JP2011248428A - Load distribution apparatus and load distribution control method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a load distribution apparatus and a load distribution control method for distributing processing requests in a processing system using virtual hosts or the like, while reducing unnecessary distributions that are likely to fail to improve efficiency of load distribution processing.SOLUTION: A virtual server reception part 11 receives processing requests from terminals 31-3n. A load distribution control part 12 distributes the processing requests to virtual hosts 21-1 to 23-3. A statistical information management part associates the virtual hosts 21-1 to 23-3 with physical hosts 21-23 in which each of the virtual hosts operates to collect statistical information on the virtual host that has failed in distributing the processing request, for each physical host associated with the virtual host, and stores them in a statistical information storage part 14. An obstruction management part 15 determines whether the statistical information on the failed virtual host is larger than a predetermined threshold or not, to obstruct the virtual host operating on the physical host and whose statistical information is larger than the threshold, so as not to receive the processing request.

Description

  The present invention relates to a load distribution apparatus and a load distribution control method. The present invention can be applied to, for example, a load distribution apparatus and a load distribution control method in which processing requests are distributed and distributed in a processing system using a virtual host or the like.

  FIG. 9 shows a configuration example of a processing system that performs load distribution processing. As shown in FIG. 9, the processing system that performs load distribution processing includes a load distribution apparatus 10, a plurality of physical hosts 21, 22, 23, and a plurality of terminals 31, 32, 3n that are real servers to which processing is distributed. Have. These are connected to each other via the network 40.

  Each physical host 21, 22, 23 has a virtualization execution environment, and a plurality of virtual hosts are operated. For example, the first physical host 21 operates a virtual host group of virtual hosts 21-1, 21-2, and 21-3, and the second physical host 22 includes virtual hosts 22-1, 22-2, and 22 -3 virtual host group operates, and the third physical host 23 operates the virtual host group of virtual hosts 23-1, 23-2, and 23-3.

  The load distribution apparatus 10 has a plurality of load distribution groups # 1, # 2, and # 3 as virtual servers, and upon receiving a processing request (such as an access request to the server) from the terminals 31 to 3n, a plurality of virtual hosts The processing request is distributed to any one of 21-1 to 23-3. Any one of the virtual hosts 21-1 to 23-3 to which the processing request is distributed executes the processing requested from the terminals 31 to 3n. The load balancing processing system shown in FIG. 9 is applied to a server system used in other load balancing environments in addition to a Web server, an application server, and the like.

  It is expected that server virtualization environment processing systems will become widespread due to increasing demands for efficient use of server resources at low loads with low user traffic and improved operation management technology. In addition, the number of virtual hosts operating on a physical host is expected to increase more and more due to the multi-core server CPU and the improvement of server virtualization technology.

  In addition, the range of increase / decrease of user traffic is large depending on the use time and use time zone, and the increase / decrease or relocation of virtual hosts is dynamically performed according to the increase / decrease of user traffic. It is expected to become increasingly complex in the future.

  In a web system in which a web server and an application server are connected, a dynamic configuration control device, a dynamic configuration control method, and the like that dynamically change a service configuration based on the load status of each server are disclosed in, for example, Patent Document 1 below. Etc. are described. In addition, a configuration in which requests from a terminal are evenly distributed to a plurality of application servers by a load balancer is described in, for example, Patent Document 2 below.

JP 2003-281007 A JP 2007-199829 A

  In a load balancer in a virtual environment, a processing request to a server is distributed using a virtual host instead of a physical host as a distribution destination host. In general, the load balancer does not particularly manage the correspondence between a physical host and a virtual host. However, since the virtual host actually operates on the physical host, when the physical host is in a function stop state due to a hardware failure or the like, the distribution operation of processing requests to the virtual host operating on the physical host is as follows: It becomes useless operation.

  The load balancer can detect a failure of a virtual host that is a distribution destination only by distributing a processing request to a virtual host and then detecting no response to the processing request that has been distributed. Therefore, if there is a high possibility that the physical host is out of order, the invalid processing request will be distributed even though the distribution to the virtual host running on the physical host is likely to fail. There is a case.

