CN102609324A - Method, device and system for restoring deadlock of virtual machine - Google Patents

Abstract

An embodiment of the invention provides a method, a device and a system for restoring deadlock of a virtual machine, which solve the problem of restarting an integral server to some extent. The method for restoring deadlock of the virtual machine includes: transmitting non-maskable interrupt to the virtual machine and configuring virtual processor resources of the virtual machine in case of deadlock of the virtual machine; dispatching a first virtual processor from the configured virtual processor resources comprising at least one virtual processor and loading an inner core guiding program of the virtual machine to the first virtual processor after the virtual machine makes response to the non-maskable interrupt and collects last words, so that the virtual machine can restore service progress according to the collected last words. The method, the device and the system for restoring deadlock of the virtual machine in the embodiment are applicable to cases of deadlock of the virtual machine.

Description

Method, device and system for recovering virtual machine after deadlock

technical field

The invention relates to the field of computer technology, in particular to a recovery method, device and system for a virtual machine after a deadlock.

Background technique

In a cloud computing system, a large number of computing nodes adopt virtualization solutions. In a virtualized environment, generally only one special virtual machine (also called a privileged virtual machine) among all virtual machines undertakes management functions and controls real physical resources. access.

When a deadlock occurs in a privileged virtual machine, the network card or disk device driver cannot work normally, and cannot meet the IO request of the guest operating system, resulting in the failure of all virtual machine services on the entire computing node. In order to solve this problem, in the prior art, the normal operation of the privileged virtual machine is restored by restarting the entire computing node such as a server. (Intelligent Platform Management Interface, Intelligent Platform Management Interface) monitors the health of the entire server, and sends a fixed frequency to the watchdog timer (usually called a hardware dog) of the BMC (Baseboard Management Controller, Baseboard Management Controller) of the server. ) to send heartbeats. When a deadlock occurs in a privileged virtual machine, the software dog cannot continue to run and cannot send heartbeats to the hardware dog of the BMC. After the hardware dog does not receive the heartbeat for a certain period of time, the BMC will power on the entire server to restart it. .

However, the method of restarting the entire server in the prior art causes phenomena such as out-of-synchronization of stored data, and the restart process takes a long time, resulting in long-term service interruption.

Contents of the invention

Embodiments of the present invention provide a recovery method, device and system for a virtual machine after a deadlock, so as to avoid problems caused by restarting the entire server to a certain extent.

Embodiments of the invention adopt the following technical solutions:

On the one hand, a recovery method after a virtual machine deadlock is provided, including:

In the case of a virtual machine deadlock, sending a non-maskable interrupt to the virtual machine and configuring virtual processor resources of the virtual machine;

After the virtual machine responds to the non-maskable interrupt and collects the last words, schedule the first virtual processor from one or more virtual processors included in the configured virtual processor resources and Load the kernel boot program of the virtual machine on the virtual processor, so that the virtual machine resumes the business process according to the collected last words; the last last words include stack information and stack information of each virtual processor corresponding to the virtual machine when deadlocked Register information.

On the other hand, a recovery method after a virtual machine deadlock is provided, including:

Respond to non-maskable interrupts sent by the virtual machine monitor VMM;

Collecting the last words of the virtual machine, wherein the last words include stack information and register information of each virtual processor corresponding to the virtual machine at the time of deadlock;

After the kernel boot program of the virtual machine is loaded, the service process is resumed according to the last words of the virtual machine.

Correspondingly, on the one hand, a virtual machine monitor is provided, including:

An interrupt simulator, configured to send a non-maskable interrupt to the virtual machine when a deadlock occurs in the virtual machine;

a configuration unit, configured to configure virtual processor resources of the virtual machine;

A scheduling loading unit, configured to schedule a first virtual processor from one or more virtual processors included in the configured virtual processor resource after the virtual machine responds to the non-maskable interrupt and collects the last words And load the kernel boot program of the virtual machine on the first virtual processor, so that the virtual machine resumes the business process according to the collected last words.

On the other hand, a virtual machine is provided, including:

The interrupt processing unit is used to respond to the non-maskable interrupt sent by the virtual machine monitor VMM;

A collection unit, configured to collect the last words of the virtual machine, wherein the last words include stack information and register information of each virtual processor corresponding to the virtual machine at the time of deadlock;

The service restoration unit is configured to restore the service process according to the last words of the virtual machine after the kernel boot program of the virtual machine is loaded.

An embodiment of the present invention also provides a host machine, including a virtual machine monitor and a privileged virtual machine, where the virtual machine monitor is the above-mentioned virtual machine monitor, and the privileged virtual machine is the above-mentioned virtual machine.

The embodiment of the present invention also provides a computing node, which includes: a hardware layer, a virtual machine monitor VMM running on the hardware layer, and a virtual machine running on the VMM; the virtual Machines include privileged virtual machines and at least one business virtual machine; where:

The VMM is configured to send a non-maskable interrupt to the privileged virtual machine and configure virtual processor resources of the privileged virtual machine when a deadlock occurs in the privileged virtual machine; when the privileged virtual machine responds to the non-maskable After interrupting and collecting the last words, scheduling the first virtual processor from one or more virtual processors included in the configured virtual processor resource and loading the privileged virtual machine on the first virtual processor The corresponding kernel bootloader;

The privileged virtual machine is used to respond to the non-maskable interrupt sent by the VMM; collect the last words of the privileged virtual machine, and the last words include stack information and register information of each virtual processor corresponding to the privileged virtual machine at the time of deadlock ; After the kernel boot program of the privileged virtual machine is loaded, resume the service process according to the last words of the privileged virtual machine.

An embodiment of the present invention also provides a computer system, including at least one computing node described above.

It can be seen that the embodiments of the present invention provide a recovery method, device and system for a virtual machine after a deadlock, by sending a non-maskable interrupt to the virtual machine and configuring the virtual processor of the virtual machine in the case of a deadlock of the virtual machine resources; when the virtual machine responds to the non-maskable interrupt and collects the last words, the virtual machine monitor schedules the first virtual processor from one or more virtual processors included in the configured virtual processor resource and executes the The kernel boot program of the virtual machine is loaded on the first virtual processor, so that the virtual machine can restore the business process according to the collected last words; compared with the method of powering off and restarting the entire computing node in the prior art, the present invention It can avoid restarting the entire server and restore the operation of the virtual machine in a timely manner, thus avoiding the long-term business disaster and out-of-sync data to a certain extent.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. Those skilled in the art can also obtain other drawings based on these drawings without creative work.

