CN115277006A - Management method and device for private computing nodes - Google Patents

Abstract

The embodiment of the specification provides a method and a device for managing private computing nodes. The method comprises the following steps: firstly, acquiring first state information of a first privacy computing node in a kubernets cluster, wherein the first state information indicates that the node state is abnormal; the first privacy computation node belongs to a first namespace corresponding to a first mechanism; then, under the condition that the duration time of the abnormal state of the node is determined to reach a first preset time length based on the first state information, adding a target stain mark to the first privacy computing node; the first preset duration is set for the first mechanism; next, in response to the addition of the target taint marker, scheduling a pod served on the first private computing node to a second private computing node in the kubernets cluster that is located under the first namespace; the second private computing node is in a normal state and does not have the target taint mark.

Description

Management method and device for private computing nodes

technical field

One or more embodiments of this specification relate to the field of computer technology, and in particular, to a method and device for managing privacy computing nodes.

Background technique

Kubernetes is an open source, industrial-grade container orchestration platform. Its emergence has promoted the popularization and implementation of popular technologies such as microservice architecture, making development, operation and maintenance, and delivery easier and easier. Although the standard kubernetes already provides many powerful functions, for some special application scenarios, the standard kubernetes can no longer meet the needs.

At present, for the business scenario of private multi-party secure computing, there is no kubernetes node management solution that can meet the special needs of this scenario.

Contents of the invention

One or more embodiments of this specification describe a method and device for managing private computing nodes, which can manage private computing nodes under the organization in a more fine-grained manner, trigger the rescheduling of services on invalid nodes in time, and ensure the availability and stability of services sex.

According to the first aspect, a method for managing a privacy computing node is provided, including: obtaining first status information of a first privacy computing node in a kubernetes cluster, which indicates that the node status is abnormal; the first privacy computing node belongs to the first institution corresponding the first namespace; when it is determined based on the first state information that the duration of the abnormal state of the node reaches a first preset duration, add a target taint mark to the first privacy computing node; the second A preset duration is set for the first organization; in response to the addition of the target taint mark, dispatch the pod served on the first privacy computing node to the kubernetes cluster under the first namespace The second privacy computing node; the state of the second privacy computing node is normal and does not have the target taint mark.

In one embodiment, the first status information further includes a first time stamp; wherein, when it is determined based on the first status information that the duration of the node status abnormality reaches a first preset duration, it is Adding a target taint mark to the first privacy computing node includes: obtaining second state information of the first privacy computing node after the first preset time period has elapsed since the first timestamp; If the second state information still indicates that the state of the node is abnormal, add the target taint mark for the first privacy computing node.

In one embodiment, obtaining the first state information of the first privacy computing node in the kubernetes cluster includes: listening to the first event of a state change of the first privacy computing node, and the information of the first event includes the First state information; wherein, when it is determined based on the first state information that the duration of the abnormal state of the node reaches a first preset duration, adding a target taint mark to the first privacy computing node includes: Add the Target taint marker.

In one embodiment, after adding the target taint mark to the first privacy computing node, the method further includes: obtaining third state information of the first privacy computing node, which indicates that the node state is normal; If the third state information determines that the duration of the normal state of the node reaches a second preset duration, the target taint mark is removed.

In a specific embodiment, the third status information further includes a second time stamp; wherein, when it is determined based on the third status information that the duration of the normal state of the node reaches the second preset duration Next, removing the target taint mark includes: obtaining fourth state information of the first privacy computing node after the second preset time period has elapsed since the second time stamp; in the fourth state information If it still indicates that the node status is normal, remove the target taint mark.

In another specific embodiment, obtaining the third state information of the first privacy computing node includes: listening to a second event that the state of the first privacy computing node changes, and the information of the second event includes The third state information; wherein, when it is determined based on the third state information that the duration of the normal state of the node reaches the second preset duration, removing the target taint mark includes: Within the second preset time period elapsed from the second event, if the event that the state of the first privacy computing node changes to abnormal is not detected, the target taint mark is removed.

In one embodiment, the kubernetes cluster includes a plurality of computing nodes; before obtaining the first status information of the first private computing node in the kubernetes cluster, the method further includes: filtering out bands from the plurality of computing nodes A plurality of privacy computing nodes with privacy computing labels; the plurality of privacy computing nodes belong to different namespaces corresponding to different institutions, including the first privacy computing node and the second privacy computing node.

In one embodiment, the execution subject of the method is a privacy computing node controller; wherein, when scheduling the pod served on the first privacy computing node to the pod located in the first namespace in the kubernetes cluster After the second privacy computing node, the method further includes: deleting the first privacy computing node from the kubernetes cluster in response to calling a target API provided by the privacy computing node controller.

