities on master capable nodes are to wait for a higher priority master node to make itself known.

[0032] In an alternate embodiment the configuration data is stored in LDAP. A start-up configuration file is also provided due to the inability of the LDAP server to be started prior to the master election. The start-up configuration file is stored in a flat file. This file is not an LDAP backup, and contains only the minimal configuration needed to elect a master. At first boot, or after a software upgrade, this file must be provided by software load.

[0033] In this alternate embodiment each diskfull node contains its own configuration file to avoid conflicts. Node eligibility will be determined by whether the node can access the configuration file or not. A node which cannot open, read, and write its configuration file cannot be eligible.

[0034] In the alternate embodiment the cluster membership monitor is the only entity allowed to modify this file. When a configuration change occurs in LDAP tree, the cluster membership monitor is notified and updates the configuration file if it is impacted by the change. The configuration file is also updated when the member ship role or the cluster election round number change.

[0035] As part of this alternate embodiment, a backup of the configuration file is kept in NVRAM. When a configuration change occurs in LDAP tables, or a membership role has changed, the cluster membership monitor is notified and updates the NVRAM.

[0036] The LDAP repository contains, in its preferred embodiment, the following:

[0037] the cluster domain id;

[0038] the cluster nodes table, including, the node id, the node name, and a list of administrative attributes corresponding to the node; and

[0039] the subnet masks

[0040] Referring to FIG. 2, the method of joining a cluster according to a preferred embodiment of the present invention includes the steps as follows. A peer node boots and the local cluster membership monitor entity is started as part of the carrier-grade highly available middleware, step 100. The initial default hold-off, time-out, and retry periods are pre-configured in the boot image or NVRAM. The node then attempts to join a cluster, step 110, by establishing contact with its peers via the cluster node membership monitor entities, step 120. This can be done over configured carrier-grade transport protocol (CGTP) physical links. It does not necessarily use the CGTP services. The cluster membership monitor also uses configured CGTP physical links for exchanging heartbeat and protocol messages.

[0041] If the hold-off timer period has expired, step 122, and no peers are detected, step 120, the local cluster membership monitor entity checks whether it is capable of running as a master, step 124. If it is, the node forms a singleton cluster. This requires that the cluster membership monitor entity suspend the formation of the cluster until it is able to obtain its configuration data from the cluster configuration repository. If it is unable to run as master, continued searching for peers may be appropriate.

[0042] If, however, peer entities are detected, the cluster membership consensus protocol is run with the peer nodes, step 130. If consensus is achieved, step 140, then a search for the master begins. If consensus is not achieved, step 142, the process may be repeated or terminated, or consensus may be mandated.

[0043] If there is consensus, step 130, and a master is located, step 150, the peers form a cluster and the master logs the “cluster join” event to its local system log, step 160. If there is no master, and the hold-off period has expired, step 152, the node checks whether it can become the master, step 154, and if so, does step 156. If it is not capable of acting as a master it simply continues handshaking, step 150, with its peers until a master turns up.

[0044] Upon taking the master role, the cluster membership monitor entity synchronizes with the cluster coordination function for startup sequencing, step 158. It also monitors the configured nodes that are present for viability. In order to reduce the load of monitoring traffic it may make use of agents to perform fan-out/fan-in of the heartbeat. On the failure of a node it re-initiates the cluster consensus procedure after a configurable delay. This procedure is transparent to other platform entities if no inconsistency is detected.

[0045] A cluster membership monitor entity joining a cluster obtains its configuration data from the master cluster membership monitor entity and sets its time-out, retry and consensus protocol timers accordingly. These override the default parameters used during initialization. A node can only join a cluster if its node id is configured in the cluster configuration repository.

[0046] Referring to FIG. 3, a preferred embodiment of the method for monitoring viability includes the following steps. The master node provides a heartbeat, step 200. Each node maintains a time-out clock, which provides a time period within which the node expects to receive a heartbeat from the master. If the master heartbeat is noticed by a node within the time-out period, step 210, the node responds to the master. Responses to the heartbeat are checked against the cluster configuration repository list of cluster members to ensure that all nodes are currently viable, step 230. If a non-master node no longer responds to heartbeats from the master it is removed from the cluster membership list and a cluster membership change is signaled to the cluster coordination function, step 240.

[0047] When a cluster node has not received a heartbeat message within the time-out period, step 210, indicating that the master is no longer viable, the node will check if the cluster reformation has already been started, step 250. If this procedure is not already active, the node will initiate a cluster reformation, step 260. In either event, a new cluster is formed and the node with the highest master priority is elected as the master of the reformed cluster, step 270.