FIG. 1 is a computing node provided by an embodiment of the present invention;

FIG. 2 is a structural block diagram of a virtual machine provided by an embodiment of the present invention;

FIG. 3(a) is a flow chart of a recovery method for a virtual machine after a deadlock provided by an embodiment of the present invention;

FIG. 3(b) is a flow chart of a recovery method for a virtual machine after a deadlock provided by an embodiment of the present invention;

FIG. 4(a) is a flow chart of another recovery method for a virtual machine after a deadlock provided by an embodiment of the present invention;

FIG. 4(b) is a schematic diagram of mutual reference between the data structure of the virtual machine and the data structure of the virtual processor provided by the embodiment of the present invention;

FIG. 5 is a block diagram of a virtual machine monitor provided by an embodiment of the present invention;

FIG. 6 is a block diagram of a virtual machine monitor provided by an embodiment of the present invention;

FIG. 7 is a block diagram of a virtual machine provided by an embodiment of the present invention;

FIG. 8 is a block diagram of a virtual machine provided by an embodiment of the present invention;

FIG. 9 is a block diagram of a host provided by an embodiment of the present invention;

Fig. 10 is a block diagram of a computer system provided by an embodiment of the present invention.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

In order to facilitate understanding of the embodiments of the present invention, several elements that will be introduced in the description of the embodiments of the present invention are firstly introduced here.

Deadlock: A state of operating system or software operation: In a multitasking system, when one or more processes are waiting for system resources, and the resources are occupied by the process itself or other processes, a deadlock is formed.

Virtual Machine Monitor (Virtual Machine Monitor, referred to as VMM): such as Xen Hypervisor.

Host:

As a management layer, it is used to complete the management and allocation of hardware resources; present a virtual hardware platform for virtual machines; realize the scheduling and isolation of virtual machines. Among them, the Host may be a virtual machine monitor (VMM); in addition, sometimes the VMM cooperates with a privileged virtual machine, and the combination of the two forms the Host. Wherein, the virtual hardware platform provides various hardware resources for each virtual machine running on it, such as virtual processor (such as VCPU), virtual memory, virtual disk, virtual network card and so on. Wherein, the virtual disk may correspond to a file or a logical block device of the Host. The virtual machine runs on the virtual hardware platform prepared by the host, and one or more virtual machines run on the host.

Virtual Machine (Virtual Machine, VM for short):

One or more virtual computers can be simulated on a physical computer through virtual machine software, and these virtual machines work just like real computers. Operating systems and applications can be installed on virtual machines, and virtual machines can also Access network resources. To the applications running in the virtual machine, the virtual machine looks like working on a real computer.

Hardware layer:

The hardware platform on which the virtualization environment runs. Wherein, the hardware layer may include a variety of hardware, for example, the hardware layer of a computing node may include processors (such as CPU) and memory, and may also include high-speed/low-speed input/output (I/O, Input/Output )equipment.

GuestOS: Guest operating system.

Privileged virtual machine: a special GeustOS virtual machine, which can also be called a driver domain. For example, this special virtual machine is called Dom0 on the Xen Hypervisor platform, and virtual machines such as network cards and SCSI disks are installed in the virtual machine. Drivers for physical devices can detect and directly access these real physical devices. Other virtual machines use the corresponding mechanism provided by the Hypervisor to access real physical devices through privileged virtual machines.

NMI (Nonmaskable Interrupt): non-maskable interrupt (that is, the CPU cannot be masked). Regardless of the state of the IF bit in the status register, the CPU must respond when it receives a valid NMI, and there is no interrupt response cycle when it is responded. Non-maskable interrupts are usually used for fault handling (for example: coprocessor operation error, memory verification error, I/O channel verification error, etc.).

BMC: Baseboard Management Controller, Baseboard Management Controller.

IPMI: Abbreviation of Intelligent Platform Management Interface (Intelligent Platform Management Interface), is an industrial standard used to manage peripherals used in enterprise systems based on Intel structure. Wait for the company to formulate. Users can use IPMI to monitor the physical health characteristics of the server, such as temperature, voltage, fan working status, power supply status, etc.

Embodiment one

As shown in FIG. 1, an embodiment of the present invention provides a computing node 100, including:

The hardware layer 110 , the virtual machine monitor VMM50 running on the hardware layer 110 , and the virtual machine VM running on the VMM, the virtual machine VM includes a privileged virtual machine 701 and at least one service virtual machine 702 .

The hardware layer 110 includes a processor, a hard disk, a network card, a memory, and the like. Wherein, the processor may be a central processing unit (CPU, Central Processing Unit), a digital signal processor (DSP, Digital Signal Processing), a field programmable gate array (FPGA, Field Programmable Gate Array) and the like.

As shown in Figure 2, the virtual machine mainly includes: virtual hardware 21, a guest operating system 22 running on the virtual hardware 21; the virtual hardware 21 includes: at least one virtual processor 23; virtual memory (VMEM) 24; A virtual hard disk (VDISK) 25 and at least one virtual device (VDEVICE) 26; it should be noted that the virtual processor in Figure 2 can change with different times or different business needs, and the same virtual processor can be used by multiple The virtual machine is shared, so the virtual processor in Figure 2 is essentially the virtual processor corresponding to the virtual machine at a certain moment. In addition, the privileged virtual machine 701 also includes drivers for physical devices running on the guest operating system (not shown in the figure), and the business virtual machine 702 also includes business applications running on the guest operating system. It should be noted that business applications run in business virtual machines, and business virtual machines need to use privileged virtual machines as an intermediary to indirectly access IO resources such as network and storage.

In the embodiment of the present invention, the virtual machine monitor can refer to Xen Hypervisor, and the privileged virtual machine refers to a special virtual machine on the Xen Hypervisor platform. This special virtual machine is also called Dom0 on the Xen Hypervisor platform, and is also called the driver domain on other virtualization platforms. For convenience of description, the embodiment of the present invention uses the statement of a privileged virtual machine, but the present invention does not Not limited.