  As a method of detecting a failure of a distribution-destination virtual host, you can also install a dedicated module for failure detection on the distribution-destination virtual host and detect the failure of the distribution-destination virtual host by independently coordinating with the load balancer. In this case, it depends on the hardware, operating system (OS), software, or specific technology of the virtualization execution environment used on the distribution destination host. There is a problem that it cannot be used.

  When processing request allocation fails, processing waiting for a response from the distribution-destination virtual host, retry operation for distribution, etc. occur, so processing request allocation failure will cause a delay in access response, etc. The effect of. In addition, processing delays waiting for a response from a virtual host that is a distribution destination, a retry operation for distribution, and the like itself lead to an increase in the processing load of the load balancer. It is an object of the present invention to reduce wasteful distribution operations that are highly likely to fail to distribute processing requests, and to improve the efficiency of load distribution processing.

  A load balancer as an embodiment for solving the above problem is a load balancer that distributes and distributes processing requests to a plurality of virtual hosts, and associates the virtual host with a physical host on which the virtual host operates, and performs processing The statistical information management unit that collects and manages the statistical information of the virtual host that failed to distribute the request for each physical host associated with the virtual host, and the statistical information of the virtual host that failed to distribute A block management unit that blocks a virtual host operating on a large physical host as a processing request distribution destination.

  It is possible to reduce useless processing request distribution operations that have a high possibility of failure, reduce response delay due to processing request distribution failure, and improve response to terminals (users). In addition, it is possible to reduce the processing load of the load balancer due to a response delay or a retry operation for distribution.

It is a figure which shows the structural example of the load distribution apparatus of an indication. It is a figure which shows the structural example of a load distribution policy information table. It is a figure which shows the structural example of a statistical information table. It is a figure which shows the example of a processing flow at the time of failure detection of the distribution of the processing request by a load distribution control part. It is a figure which shows the example of a flow of the block process of the virtual host at the time of a failure by the block management part. It is a figure which shows the operation example of distribution of a process request. It is a figure which shows the example of a processing flow judged by the recovery confirmation monitoring part as the failure recovery. It is a figure which shows the example of a processing flow which cancels | releases the block of the virtual host by a block management part. It is a figure which shows the structural example of the processing system which performs a load distribution process.

  The disclosed load balancer manages the correspondence relationship between a virtual host and a physical host on which the virtual host operates, collects statistical information on failure to distribute processing requests to the virtual host, and operates on a specific physical host When there are many failure to distribute to the virtual host, it is determined that a failure has occurred in the specific physical host. Then, the distribution of the processing request to the virtual host group operating on the physical host determined to have a failure is stopped, and an invalid distribution operation is avoided.

  As a means for performing the above-mentioned operations, the statistical information management unit that manages the correspondence relationship between the physical host and the virtual host, manages the statistical information of the distribution failure of the processing request to the virtual host, and operates on the physical host And a block management unit that changes a virtual host group operating on the physical host to a failure block state when the number of processing request distribution failures to the virtual host exceeds a predetermined threshold.

  A configuration example of the disclosed load distribution apparatus is shown in FIG. As illustrated in FIG. 1, the load distribution apparatus 10 includes a virtual server reception unit 11, a load distribution control unit 12, a load distribution policy information storage unit 13, a statistical information storage unit 14, a block management unit 15, and a recovery confirmation monitoring unit 16. Is provided.

  The virtual server accepting unit 11 publishes the virtual server IP address and port number to the outside of the load distribution processing system, and accepts processing requests (access requests and the like) from the terminals 31 to 3n. The load distribution control unit 12 performs load distribution distribution control according to the load distribution policy information table stored in the load distribution policy information storage unit 13.

  The blocking management unit 15 refers to the statistical information table stored in the statistical information storage unit 14 and analyzes the tendency of processing request distribution failure. For example, when the ratio of the allocation failure to, for example, the virtual host 22-1 operating on the specific physical host 22 exceeds a predetermined threshold, it is determined that a failure has occurred in the specific physical host 22. Then, the block management unit 15 controls to block distribution to the virtual host groups 22-1 to 22-3 operating on the specific physical host 22.

  The restoration confirmation monitoring unit 16 confirms the restoration of the failure of the virtual host groups 22-1 to 22-3 operating on, for example, the physical host 22 that is determined as a failure by the blocking management unit 15 and blocked as a distribution destination.