In one embodiment, the first status information further includes a first timestamp; when scheduling the pod served on the first privacy computing node to the first pod under the first namespace in the kubernetes cluster After the second privacy computing node, the method further includes: acquiring fifth status information of the first privacy computing node after a third preset time period has elapsed since the first timestamp; when the fifth status information indicates When the node state is abnormal, the first privacy computing node is deleted from the kubernetes cluster.

According to the second aspect, there is provided a management device for a private computing node, including: a state acquisition unit configured to obtain first state information of a first private computing node in a kubernetes cluster, which indicates that the state of the node is abnormal; the first private computing node The node belongs to the first namespace corresponding to the first organization; the taint adding unit is configured to, when it is determined based on the first state information that the duration of the abnormal state of the node reaches a first preset duration, add the taint to the second A privacy computing node adds a target taint mark, and in response to the addition of the target taint mark, schedules the pod served on the first privacy computing node to the second pod under the first namespace in the kubernetes cluster. A privacy computing node; the first preset duration is set for the first organization; the second privacy computing node is in a normal state and does not have the target taint mark.

According to a third aspect, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed in a computer, the computer is caused to execute the method of the first aspect.

According to a fourth aspect, there is provided a computing device, including a memory and a processor, where executable codes are stored in the memory, and when the processor executes the executable codes, the method of the first aspect is implemented.

By adopting the management method of privacy computing nodes disclosed in the embodiment of this specification, the privacy computing nodes under the organization can be managed in a more fine-grained manner, the invalid nodes among them can be found in time and the target taint mark can be added, so as to trigger the restart of the pod serving on it in time. Scheduling to ensure the availability and stability of services. Furthermore, it is also possible to clean up and recycle invalid nodes for privacy computing, and remove the target taint mark for private computing nodes that have returned to a normal state and are in a stable state, so that they can be put into use again and give full play to the use efficiency of private computing nodes.

Description of drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For Those of ordinary skill in the art can also obtain other drawings based on these drawings without making creative efforts.

FIG. 1 shows an example diagram of implementing a management solution of a privacy computing node according to an embodiment;

Fig. 2 shows a schematic flowchart of a management method of a privacy computing node according to an embodiment;

Fig. 3 shows a schematic structural diagram of a management device of a privacy computing node according to an embodiment.

Detailed ways

The solutions provided in this specification will be described below in conjunction with the accompanying drawings.

As mentioned above, for the business scenario of private multi-party secure computing, there is no kubernetes node management solution that can meet the special needs of this scenario. Specifically, due to the business particularity of private multi-party secure computing, each organization has a dedicated set of private computing nodes. When some privacy computing nodes are unavailable, it is necessary to discover these invalid privacy computing nodes in time, and reschedule the services on these privacy computing nodes to other available nodes in time, otherwise the availability of services will be affected.

Based on the above observations and analysis, in order to manage the private computing nodes under each organization in a more fine-grained manner, including timely discovery and cleaning of invalid nodes, the inventor proposes a management scheme for private computing nodes, which is used to manage the private computing nodes under each organization. The private computing nodes guarantee the availability and stability of services.

Fig. 1 shows an implementation example diagram of a management scheme of a private computing node according to an embodiment. As shown in Fig. 1, the private computing nodes in the kubernetes cluster are dynamically monitored, and when a certain private computing node ( For example, if the status of the privacy computing node S _A1 in organization A is abnormal, it will be continuously monitored. The node is marked with a predetermined taint (in Figure 1, the taint mark is visually illustrated as an irregularly shaped gray color block), thereby triggering the rescheduling of the service pod on the invalid node to other available privacy computing nodes under the same organization (for example, the organization On the privacy computing node S _A2 in A.

The implementation steps of the above solution will be introduced in detail below in conjunction with more embodiments.

Fig. 2 shows a schematic flowchart of a method for managing a private computing node according to an embodiment. The subject of execution of the method may be any device, server, or device cluster with computing and processing capabilities, for example, it may be a private computing node controller . As shown in Figure 2, the method comprises the following steps:

Step S210, obtain the first status information of the first privacy computing node in the kubernetes cluster, which indicates that the node status is abnormal, and the first privacy computing node belongs to the first namespace corresponding to the first organization; step S220, based on the first If the state information determines that the duration of the abnormal state of the node reaches a first preset duration, add a target taint mark to the first privacy computing node; the first preset duration is specific to the first institution Setting; step S230, in response to the addition of the target taint mark, dispatching the pod served on the first privacy computing node to the second privacy computing node in the kubernetes cluster under the first namespace, so The state of the second privacy computing node is normal and does not have the target taint mark.