Wherein, in the case of a deadlock in the privileged virtual machine, the VMM50 is used to send a non-maskable interrupt to the privileged virtual machine and configure the virtual processor of the privileged virtual machine in the case of a deadlock in the privileged virtual machine resources; after the privileged virtual machine responds to the non-maskable interrupt and collects the last words, schedule the first virtual processor from one or more virtual processors included in the configured virtual processor resources and Loading the kernel boot program corresponding to the privileged virtual machine on the first virtual processor;

In this embodiment, the first virtual processor may be the first virtual processor randomly scheduled from one or more virtual processors included in the configured virtual processor resource, or the first virtual processor that is configured from Among the one or more virtual processors included in the virtual processor resource, the scheduled virtual processor has the highest credit value and/or weight value.

The first virtual processor is the virtual processor that is configured first, and comes according to the order of configuration. In addition, selecting the virtual processor with the highest credit value and the highest weight value among the predetermined number of virtual processors, or the virtual processor with the highest credit value or the highest weight value facilitates the fast loading of the kernel boot program.

The privileged virtual machine 701 is used to respond to the non-maskable interrupt sent by the VMM50; collect the last words of the privileged virtual machine, and the last words include the stack information and registers of each virtual processor corresponding to the privileged virtual machine at the time of deadlock Information; after the kernel boot program of the privileged virtual machine is loaded, resume the service process according to the last words of the privileged virtual machine.

In the case of a deadlock in the service virtual machine 702, the VMM50 is also used to send a non-maskable interrupt to the service virtual machine 702 and configure the virtual Processor resource: after the service virtual machine 702 responds to the non-maskable interrupt and collects the last words, schedule the second virtual processor from one or more virtual processors included in the configured virtual processor resource And loading the kernel boot program corresponding to the service virtual machine on the second virtual processor;

In this embodiment, the second virtual processor may be the first virtual processor randomly scheduled from one or more virtual processors included in the configured virtual processor resources, or the second virtual processor is the Among the one or more virtual processors included in the virtual processor resource, the scheduled virtual processor has the highest credit value and/or weight value.

The second virtual processor is the virtual processor configured first, and comes according to the sequence of configuration. In addition, selecting the virtual processor with the highest credit value and the highest weight value among the predetermined number of virtual processors, or the virtual processor with the highest credit value or the highest weight value facilitates the fast loading of the kernel boot program.

It should be noted that the terms first and second used in the description of this embodiment do not mean to limit the sequence, but are only for convenience of distinction. The first virtual processor here is used to indicate the virtual processor scheduled by the VMM for the privileged virtual machine, and the second virtual processor here is used to indicate the virtual processor scheduled by the VMM for the service virtual machine.

The service virtual machine 702 is used to respond to the non-maskable interrupt sent by the VMM; collect the last words of the service virtual machine, and the last words include the stack information and registers of each virtual processor corresponding to the service virtual machine at the time of deadlock Information; after the kernel boot program of the service virtual machine is loaded, resume the service process according to the last words of the service virtual machine.

In the embodiment of the present invention, the virtual machine collects the last words from the hardware layer through the VMM, specifically, the virtual machine collects the stack information of each virtual processor corresponding to the virtual machine at the time of deadlock from the memory of the hardware layer through the VMM , and collect the register information of each virtual processor corresponding to the virtual machine at the time of deadlock from the registers of the processor at the hardware layer. It should be pointed out that both the special virtual machine and the business virtual machine can collect last words from the hardware layer through the VMM. The stack information of each VCPU corresponding to the virtual machine at the time of deadlock is collected from the memory of the hardware layer, and the register information of each VCPU corresponding to the virtual machine at the time of deadlock is collected from the registers of the CPU.

It can be seen that the embodiment of the present invention provides a computing node. The virtual machine monitor sends a non-maskable interrupt to the virtual machine and configures the virtual processor resources of the virtual machine. The virtual machine responds to the non-maskable interrupt and collects the last words. Then use the virtual machine monitor to schedule a certain virtual processor (such as the aforementioned first virtual processor or second virtual processor) from one or more virtual processors included in the configured virtual processor resource, and The kernel boot program of the virtual machine is loaded on the certain virtual processor, so that the virtual machine can recover the business process according to the collected last words; In the method of powering off the entire server, the present invention detects the deadlock of the privileged virtual machine and restores the privileged virtual machine online through the VMM, which can restore the operation of the privileged virtual machine in time and maintain the running state of the entire computing node, thereby avoiding the entire computing node such as the entire The server is restarted, thereby avoiding business disasters that take too long and data out of sync. In addition, due to the rapid recovery of privileged virtual machines, the IO functions of other business virtual machines to access network and storage IO resources such as IO resources are automatically restored. The business virtual machine does not affect; on the other hand, when the business virtual machine is deadlocked, the present invention can restore the business virtual machine online through the VMM after detecting the deadlock of the business virtual machine, so as to avoid restarting the entire server and restore the business virtual machine in time The operation of the virtual machine realizes the rapid recovery of the business in the virtual machine, and avoids the business disaster of the business virtual machine for too long, and the data is not synchronized.

Embodiment two

As shown in Figure 3 (a), the embodiment of the present invention provides a recovery method after a virtual machine deadlock, the execution subject of the method may be a virtual machine monitor, specifically, it may be XenHypervisor, including:

301. When a deadlock occurs in a virtual machine, send a non-maskable interrupt to the virtual machine and configure virtual processor resources of the virtual machine.

302. After the virtual machine responds to the non-maskable interrupt and collects the last words, schedule the first virtual processor from one or more virtual processors included in the configured virtual processor resources and execute the Load the kernel boot program of the virtual machine on the first virtual processor, so that the virtual machine resumes the business process according to the collected last words; the last words include the stacks of each virtual processor corresponding to the virtual machine at the time of deadlock information and register information.