  Although FIG. 1 shows a configuration example in which the load distribution apparatus 10 and the physical hosts 21 to 23 are separated from each other, they may be partially or entirely integrated into one. Moreover, it is good also as a structure which isolate | separated each function block in the load distribution apparatus 10 into another system partially, or integrated the whole with another system.

  FIG. 2 shows a configuration example of the load balancing policy information table. As shown in FIG. 2, the load distribution policy information storage unit 13 associates the load distribution group name, the load distribution group ID, and the virtual server IP address / port number with each load distribution group in the load distribution policy information table. Store.

  The load balancing policy information table further stores the virtual host IP address / port number for each virtual host in each load balancing group, and information on the operating state of each virtual host. Here, “Online” in the operating status indicates that the corresponding virtual host is set to a valid status as a processing request distribution destination, and “Fault blocked” indicates that the distribution destination is blocked due to a failure. Indicates that it is set. “Maintenance block” indicates that the block is set as a distribution destination for maintenance.

  FIG. 3 shows a configuration example of the statistical information table. As shown in FIG. 3, the statistical information storage unit 14 stores a statistical information table in which a physical host name and a physical host IP address are associated with each physical host. The statistical information table further includes information on the virtual host name, virtual host IP address / port number of each virtual host operating on each physical host, a failure detection flag indicating failure of allocation, and recovery confirmation monitoring status. Is set.

  Here, “OFF” of the failure detection flag indicates that a distribution failure has not been detected, and “ON” indicates that a distribution failure has been detected. Further, “recovered” in the recovery confirmation monitoring state indicates that the failure has been recovered, and “unconfirmed” indicates that the failure has not been confirmed yet.

  Hereinafter, processing when a processing request distribution failure is detected by the load distribution control unit 12, processing for blocking a virtual host on a failed physical host by the blocking management unit 15, and processing for determining failure recovery by the recovery confirmation monitoring unit 16 Next, the process of releasing the block of the virtual host on the failure recovery physical host by the block manager 15 will be described.

<< Processing when load distribution control unit detects distribution failure >>
FIG. 4 shows an example of a processing flow when the load distribution control unit 12 detects a processing request distribution failure. The load distribution control unit 12 distributes and distributes processing requests such as access requests to the virtual hosts 21-1 to 23-3. When an error response is received from the virtual hosts 21-1 to 23-3 in response to the distributed processing request, or when the expiration of the response waiting timer for the processing request transmitted to the virtual hosts 21 to 23 is detected (4- 1) Add (increment) one failure count for each virtual host held inside (4-2).

  As a result of adding (incrementing) the number of failures, it is determined whether or not the number of failures is equal to or greater than a predetermined retry count threshold (4-3). If the number of failures is less than the retry count threshold value, retry processing for distributing processing requests to the virtual host is performed (4-4). If the number of failures is greater than or equal to the retry count threshold, the failure detection flag in the statistical information table is changed from “OFF” to “ON” (4-5), and the failure detection flag is changed to the block management unit 15 Notify (4-6).

  Subsequently, referring to the load balancing policy table, a virtual host as a next distribution destination candidate is determined (4-7), a processing request is transmitted to the determined virtual host, and a response waiting timer for the processing request is started. (4-8).

<< Processing of virtual host failure shutdown by the shutdown management unit >>
FIG. 5 shows a flow example of the block process of the virtual host at the time of failure by the block manager 15. When receiving the failure detection flag update from the load distribution control unit 12 (5-1), the blocking management unit 15 calculates the ratio of the number of failure detection flags ON for each virtual host group operating on each physical host, It is determined whether or not the physical host failure determination threshold value is predetermined (for example, 50%) (5-2).

  If the ratio of the number of failure detection flags ON is greater than or equal to the physical host failure determination threshold, it is determined that some failure has occurred in the physical host of the virtual host group (5-3), and the virtual running on the physical host For the host, the operating state of the load balancing policy information table is changed from “online” to “failure blockage” (5-4).

  The lock management unit 15 changes the operation state of the load balancing policy information table to “failure block”, and then performs a recovery check for the physical host that has been determined that a failure has occurred. The recovery confirmation monitoring unit 16 is notified of the recovery monitoring request of the physical host that is determined to be in the process (5-5).