[0048] Variations in the manner a cluster is created can occur. The following are examples including the manner in which a preferred embodiment of the present invention would handle each event. In the event two clusters merge, the node with the highest master priority among the two clusters will be elected as the new master. When a new node is detected by the hardware (e.g. Hot Swap Manager) the node is booted according to its software load configuration parameters. Once its cluster membership monitor entity is started it will join the cluster if it has been configured as a member in the cluster configuration repository.

[0049] Management can remove or add nodes to the cluster by updating the cluster membership table in the cluster configuration repository. These updates are automatically forwarded to the master cluster membership monitor, which propagates the change throughout membership configuration parameters.

[0050] A node is not required to be a member of a cluster in order to communicate with cluster members using raw RTP or standard protocols such as UDP or TCP/IP.

[0051] Additional embodiments include a user space process operating over UDP. Each cluster membership monitor entity may also comprise two entities. A kernel entity that deals with the heartbeat exchanges and a user level daemon that deals with consensus and master election procedures. A kernel implementation for the low level protocol elements allows tighter values on the timers and improves node failure detection times as well as reducing the time and overhead required to achieve cluster membership consensus.

[0052] In order to support split mode software upgrades the cluster membership monitor is required to be service domain aware. The domain name and or cluster incarnation number could be used by the cluster membership monitor algorithms to logically partition the cluster and would be required at boot time. Candidate peers for a cluster must belong to the same domain and or cluster incarnation.

[0053] In an alternate embodiment to improve reconfiguration responsiveness to node failures the cluster membership monitor may provide a mechanism to allow external entities to inform it of a node failure. This is useful in cases where the external mechanism can detect a node failure sooner than the cluster membership monitor heartbeat protocol and it relies on the cluster membership monitor to signal the change in cluster membership to the availability management services.

[0054] System services and highly available aware applications using whatever redundancy model may be appropriate and running on any peer node combination that is required can use a cluster membership API. Cluster membership participation information can be retrieved synchronously while cluster membership changes can be received asynchronously by registering for notifications. The API also contains interfaces that control the membership of the cluster.

[0055] In a preferred embodiment, the cluster membership monitor API exports interfaces to allow other cluster system services and highly available aware applications to retrieve information about current cluster members and to be notified of cluster membership changes. These interfaces include: cmm_connection_prepare, cmm_master_getinfo, cmm_vicemaster_getinfo, cmm_member_getinfo, cmm_node_getid, cmm_member_getall, cmm_member_getcount.

[0056] Cmm_connection_prepare( ) prepares the connection to the cluster membership monitor. Cmm_master_getinfo( ) returns the current master node information. Cmm_vicemaster_getinfo( ) returns the current vice-master node information. Cmm_member_getinfo( ) returns information about the cluster member identified by the node identifier. Cmm_node_getid( ) returns the node identifier of the caller's node. Cmm_member_getall( ) returns a list of currently participating cluster members, a list entry for each node that is responding to heartbeat messages. Cmm_member_getcount( ) returns the number of current cluster members. The cluster member information returned includes the name, IP address, identifier, and type of node. The caller is responsible for allocating enough space for the cluster member table.

[0057] Cmm_cmc_register registers the function indicated by the callback argument with the local cluster membership monitor entity. A highly available system service is provided with the cluster membership change information when the function is called. The highly available system service must use the cmm_notify_getfd( ) and cmm_notify_dispatch( ) functions to receive and dispatch messages from the local cluster membership monitor entity. Registering sets up the callback function to be invoked in response to messages sent to the service from the local cluster membership monitor entity. A service need only register once to receive cluster membership change notifications. An error is returned when a service attempts to register more than once.

[0058] Cmm_cmc_unregister( ) removes the service's registration with the local cluster membership monitor entity so that no further delivery of cluster membership change notifications are made. If the system service's callback function is active when the unregister request is made, the request will be performed after the callback returns to completion.

[0059] The cluster membership monitor API also exports interfaces to allow certain highly available services the ability to administratively control cluster membership.

[0060] It will be apparent to those skilled in the art that various modifications and variations can be made in the system for managing, maintaining, and using cluster membership data within a clustered computer network of the present invention without departing from the spirit or scope of the invention. Thus, it is intended that the present invention covers the modifications and variations of this invention provided they come within the scope of the appended claims and their equivalents.