Wherein, the first virtual processor may be the first virtual processor randomly scheduled from one or more virtual processors included in the configured virtual processor resources, or may be the first virtual processor from the configured The virtual processor with the highest credit value and/or weight value scheduled out of one or more virtual processors included in the virtual processor resource is not limited in this embodiment of the present invention, and different scheduling can be adopted according to actual application conditions Algorithms are implemented, such as fair scheduling algorithms.

In one implementation manner, the configuring the virtual processor resources of the virtual machine in step 301 includes:

Acquiring a predetermined number of virtual processors, wherein the predetermined number is the number of virtual processors corresponding to the virtual machine at the time of deadlock;

According to the recorded first resource information of one or more virtual processors corresponding to the virtual machine, set the second resource information of the predetermined number of virtual processors, the first resource information and the second The resource information is used to represent virtual processor resources with the same distribution, and both the first resource information and the second resource information include a credit value and a weight value.

It should be noted that, in the embodiment of the present invention, the "first" in the "first virtual processor" does not mean an order, but is only for convenience of distinction.

It can be seen that the embodiment of the present invention provides a recovery method for a virtual machine after a deadlock. From the side of the virtual machine monitor, a non-maskable interrupt is sent to the virtual machine and the virtual processor resources of the virtual machine are configured, and then Scheduling a first virtual processor from one or more virtual processors included in the configured virtual processor resource and loading a kernel boot program of the virtual machine on the first virtual processor, so that the virtual machine can Restore the business process according to the collected last words. Compared with the method of powering off and restarting the entire server in the prior art, the present invention can restore the operation of the virtual machine in time without restarting the entire server, thus avoiding the long time of business disaster and the loss of data Asynchrony etc.

As shown in FIG. 3(b), the embodiment of the present invention provides a method for recovering a virtual machine after a deadlock. The execution subject of the method is a virtual machine. Specifically, it may be a privileged virtual machine or a service virtual machine. The method can include:

311. Respond to the non-maskable interrupt sent by the virtual machine monitor VMM.

Specifically, in response to the non-maskable interrupt sent by the virtual machine monitor VMM, end the business running on the virtual machine

312. Collect the last words of the virtual machine, where the last words include stack information and register information of each virtual processor corresponding to the virtual machine at the time of deadlock.

313. After the kernel boot program of the virtual machine is loaded, resume the service process according to the last words of the virtual machine.

In one implementation manner, step 312 may be: collecting the last words of the virtual machine from the hardware layer through the VMM.

In one implementation, step 313 may include: after the kernel boot program of the virtual machine is loaded, identify the The running instruction and address of the virtual machine before the deadlock, so that the virtual machine continues to execute the next instruction of the running instruction after recovery; it should be noted that the running instruction and address of the virtual machine before the deadlock here refers to is the running instructions and addresses processed by the virtual machine at the last moment before the deadlock.

More specifically, after the kernel boot program of the virtual machine is loaded, initialize the linked list of the virtual processor according to the stack information and register information of each virtual processor corresponding to the virtual machine at the time of deadlock, and according to The stack information and the register information in the linked list identify the running instruction and address of the virtual machine before deadlock, so that the virtual machine continues to execute the next instruction after the running instruction after recovery.

It can be seen that the embodiment of the present invention provides a recovery method for a virtual machine after a deadlock. From the virtual machine side, the virtual machine responds to the non-maskable interrupt sent by the virtual machine monitor and collects the last words of the virtual machine. After the server completes the loading of the boot program, the virtual machine realizes the recovery of the business process according to the last words. Compared with the method of powering off and restarting the entire server in the prior art, the present invention can restore the virtual machine in time without restarting the entire server. Running, thus avoiding business disasters for too long, data out of synchronization and other phenomena.

Embodiment Three

As shown in Figure 4(a), the embodiment of the present invention provides another recovery method after a virtual machine deadlock, including:

401. The virtual machine monitor VMM initializes a monitoring timer.

402. The virtual machine VM generates heartbeat information according to a predetermined first time interval. The virtual machine VM here may be a privileged virtual machine or a business virtual machine.

403. In the calling mode provided by the VMM, the VM sends the heartbeat information to the VMM at a predetermined second time interval. Preferably, the call mode is a hyper call.

404. The VMM receives heartbeat information sent by the VM at a predetermined second time interval; the VMM updates a monitoring timer according to the heartbeat information.

It should be noted that the first time interval and the second time interval involved in steps 402, 403, and 404 may be the same time interval, or may be different time intervals. It should be understood that the time interval in the embodiment of the present invention can be flexibly set according to actual application or experience value, which is not limited in the present invention.

405. The VMM judges whether the VM is deadlocked according to the update frequency of the monitoring timer; if the VM does not have a deadlock, execute steps 401-405 in a loop, and if the VM is deadlocked, execute the following Steps 406-412.

406. The VMM sends a non-maskable interrupt to the VM where the deadlock occurs.

407. The VM responds to the non-maskable interrupt sent by the virtual machine monitor VMM.

Wherein, in the embodiment of the present invention, the VM responds to the non-maskable interrupt to end services running on the VM, specifically, to end all processes currently running on the VM.

408. The VM collects the last words of the VM, where the last words include stack information and register information of each virtual processor corresponding to the VM at the time of deadlock.

Specifically, the VM collects the last words from the hardware layer through the VMM. More specifically, the virtual machine collects the stack information of each virtual processor corresponding to the virtual machine at the time of the deadlock from the memory of the hardware layer through the VMM, and the processing information from the hardware layer The registers of the processor collect the register information of each virtual processor corresponding to the virtual machine when the deadlock occurs. In this embodiment, taking the processor as the CPU and the virtual processor as the VCPU as an example, the virtual machine collects the stack information of each VCPU corresponding to the virtual machine at the time of deadlock from the memory of the hardware layer through the VMM, and collects the deadlock information from the CPU register. The register information of each VCPU corresponding to the time-lock virtual machine.

It should be pointed out that both the special virtual machine and the service virtual machine can collect the last words from the hardware layer through the VMM. The embodiment of the present invention is applicable to the recovery after the deadlock of the special virtual machine, and is also applicable to the recovery after the deadlock of the service virtual machine.

After 406, the VMM configures virtual processor resources of the VM, specifically through the following steps 409 and 410.