  FIG. 6 shows an example of processing request distribution. As illustrated in FIG. 6, if the distribution of processing requests to the virtual host 22-1 and the virtual host 22-2 is unsuccessful (NG), it is determined that a failure has occurred in the physical host 22. Then, based on the load balancing policy information, the operating state of the virtual hosts 22-1 to 22-3 operating on the physical host 22 is changed to “failure blockage”. By doing so, the virtual hosts 22-1 to 22-3 are excluded from the selection targets of the processing request distribution destination by the load distribution control unit 12, and the distribution operation to the virtual host 22-3 that is likely to cause the distribution failure Will not be implemented.

<< Processing for recovery from failure confirmation monitoring unit >>
FIG. 7 shows an example of a processing flow in which the recovery confirmation monitoring unit 16 determines that the failure is recovered. When the recovery confirmation monitoring unit 16 receives the notification of the recovery monitoring request of the physical host determined to have failed from the block management unit 15 (7-1), the recovery confirmation of each virtual host operating on the physical host in the statistical information table The monitoring state is changed to “unconfirmed” (7-2).

  A virtual host whose recovery confirmation monitoring status is “unconfirmed” is selected from the virtual hosts in the statistical information table (7-3), a signal indicating a recovery confirmation request is transmitted to the selected virtual host, and the recovery confirmation is performed. A timer for waiting for a response to the request is started (7-4). Specific means for recovery confirmation monitoring include recovery confirmation monitoring by communication protocol, Ping response monitoring, TCP port response monitoring, UDP port response monitoring, application level response monitoring (HTTP, FTP, SIP, TELNET, SSL, etc.) SNMP monitoring or the like can be used.

  The recovery confirmation monitoring unit 16 determines whether a normal response to the recovery confirmation request is received from the virtual host (7-5). When a normal response is not received and an error response is received from the virtual host or when the response wait timer expires (7-6), after waiting for a certain time (7-7), a recovery confirmation request signal is sent to the virtual host. The response waiting timer is started (7-8). This operation is repeated until a normal response is received from the virtual host (7-5). The destination virtual host for repeatedly transmitting the restoration confirmation request signal may be the same virtual host, or may be changed to another unconfirmed virtual host.

  When a normal response to the recovery confirmation request is received from the virtual host, it is determined that the failure of the virtual host has been recovered, and the recovery confirmation monitoring state of the statistical information table is changed to “recovered” (7-9). Then, the ratio of the number of virtual hosts whose recovery confirmation monitoring state is “recovered” in the virtual host group of the physical hosts on which the virtual hosts changed to “recovered” operate is calculated. It is determined whether the ratio of the number of “recovered” virtual hosts is equal to or greater than a predetermined physical host recovery determination threshold (for example, 50%) (7-10).

  If the ratio of the number of “recovered” virtual hosts is equal to or greater than the physical host recovery determination threshold (for example, 50%), it is determined that the corresponding physical host has been recovered (7-11), and the load balancing policy information table The recovery confirmation monitoring status of the virtual host operating on the corresponding physical host is changed to “recovered” (7-12). Thereafter, the recovery confirmation monitoring unit 16 notifies the block management unit 15 of the recovery monitoring response of the physical host that has been determined to have been recovered (7-13).

  On the other hand, when the ratio of the number of “recovered” virtual hosts is smaller than a predetermined physical host recovery determination threshold value (for example, 50%) in the determination of the processing flow 7-10, the “recovered” virtual host The processing returns to the processing flow 7-3 and the same processing is repeated until the ratio of the number of hosts becomes equal to or greater than a predetermined physical host recovery determination threshold (for example, 50%).

<< Process to release virtual host shutdown by shutdown manager >>
FIG. 8 is a processing flow for releasing the block of the virtual host operating on the physical host by the block management unit 15 and setting it to “online state” for the physical host that the recovery confirmation monitoring unit 16 determines to have recovered. An example is shown.

  When receiving the recovery monitoring response of the physical host that has been determined that the failure has been recovered from the recovery confirmation monitoring unit 16 (8-1), the blocking management unit 15 refers to the statistical information table and determines the physical that has been determined to have recovered from the failure. A virtual host operating on the host is extracted (8-2). Then, the operating state in the load distribution policy information table of the extracted virtual host is changed from “failure blocked” to “online” (8-3).

