JP2007334668A - Memory dumping method, cluster system, node constituting the system, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide technique for evading the occurrence of a fault in a magnetic disk interface during the sampling of a memory dump. <P>SOLUTION: When a fault such as an OS panic occurs in a node 10a, a microprocessor 12a samples information contents to be dumped in a control memory 13a of the node 10a and executes retreating processing to a magnetic disk 20a. When the retreating processing to the magnetic disk 20a is executed and abnormality in disk access is detected, a network interface 11a is accessed. Then data communication with a node 10b is executed through a LAN 30 and a transmission request of information contents to be dumped is output to the node 10b. The node 10b temporarily stores a part of the information contents received from the node 10a in its own control memory 13b, transfers the stored information to a disk interface 14b and writes the information to the magnetic disk 20a. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a memory dump method, a cluster system, and nodes and programs constituting the memory dump method, and more particularly to a technique for storing memory dump data of each node in a shared storage device provided in a cluster system having a plurality of nodes.

  Conventionally, in a server device or the like, when a failure such as an OS panic occurs due to a software or hardware problem, the memory contents are saved to a disk (memory dump) and later saved to the disk. To investigate the cause of the failure.

  As a system for performing such a memory dump, for example, in Patent Document 1, it is possible to collect a memory dump on a disk that can be referred to from all the nodes in the cluster and to refer to it from other nodes including the cluster management node. A memory dump collection method is disclosed. This memory dump collection method uses a cluster management node connected to a plurality of nodes via a network and a shared disk connected to each node including the cluster management nodes via a dedicated bus in a cluster system having a plurality of nodes. And a means for instructing the cluster management node to store the memory dump of each node at the time of failure in the shared disk and refer to the memory dump stored in the shared disk from each node.

  As a related technique, Patent Document 2 discloses that in a memory dump method in a distributed processing system, an abnormal failure that has occurred in each terminal processing device connected to a host computer is transmitted to an external storage device on the host computer side via a communication line. The memory dump method to dump is described.

JP 2003-30011 A JP 60-218141 A

  By the way, Patent Document 1 discloses a technique for transmitting dump information to be collected to another node and sharing it with another node. However, no workaround is disclosed when a disk access failure occurs during dump collection. In other words, if an OS panic occurs due to a failure of the disk control unit, etc., the process of saving the memory contents to the disk cannot access the disk, and the memory contents saving process fails. There is a risk of difficulty.

  Accordingly, an object of the present invention is to provide a technique for avoiding a failure in disk access when collecting dumps in a cluster system.

  In a memory dump method according to one aspect of the present invention, in a cluster system in which a plurality of nodes are connected to each other via a communication path, when the first node fails, the memory contents of the first node are transferred to the second node. In the memory dump method for saving to a shared storage device, the step of determining whether or not the access path to the storage device in the first node is accessible, and the access path is inaccessible, Transmitting the memory contents from the first node to the second node using the communication path; and writing the memory contents to the storage device using the access path to the storage device in the second node. Including.

  In the memory dump method according to the first development mode of the present invention, the storage device is composed of multiplexed storage devices, and determines whether or not the access path is accessible for all of the multiplexed storage devices. You may do it.

  A cluster system according to one aspect of the present invention is connected to and shared by a first and second node and a first and second node in a cluster system in which a plurality of nodes are connected to each other by a communication path. The first node determines whether the access path to the storage device is accessible when saving the memory contents of the first node to the storage device when a failure occurs, If the access path is inaccessible, the memory contents are transmitted from the first node to the second node using the communication path, and the second node accesses the storage device in the second node. To write the memory contents to the storage device.

  The nodes constituting the cluster system according to one aspect of the present invention constitute a cluster system in which a plurality of nodes are mutually connected by a communication path. This node includes a memory that holds information to be dumped when a failure occurs, a storage device interface unit to a storage device that is connected to and shared with other nodes, and a communication path with other nodes. A communication interface unit for connecting to each other and a control unit. The control unit determines whether the storage device interface unit is accessible when saving the information held in the memory to the storage device when a failure occurs. If the access to the storage device interface unit is impossible, information is transmitted to other nodes using the communication interface unit, and a write request from other nodes to the storage device has arrived during normal operation. In this case, the storage device interface unit is used to control to write information accompanying the write request to the storage device.