409. The VMM acquires a predetermined number of virtual processors; the predetermined number is the number of virtual processors corresponding to the VM at the time of deadlock;

It should be noted that the present invention does not limit whether the predetermined number of virtual processors is the corresponding virtual processor before the deadlock, and the predetermined number of virtual processors may be the predetermined number of virtual processors used when the VM deadlock occurs. The virtual processor can also be a predetermined number of virtual processors acquired again. For example, when the virtual processor is a VCPU, VCPU1, VCPU2 and VCPU3 are used before the VM deadlock occurs, and can be obtained in step 409 VCPU1, VCPU2 and VCPU3 can also be obtained in step 409 to VCPU4, VCPU5 and VCPU6.

410. The VMM sets the second resource information of the predetermined number of virtual processors according to the recorded first resource information of one or more virtual processors corresponding to the VM, and the first resource information is consistent with the second The resource information is used to represent virtual processor resources with the same distribution, and both the first resource information and the second resource information include a credit value and a weight value.

It should be noted that the terms "first" and "second" used in all the embodiments of the present invention do not mean to limit the order, but are only for convenience of distinction.

In the embodiment of the present invention, the distribution of the first resource information is the same as that of the second resource information, and the newly obtained second resource information of the virtual processor is set according to the recorded first resource information.

However, the embodiment of the present invention does not limit the corresponding relationship between the newly obtained virtual processors and the recorded virtual processors. For example, if the predetermined number of virtual processors acquired are VCPU4, VCPU5 and VCPU6, then in In step 410, the resource information of VCPU4, VCPU5 and VCPU6 needs to be sequentially set as the recorded resource information of VCPU1, VCPU2 and VCPU3 at the time of deadlock, and the resource information of VCPU4, VCPU5 and VCPU6 can also be sequentially set as recorded deadlock time Resource information of VCPU3, VCPU2, and VCPU1.

More specifically, when the credit values of VCPU1, VCPU2, and VCPU3 are respectively 20, 30, and 50 at the time of the recorded deadlock, and the weight values are respectively 25, 35, and 40, when the acquired predetermined number of virtual processors is For VCPU4, VCPU5 and VCPU6, the credit values of VCPU4, VCPU5 and VCPU6 can be set to 20, 30 and 50 respectively, and the weight values can be set to 25, 35 and 40 respectively, or the credit values of VCPU4, VCPU5 and VCPU6 can be set respectively are 50, 20, and 30, and the weight values are set to 40, 25, and 35, respectively. In the embodiment of the present invention, the credit value and weight value of each virtual processor may be the same or different, depending on the algorithm adopted in the actual application, which is not limited in the present invention.

According to the recorded resource information of each virtual processor corresponding to the virtual machine at the time of deadlock, the resource information of the reacquired virtual processor is set to ensure that the reacquired predetermined number of virtual processors have the same value as before the VM deadlock. original information.

As shown in Figure 4(b), in a VMM (such as Xen Hypervisor), all virtual machine-related information has a specific data structure, and the data structure of each virtual machine and the data structure of the vcpu refer to each other. VMM (such as Xen Hypervisor) manages all data structures. After a deadlock occurs in a virtual machine, it directly reclaims the vcpu linked list corresponding to the virtual machine, and re-initializes the corresponding number of VCPUs according to the number of VCPU configurations recorded in the data structure of the virtual machine. VCPU, set the same credit value, weight, etc. as before, and re-schedule.

For example, VMM (such as Xen Hypervisor) reclaims the VCPU resources of the privileged virtual machine, and re-initializes the corresponding number of VCPUs. After initialization, the kernel boot program of the privileged virtual machine is scheduled to the first VCPU object (referred to as VCPU0) After loading and running, VCPU0 gets the time slice and executes the kernel boot program of the privileged virtual machine. Correspondingly, after the kernel of the privileged virtual machine is successfully started, the native driver is loaded, and the business process is resumed according to the last words information, and the front-end and back-end IO channels are restored.

After the virtual machine responds to the non-maskable interrupt and collects the last words, the following step 411 is performed.

411. The VMM schedules a first virtual processor from one or more virtual processors included in the configured virtual processor resource, and loads a kernel boot program of the virtual machine on the first virtual processor, wherein ,

The first virtual processor may be the first virtual processor randomly scheduled from one or more virtual processors included in the configured virtual processor resource, or the first virtual processor that is configured from the configured virtual processor Among the one or more virtual processors included in the processor resource, the scheduled virtual processor has the highest credit value and/or weight value.

The first virtual processor is the virtual processor that is configured first, and comes according to the order of configuration. In addition, selecting the virtual processor with the highest credit value and the highest weight value among the predetermined number of virtual processors, or the virtual processor with the highest credit value or the highest weight value facilitates the fast loading of the kernel boot program.

412. The VM resumes the service process according to the last words of the virtual machine.

Specifically, after the kernel boot program of the virtual machine is loaded, initialize the linked list of the virtual processor according to the stack information and register information of each virtual processor corresponding to the virtual machine at the time of deadlock, and according to the The stack information and the register information in the linked list identify the running instruction and address of the virtual machine before deadlock, so that the virtual machine continues to execute the next instruction after the running instruction after recovery.

It can be seen that the embodiment of the present invention provides a recovery method for a virtual machine after a deadlock. The monitoring timer is updated through the heartbeat information sent by the virtual machine, so that the virtual machine monitor can judge whether a deadlock occurs according to the update frequency of the monitoring timer. , which provides a basis for the deadlock recovery process; further, the virtual processor resources of the virtual machine are configured by sending a non-maskable interrupt to the virtual machine through the virtual machine monitor, and the virtual processor resources included in the configuration are completed Scheduling the first virtual processor among one or more virtual processors and loading the kernel boot program of the virtual machine on the first virtual processor, so that the virtual machine can resume the business process according to the collected last words, compared with the current There is a method of powering off and restarting the entire server in the prior art, but the present invention can avoid restarting the entire server and restore the operation of the virtual machine in time, thereby avoiding the phenomena such as long service disaster time and out-of-sync data.

Embodiment four

As shown in FIG. 5 , an embodiment of the present invention provides a virtual machine monitor 50 , including: an interrupt simulator 51 , a configuration unit 52 and a scheduling loading unit 53 .