  By changing the operating status of the load balancing policy information table from “failure blocked” to “online”, the virtual host that has been changed to “online” is incorporated into selection of processing request distribution destination candidates by the load balancing control unit 12 It comes to be. For a virtual host whose operation status has been changed from “failure blocked” to “online” and the statistical information failure detection flag is set to “ON”, the failure detection flag is set to “OFF”. Set (8-4).

  In the above-described embodiment, a mode in which one service operates on the virtual host is shown. However, the number of services that operate on the virtual host is not limited to one, and a plurality of virtual IP addresses and port numbers are provided. Thus, the present invention can be applied to a system in which a plurality of services operate on a virtual host.

  In the above embodiment, the failure determination threshold and the recovery determination threshold of the physical host are set to predetermined values for each system in advance. However, if the physical host failure frequently occurs, the recovery determination is performed. It is possible to adopt a configuration in which the determination threshold is dynamically changed according to the situation where a physical host failure has occurred, such as lowering the threshold.

  Also, when the load distribution traffic increases, the recovery destination threshold is lowered when the allocation destination resource is less, and when there is little load balancing traffic and there is a margin, the failure judgment threshold is lowered and the failure status is more strictly For example, the failure determination threshold value or the recovery determination threshold value can be dynamically changed according to the operation status of the entire system.

  The block control method in the load balancer described above does not depend on the hardware, operating system (OS), software, or specific technology of the virtualization execution environment used by the host to which processing requests are distributed. Can be used universally as a target.

DESCRIPTION OF SYMBOLS 10 Load distribution apparatus 11 Virtual server reception part 12 Load distribution control part 13 Load distribution policy information storage part 14 Statistical information storage part 15 Blocking management part 16 Recovery confirmation monitoring part 21-23 Physical host 21-1-23-3 Virtual host 31 ~ 3n terminal 40 network

Claims (5)

In a load balancer that distributes and distributes processing requests to multiple virtual hosts,
Statistical information for associating the virtual host with the physical host on which the virtual host operates, and collecting and managing statistical information of the virtual host for which processing request distribution has failed for each physical host associated with the virtual host The management department,
A block management unit that blocks a virtual host operating on a physical host whose statistical information of the virtual host that has failed to be allocated is larger than a predetermined threshold as a processing request distribution destination;
A load balancer comprising:   The number of failures is accumulated for each virtual host for each occurrence of an error response from the virtual host in response to the processing request or the expiration of the response wait timer, and processing is performed for the virtual host whose accumulated value exceeds a predetermined threshold value. The load distribution control unit configured to set a failure detection flag indicating that the virtual host has failed to distribute a request and notify the statistical information management unit of the setting state of the failure detection flag. 2. The load balancer according to 1.   The block management unit is configured so that the ratio of the number of virtual hosts in which the failure detection flag is set to the number of virtual hosts associated with the same physical host is greater than a predetermined threshold on the physical host. The load balancing apparatus according to claim 2, wherein a virtual host operating in a network is blocked as the distribution destination. A signal indicating a normal response to the recovery confirmation request by transmitting a signal indicating a recovery confirmation request to a virtual host operating on a physical host whose statistical information of the virtual host that has failed to be distributed is greater than a predetermined threshold Is received, the ratio of the number of virtual hosts determined to have recovered from the number of virtual hosts associated with the same physical host is determined from a predetermined threshold. A recovery confirmation monitoring unit is provided for setting a virtual host running on a physical host as having been recovered from a failure for a large physical host,
The load distribution apparatus according to any one of claims 1 to 3, wherein the block management unit sets the processing request as an effective distribution destination for the virtual host set as having been recovered from the failure. . In a load balancing control method that distributes and distributes processing requests to multiple virtual hosts,
Associating the virtual host with a physical host on which the virtual host operates, and collecting and managing statistical information of a virtual host that has failed to distribute processing requests for each physical host associated with the virtual host; ,
Blocking a virtual host operating on a physical host whose statistical information of the virtual host that has failed to be distributed is larger than a predetermined threshold as a processing request distribution destination;
A load distribution control method comprising:

Images