  A program according to one aspect of the present invention is a node constituting a cluster system in which a plurality of nodes are connected to each other by a communication path, and a memory that holds information to be dumped when a failure occurs; A failure occurs in a computer that constitutes a node including a storage device interface unit connected to a shared storage device connected to another node and a communication interface unit connected to the other node via a communication path. When the information stored in the memory is sometimes saved to the storage device, a process for determining whether or not the storage device interface unit is accessible, and if the access to the storage device interface unit is impossible, the communication interface unit The process of sending information to other nodes using and the writing from other nodes to the storage device during normal operation When the viewing request has come, the process of writing the information accompanying the write request to the storage device using the storage device interface unit, thereby to execute.

  According to the present invention, even when the access path to the storage device becomes a failure in the saving process of the memory contents to the storage device at the time of OS panic, the other data transfer path between the nodes is used. By transmitting to the node and using the access path of the storage device in the other node, the memory contents at the time of OS panic can be saved to the storage device.

  The cluster system according to the embodiment of the present invention includes a first and second nodes (10a and 10b in FIG. 1) in a cluster system in which a plurality of nodes are mutually connected by a communication path (30 in FIG. 1), And a storage device connected to the first and second nodes (magnetic disks 20a and 20b in FIG. 1). When the first node (10a in FIG. 1) saves the memory contents in the control memory (13a in FIG. 1) to the storage device (20a, 20b in FIG. 1) when a failure such as an OS panic occurs, the first node (10a in FIG. 1) It is determined whether the access path is accessible, that is, whether it is accessible via the disk interface (14a, 15a in FIG. 1). If the access path to the storage device is inaccessible, the memory contents are transmitted from the first node to the second node using the communication path (30 in FIG. 1). In addition, the second node (10b in FIG. 1) uses the access path to the storage device in the second node, that is, the memory contents to the storage device via the disk interface (14b, 15b in FIG. 1). Write.

  In a cluster system such as a cluster server that achieves high availability as described above, if an OS panic occurs on one node and disk access is difficult on the failed node, the other normally operating node Use the disk access path to save the memory contents to the magnetic disk. In other words, OS panics in a cluster system such as a cluster server where a single server is connected to each other via a high-speed data communication path such as LAN or RDMA (Remote Direct Memory Access), and cluster management is performed by cluster software, thereby improving availability. The memory contents are saved to disk when a failure such as the above occurs. In this save process, it is determined whether the access path to the disk is accessible. If the disk cannot be accessed, the communication path between nodes is used to communicate with other nodes, and the memory contents are transferred to the other node. Send and save to disk using the disk access path of the other node.

  According to such a cluster system, even when an OS panic occurs due to a failure of the disk control unit or the like, even when the access path to the storage device becomes a failure, the process of saving the memory contents to the disk is performed. Is possible. Hereinafter, description will be made with reference to the drawings in accordance with embodiments.

  FIG. 1 is a block diagram showing a configuration of a cluster system according to an embodiment of the present invention. The cluster system shown in FIG. 1 is a configuration example of a two-node cluster in which single servers are connected by a LAN 30 and can perform data communication with the other node. The servers are referred to as a node 10a and a node 10b. Each node 10a, 10b exchanges data with the other node via a network interface 11a (11b) that communicates via the LAN 30, a disk interface 14a (14b) that exchanges data with the magnetic disk 20a, and a magnetic disk 20b. A disk interface 15a (15b), a microprocessor 12a (12b) that executes a program for controlling a node, and a control memory 13a (13b) that holds control information are commonly connected by a bus. The node 10a (10b) is connected to the common magnetic disk 20a via the disk interface 14a (14b), and is connected to another common magnetic disk 20b via the disk interface 15a (15b). With such a connection configuration, it is possible to access the common magnetic disks 20a and 20b from the nodes 10a and 10b. The magnetic disks 20a and 20b have a multiplexed configuration in order to increase availability. Here, an example of a duplex configuration is shown. If a redundant configuration is not required, a single system may be used.

  In the cluster system having such a configuration, it is assumed that a failure such as an OS panic occurs in the node 10a. Processing in this case will be described. FIG. 2 is a flowchart showing an operation in the cluster system according to the embodiment of the present invention. The microprocessor 12a of the node 10a collects information contents to be dumped in the control memory 13a of the node 10a (step S11), and executes a process of saving to the magnetic disk 20a. At this time, if a disk access abnormality is detected by executing a save process to the magnetic disk (step S12), the network interface 11a is accessed. Then, data communication with the node 10b is performed via the LAN 30, and a transmission request for information content to be dumped is made to the node 10b (step S13). The node 10b temporarily stores part of the information content received from the node 10a in its own control memory 13b (step S14), passes the stored information to the disk interface 14b, and writes to the magnetic disk 20a (step S14). S15). The nodes 10a and 10b repeat these processes until the final data of the information contents, and write all the information contents to be dumped in the control memory 13a to the magnetic disk 20a.