The interrupt emulator 51 is configured to send a non-maskable interrupt to the virtual machine when the virtual machine is deadlocked.

The configuration unit 52 is configured to configure virtual processor resources of the virtual machine.

The scheduling loading unit 53 is configured to schedule the first virtual processor from one or more virtual processors included in the configured virtual processor resources after the virtual machine responds to the non-maskable interrupt and collects the last words. virtual processor and load the kernel boot program of the virtual machine on the first virtual processor, so that the virtual machine resumes the business process according to the last words collected; when the last words include a deadlock, the virtual machine corresponds to The stack information and register information of each virtual processor. in,

The first virtual processor is the first virtual processor randomly scheduled from one or more virtual processors included in the configured virtual processor resource, or the virtual processor configured from the Among the one or more virtual processors included in the server resource, the scheduled virtual processor has the highest credit value and/or weight value.

The resource information of the first virtual processor includes a first credit value and a first weight value.

Further, the configuration unit 52 is specifically configured to obtain a predetermined number of virtual processors, wherein the predetermined number is the number of virtual processors corresponding to the virtual machine at the time of deadlock; according to the recorded The first resource information of one or more virtual processors corresponding to the virtual machine, setting the second resource information of the predetermined number of virtual processors, the first resource information and the second resource information are used to represent For virtual processor resources with the same distribution, both the first resource information and the second resource information include a credit value and a weight value.

Further, the scheduling loading unit 53 is specifically configured to randomly schedule virtual processors from one or more virtual processors included in the configured virtual processor resources after the virtual machine responds to the non-maskable interrupt and collects the last words. dispatch the first virtual processor, or dispatch the virtual processor with the highest credit value and/or weight value from one or more virtual processors included in the configured virtual processor resources and Load the kernel boot program of the virtual machine on the server, so that the virtual machine resumes the business process according to the collected last words.

Further, as shown in FIG. 6 , the virtual machine monitor 50 further includes: a monitoring timer 54 , an updating unit 55 and a judging unit 56 .

The monitoring timer 54 is configured to receive heartbeat information sent by the virtual machine 70 according to a predetermined second time interval.

The updating unit 55 is configured to update the monitoring timer 54 according to the heartbeat information.

The judging unit 56 is configured to judge whether a deadlock occurs in the virtual machine 70 according to the update frequency of the monitoring timer.

The interrupt simulator 51 is specifically configured to send a non-maskable interrupt to the virtual machine 70 when the judging unit 56 determines that a deadlock occurs in the virtual machine.

It can be seen that the embodiment of the present invention provides a virtual machine monitor, which can update the monitoring timer according to the heartbeat information sent by the virtual machine, and judge whether the virtual machine is deadlocked according to the update frequency of the monitoring timer. The lock recovery process provides a basis. Further, the virtual machine monitor sends a non-maskable interrupt to the virtual machine and configures the virtual processor resources of the virtual machine, and one or more Scheduling the first virtual processor among the virtual processors and loading the kernel boot program of the virtual machine on the first virtual processor, so as to help the virtual machine recover the business process according to the collected last words, compared with the existing According to the method of powering off and restarting the entire server in the technology, the present invention can avoid restarting the entire server and restore the operation of the virtual machine in time, thus avoiding the phenomena such as business disaster time is too long and data is not synchronized.

As shown in FIG. 7 , the embodiment of the present invention provides a virtual machine 70 , and the virtual machine 70 may be a privileged virtual machine 701 or a service virtual machine 702 .

The virtual machine 70 includes: an interrupt processing unit 71 , a collection unit 72 and a service recovery unit 73 .

The interrupt processing unit 71 is configured to respond to the non-maskable interrupt sent by the virtual machine monitor VMM;

The collection unit 72 is configured to collect the last words of the virtual machine, wherein the last words include stack information and register information of each virtual processor corresponding to the virtual machine at the time of deadlock;

The service restoration unit 73 is configured to restore the service process according to the last words of the virtual machine after the kernel boot program of the virtual machine is loaded.

Further, the collection unit 72 is specifically configured to collect the last words of the virtual machine from the hardware layer through the VMM.

The service recovery unit 73 is specifically configured to initialize the linked list of virtual processors according to the stack information and register information of each virtual processor corresponding to the virtual machine when the virtual machine is deadlocked after the kernel boot program of the virtual machine is loaded. , and identify the running instruction and address of the virtual machine before deadlock according to the stack information and the register information in the linked list, so that the virtual machine continues to execute the next instruction of the running instruction after recovery.

Further, as shown in FIG. 8 , the virtual machine 70 further includes: a heartbeat generating and sending unit 74 .

The heartbeat generating and sending unit 74 is configured to generate heartbeat information according to a predetermined first time interval; in the calling mode provided by the VMM, send the heartbeat information to the VMM according to a predetermined second time interval, so that The VMM judges whether a deadlock occurs in the virtual machine. Specifically, the calling mode is a hypercall.

It can be seen that the embodiment of the present invention provides a virtual machine, which generates heartbeat information according to a predetermined first time interval, so that the virtual machine monitor can make a judgment on whether a deadlock occurs, and provides a basis for deadlock recovery. Further , the virtual machine responds to the non-maskable interrupt sent by the virtual machine monitor and collects the last words of the virtual machine. After the virtual machine monitor completes the loading of the boot program, the recovery of the business process can be realized. Compared with the prior art, the entire server In the method of restarting after power off, the present invention can avoid restarting the entire server and restore the operation of the virtual machine in time, thereby avoiding the phenomena such as long service disaster time and out-of-sync data.

The embodiment of the present invention also provides a host machine 90, including: a virtual machine monitor VMM50 and a privileged virtual machine VM701, wherein the VMM is used to send Non-maskable interrupts and configuring the virtual processor resources of the privileged virtual machine; after the privileged virtual machine responds to the non-maskable interrupts and collects the last words, one or more of the configured virtual processor resources include Schedule the first virtual processor among the virtual processors and load the kernel boot program corresponding to the privileged virtual machine on the first virtual processor.

The privileged virtual machine is used to respond to the non-maskable interrupt sent by the VMM; collect the last words of the privileged virtual machine, and the last words include stack information and register information of each virtual processor corresponding to the privileged virtual machine at the time of deadlock ; After the kernel boot program of the privileged virtual machine is loaded, resume the service process according to the last words of the privileged virtual machine.

In addition, it should be noted that in the embodiment of the present invention, the virtual machine monitor and the privileged virtual machine cooperate to form the host machine. In addition, the host machine may also include the virtual machine monitor instead of the privileged virtual machine.

It can be seen that the embodiment of the present invention provides a host computer that sends a non-maskable interrupt to a virtual machine through a virtual machine monitor and configures virtual processor resources of the virtual machine, and one of the configured virtual processor resources includes or a plurality of virtual processors to schedule the first virtual processor and load the kernel boot program of the virtual machine on the first virtual processor, so that the virtual machine can resume the business process according to the collected last words, in the privileged virtual machine In the case of a deadlock, compared with the method of powering off and restarting the entire server in the prior art, the present invention can restore the operation of the privileged virtual machine in time, thereby avoiding restarting the entire server, thereby avoiding the business disaster time being too long, Data out of sync and so on.

As shown in FIG. 10 , an embodiment of the present invention further provides a computer system, including: at least one computing node 100 as described in the foregoing embodiments.

It can be seen that in the computer system provided by the embodiment of the present invention, specifically in each computer node, the virtual machine monitor sends a non-maskable interrupt to the virtual machine and configures the virtual processor resources of the virtual machine, and the virtual machine responds to the non-maskable interrupt. Shield the interrupt and collect the last words, then schedule a certain virtual processor (such as the aforementioned first virtual processor or second virtual processor) and load the kernel boot program of the virtual machine on the certain virtual processor, so that the virtual machine can recover the business process according to the collected last words; on the one hand, when a deadlock occurs in the privileged virtual machine , compared to the method of powering off the entire server in the prior art, the present invention detects the deadlock of the privileged virtual machine and restores the privileged virtual machine online through the VMM, which can restore the operation of the privileged virtual machine in time and maintain the running state of the entire computing node, thereby It avoids restarting the entire computing node such as the entire server, thereby avoiding the business disaster for a long time, data out of synchronization, etc.; in addition, due to the rapid recovery of the privileged virtual machine, it automatically restores other business virtual machines to access the network and storage IO The IO function of resources will not affect other business virtual machines; on the other hand, when a deadlock occurs in a business virtual machine, the present invention can restore the business virtual machine online through the VMM after detecting the deadlock of the business virtual machine, avoiding the entire The server restarts to restore the operation of the business virtual machine in time, realizes the rapid recovery of the business in the virtual machine, and avoids the phenomenon that the business disaster of the business virtual machine takes too long and the data is not synchronized.

It should be noted that the terms first and second used in the description of the foregoing embodiments do not mean to limit the sequence, but are only for convenience of distinction.

It should be noted that for the foregoing method embodiments, for the sake of simple description, they are expressed as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence. Because of the present invention, certain steps may be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification belong to preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

In the foregoing embodiments, the descriptions of each embodiment have their own emphases, and for parts not described in detail in a certain embodiment, reference may be made to relevant descriptions of other embodiments.

Those of ordinary skill in the art can understand that all or part of the steps for realizing the above-mentioned method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the It includes the steps of the above method embodiments; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk and other various media that can store program codes.

The above is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Anyone skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present invention. Should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the protection scope of the claims.

Claims (21)

1. A recovery method after deadlock of a virtual machine is characterized by comprising the following steps:

sending an unmasked interrupt to a virtual machine and configuring virtual processor resources of the virtual machine under the condition that the virtual machine is deadlocked;

after the virtual machine responds to the unmasked interrupt and collects the collected complete telepresence, a first virtual processor is dispatched from one or more virtual processors included in the configured virtual processor resources, and a kernel bootstrap program of the virtual machine is loaded on the first virtual processor, so that the virtual machine can recover a business process according to the collected telepresence.

2. The method of claim 1,

the configuring the virtual processor resources of the virtual machine comprises:

acquiring a preset number of virtual processors, wherein the preset number is the number of the virtual processors corresponding to the virtual machine in deadlock;

and setting second resource information of the predetermined number of virtual processors according to recorded first resource information of one or more virtual processors corresponding to the virtual machine, wherein the first resource information and the second resource information are used for representing virtual processor resources with the same distribution situation, and both the first resource information and the second resource information comprise credit values and weighted values.

3. The method of claim 1,

the first virtual processor is a first virtual processor randomly scheduled from one or more virtual processors included in the configured virtual processor resource, or a virtual processor with the highest credit value and/or weight value scheduled from one or more virtual processors included in the configured virtual processor resource.

4. The method according to any one of claims 1 to 3, further comprising:

receiving heartbeat information sent by the virtual machine according to a preset second time interval;

updating a monitoring timer according to the heartbeat information;

and judging whether the virtual machine is deadlocked or not according to the updating frequency of the monitoring timer, and if the virtual machine is deadlocked, executing the step of sending the unmasked interrupt to the virtual machine.

5. A recovery method after deadlock of a virtual machine is characterized by comprising the following steps:

responding to a non-maskable interrupt sent by a Virtual Machine Monitor (VMM);

collecting a telepresence of a virtual machine, wherein the telepresence comprises stack information and register information of each virtual processor corresponding to the virtual machine during deadlock;

and after the kernel bootstrap program of the virtual machine is loaded, recovering the business process according to the terminal-bound heritage of the virtual machine.

6. The method of claim 5, wherein collecting the telepresence of the virtual machine comprises: and collecting a last-in-place talent of the virtual machine from a hardware layer through the VMM, wherein the last-in-place talent comprises stack information and register information of each virtual processor corresponding to the virtual machine during deadlock.

7. The method of claim 5, wherein after the loading of the kernel boot program of the virtual machine is completed, resuming the business process according to the telepresence of the virtual machine comprises:

after the kernel bootstrap program of the virtual machine is loaded, initializing a linked list of the virtual processor according to stack information and register information of each virtual processor corresponding to the virtual machine in deadlock, and identifying an operation instruction and an address of the virtual machine before deadlock according to the stack information and the register information in the linked list, so that the virtual machine continues to execute a next instruction of the operation instruction after recovery.