  In the above description, the processing in the case where it is impossible to access the magnetic disk 20a when saving the information content to be dumped to the magnetic disk 20a in the node 10a has been described. If the magnetic disk is duplicated with the magnetic disks 20a and 20b, the access to the magnetic disk 20a is tried. If the access is impossible, the access to the magnetic disk 20b is further tried and the access is still not possible. If possible, information contents to be dumped may be written using the disk interface 14b or 15b to the magnetic disk 20a or the magnetic disk 20b in the node 10b via the LAN 30.

It is a block diagram which shows the structure of the cluster system which concerns on the Example of this invention. It is a flowchart which shows the operation | movement in the cluster system based on the Example of this invention.

Explanation of symbols

10a, 10b Nodes 11a, 11b Network interfaces 12a, 12b Microprocessors 13a, 13b Control memories 14a, 14b, 15a, 15b Disk interfaces 20a, 20b Magnetic disk 30 LAN

Claims (8)

In a cluster system in which a plurality of nodes are connected to each other via a communication path, a memory dump method is used to save the memory contents of the first node to a shared storage device with the second node when a failure occurs in the first node. And
Determining whether an access path to the storage device in the first node is accessible;
If the access path is inaccessible, transmitting the memory contents from the first node to the second node using the communication path;
Writing the memory contents to the storage device using an access path to the storage device in the second node;
A memory dump method comprising:   2. The memory dump according to claim 1, wherein the storage device is composed of multiplexed storage devices, and judges whether or not an access path is accessible for all of the multiplexed storage devices. Method. In a cluster system in which multiple nodes are connected to each other via communication paths,
First and second nodes;
A storage device connected to and shared with the first and second nodes;
With
When the first node saves the memory contents of the first node to the storage device when a failure occurs, the first node determines whether the access path to the storage device is accessible. If inaccessible, send the memory contents from the first node to the second node using the communication path;
The cluster system, wherein the second node writes the memory contents to the storage device using an access path to the storage device in the second node.   The storage device includes a multiplexed storage device, and the first node determines whether an access path is accessible for all of the multiplexed storage devices. The cluster system according to claim 3. In a node constituting a cluster system in which a plurality of nodes are connected to each other via a communication path,
A memory that holds information to be dumped at the time of failure; and
A storage device interface unit to a storage device connected to another node and shared;
A communication interface unit for connecting to the other node via the communication path;
A control unit;
With
The controller determines whether or not the storage device interface unit is accessible when evacuating information stored in the memory to the storage device when a failure occurs, and cannot access the storage device interface unit In this case, the information is transmitted to the other node using the communication interface unit. When a write request from the other node to the storage device arrives during normal operation, the storage device interface A node constituting a cluster system, wherein a control is performed so that information accompanying a write request is written to the storage device using a storage unit.   The storage device is configured by a multiplexed storage device, the storage device interface unit includes a plurality of interface units corresponding to the multiplexed storage device, and the control unit is the multiplexed 6. The node constituting the cluster system according to claim 5, wherein it is determined whether or not all of the storage devices are accessible. A node that constitutes a cluster system in which a plurality of nodes are connected to each other via communication paths, and is connected to a memory that holds information to be dumped when a failure occurs, and is connected to other nodes and shared To a computer constituting a node comprising a storage device interface unit to a storage device and a communication interface unit for connecting to the other node via the communication path,
A process of determining whether or not the storage device interface unit is accessible when saving the information held in the memory when a failure occurs to the storage device;
When access to the storage device interface unit is impossible, processing to transmit the information to the other node using the communication interface unit;
When a write request to the storage device comes from the other node during normal operation, a process of writing information accompanying the write request to the storage device using the storage device interface unit;
A program that executes The storage device includes a multiplexed storage device, and the storage device interface unit includes a plurality of interface units corresponding to the multiplexed storage device, and the storage device interface unit is accessible. 8. The program according to claim 7, wherein in the process of determining whether or not all of the multiplexed storage devices are accessible, it is determined.

Images