8. The method of any of claims 5 to 7, further comprising:

generating heartbeat information according to a preset first time interval;

and sending the heartbeat information to the VMM according to a preset second time interval under a calling mode provided by the VMM so that the VMM can judge whether the virtual machine is deadlocked or not.

9. The method of claim 8, wherein the calling mode is a hypercall.

10. A virtual machine monitor, comprising:

the interrupt simulator is used for sending the unmasked interrupt to the virtual machine under the condition that the virtual machine is deadlocked;

a configuration unit, configured to configure virtual processor resources of the virtual machine;

and the scheduling and loading unit is used for scheduling a first virtual processor from one or more virtual processors included in the configured virtual processor resources and loading a kernel bootstrap program of the virtual machine on the first virtual processor after the virtual machine responds to the unmasked interrupt and collects the collected telepresence, so that the virtual machine can recover a business process according to the collected telepresence.

11. The virtual machine monitor according to claim 10, wherein the configuration unit is specifically configured to obtain a predetermined number of virtual processors, where the predetermined number is the number of virtual processors corresponding to the virtual machine in the deadlock; and setting second resource information of the predetermined number of virtual processors according to recorded first resource information of one or more virtual processors corresponding to the virtual machine, wherein the first resource information and the second resource information are used for representing virtual processor resources with the same distribution situation, and both the first resource information and the second resource information comprise credit values and weighted values.

12. The virtual machine monitor according to claim 10,

the scheduling and loading unit is specifically configured to, after the virtual machine responds to the unmasked interrupt and collects a complete epigraph, randomly schedule a first virtual processor from one or more virtual processors included in the configured virtual processor resource, or schedule a virtual processor with a highest credit value and/or weight value from one or more virtual processors included in the configured virtual processor resource and load a kernel bootstrap of the virtual machine on the virtual processor, so that the virtual machine resumes a business process according to the collected epigraph.

13. A virtual machine monitor according to any one of claims 10 to 12, wherein the virtual machine monitor further comprises:

the monitoring timer is used for receiving heartbeat information sent by the virtual machine according to a preset second time interval;

the updating unit is used for updating the monitoring timer according to the heartbeat information;

the judging unit is used for judging whether the virtual machine is deadlocked or not according to the updating frequency of the monitoring timer;

the interrupt simulator is specifically configured to send an unmasked interrupt to the virtual machine when the determination unit determines that the virtual machine is deadlocked.

14. A virtual machine, comprising:

the interrupt processing unit is used for responding to the unmasked interrupt sent by the virtual machine monitor VMM;

the system comprises a collecting unit, a processing unit and a processing unit, wherein the collecting unit is used for collecting the terminal talent of the virtual machine, and the terminal talent comprises stack information and register information of each virtual processor corresponding to the virtual machine when the terminal talent is deadlocked;

and the service recovery unit is used for recovering the service process according to the terminal-bound heritage of the virtual machine after the kernel bootstrap program of the virtual machine is loaded.

15. The virtual machine according to claim 14, wherein the collecting unit is specifically configured to collect, by the VMM, a lifetime statement of the virtual machine from a hardware layer, wherein the lifetime statement includes stack information and register information of each virtual processor corresponding to the virtual machine at the time of the deadlock.

16. The virtual machine according to claim 14, wherein the service recovery unit is specifically configured to initialize a linked list of the virtual processors according to stack information and register information of each virtual processor corresponding to the virtual machine in the deadlock, and identify an operation instruction and an address before the deadlock of the virtual machine according to the stack information and the register information in the linked list, after the kernel boot program of the virtual machine is completely loaded, so that the virtual machine continues to execute a next instruction of the operation instruction after recovery.

17. The virtual machine according to any one of claims 14 to 16, further comprising:

a heartbeat generating and sending unit, which is used for generating heartbeat information according to a preset first time interval; and sending the heartbeat information to the VMM according to a preset second time interval under a calling mode provided by the VMM so that the VMM can judge whether the virtual machine is deadlocked or not.

18. A host machine comprising a virtual machine monitor and a privileged virtual machine, wherein the virtual machine monitor is the virtual machine monitor of any one of claims 10 to 13 and the privileged virtual machine is the virtual machine of any one of claims 14 to 17.

19. A computing node, wherein the computing node comprises: a hardware layer, a Virtual Machine Monitor (VMM) running on top of the hardware layer, and a virtual machine running on top of the VMM; the virtual machines include a privileged virtual machine and at least one business virtual machine, wherein:

the VMM is configured to send a non-maskable interrupt to the privileged virtual machine and to configure virtual processor resources of the privileged virtual machine if the privileged virtual machine is deadlocked; after the privileged virtual machine responds to the unmasked interrupt and collects and completes the terminal heritage, scheduling a first virtual processor from one or more virtual processors included in the configured virtual processor resources and loading a kernel bootstrap program corresponding to the privileged virtual machine on the first virtual processor;

the privileged virtual machine is to respond to a non-maskable interrupt sent by the VMM; collecting the telepresence of the privileged virtual machine, wherein the telepresence comprises stack information and register information of each virtual processor corresponding to the privileged virtual machine during deadlock; and after the kernel bootstrap program of the privileged virtual machine is loaded, recovering the business process according to the terminal heritage of the privileged virtual machine.

20. The apparatus of claim 19,

the VMM is also used for sending an unmasked interrupt to the service virtual machine and configuring the virtual processor resource of the service virtual machine under the condition that the service virtual machine is deadlocked; after the business virtual machine responds to the unmasked interrupt and collects and completes the terminal language, scheduling a second virtual processor from one or more virtual processors included in the configured virtual processor resources and loading a kernel bootstrap program corresponding to the business virtual machine on the second virtual processor;

the business virtual machine is used for responding to the non-maskable interrupt sent by the VMM; collecting a terminal language of the service virtual machine, wherein the terminal language comprises stack information and register information of each virtual processor corresponding to the service virtual machine during deadlock; and after the kernel bootstrap program of the service virtual machine is loaded, recovering the service process according to the terminal-up heritage of the service virtual machine.

21. A computer system, comprising: at least one computing node as claimed in any one of claims 19 to 20.

Images