The expansion of the above steps is as follows:

First, in step S210, the first status information of the first privacy computing node in the kubernetes cluster is acquired. It should be noted that in the kubernetes cluster disclosed in the embodiment of this specification, there are multiple private computing nodes including the first private computing node, and the method disclosed in the embodiment of this specification aims to manage these private computing nodes.

The multiple privacy computing nodes mentioned above belong to different organizations. In other words, a single privacy computing node belongs to a certain organization and can only be scheduled and used by the organization to perform computing tasks in the organization. Further, on the one hand, in terms of technical implementation, different organizations use different namespaces (kubernetes namespace) for resource isolation, and correspondingly, different organizations are expressed as different namespaces in kubernetes. For brevity and distinction, the organization and namespace to which the first privacy computing node belongs are called the first organization and the first namespace, respectively. On the other hand, computing tasks performed on private computing nodes can be called private computing tasks. Typically, private computing tasks can be the computing tasks that multiple organizations need to perform when performing multi-party secure computing. For example, in the scenario of federated learning, each organization needs to use its own local private data to perform local deployment. train.

Usually, a private computing node also has a private computing label, which enables it to be distinguished from non-private computing nodes in the kubernetes cluster. It can be understood that the label content of the privacy computing label can be customized by the staff, for example, defined as "kubernetes.io/nueva-agent". In addition, both privacy computing nodes and non-privacy computing nodes can be implemented as virtual machines or physical machines.

In one embodiment, in the scenario where common computing nodes are also included in the kubernetes cluster, before performing this step, the method may further include: filtering out the private computing tags from the plurality of computing nodes included in the kubernetes cluster The above multiple privacy computing nodes. In this way, the multiple privacy computing nodes located can be accurately supervised in the future.

Above, a plurality of privacy computing nodes including the first privacy computing node are mainly introduced.

For the implementation of this step, in one embodiment, the status information of each privacy computing node in the above-mentioned multiple privacy computing nodes can be obtained, and the state information shows the status of several (several refer to one or more) privacy computing nodes among them Abnormal (or invalid state), for example, the node state is Not Ready. At this time, the above-mentioned first privacy computing node may be any computing node among the several privacy computing nodes. Exemplarily, the batch acquisition of state information of multiple privacy computing nodes can be realized based on the list mechanism of kubernetes.

In another embodiment, an event that a state of a certain computing node among the plurality of privacy computing nodes is changed is monitored. At this time, the above-mentioned first privacy computing node may be the certain computing node, and this event may be called the first event, and the changed state information included in the first event is called the first state information, which indicates the node state changed to exception. Exemplarily, the monitoring of the first event may be implemented based on the list-watch mechanism of kubernetes.

In this way, the status monitoring of multiple privacy computing nodes can be realized, and the first status information indicating that the first privacy computing node among them is in an abnormal state can be obtained.

Then, in step S220, if it is determined based on the above-mentioned first state information that the duration of the abnormal state of the node reaches the first preset duration, a target taint mark is added to the first privacy computing node. It should be noted that the first preset duration is set for the first organization. Exemplarily, this setting can be realized through a related API provided by the privacy computing node controller. For example, the upper platform passes in the name of the organization and Complete the management time setting for a specific organization. In this way, configuring different node management time for different organizations can meet the actual needs of different organizations and realize more fine-grained management of privacy computing invalid nodes under organizations.

After the first state information is obtained, the judgment on the duration of the abnormal state of the first privacy computing node can be realized by means of query or monitoring. Specifically, the first status information corresponds to the first timestamp, that is, at the moment corresponding to the first timestamp, the status of the first privacy computing node is abnormal. In an implementation situation, the first state information is included in the information of the first event, and at this time, the first time stamp also corresponds to the occurrence time of the first event.

In an embodiment, after a first preset period of time elapses from the first timestamp, the second state information of the first privacy computing node may be queried in real time. For example, the current status information of the first privacy computing node can be queried according to its name, for example, the query can be performed through the get method, and for example, the first privacy computing node can be located through the list resource list in kubernetes, and then its status can be obtained information. Further, when the second state information still indicates that the state of the node is abnormal, add a target taint mark for the first privacy computing node. Otherwise, the addition of the target taint mark is not performed. It should be understood that the target taint mark can be defined by the worker, such as nueva.alipay.com/agent=NoExecute.

In another embodiment, within the first preset period of time elapsed from the first time stamp, no event that the state of the first privacy computing node changes to normal is detected, and at this time it can be determined that the state of the first privacy computing node is still Abnormally, then, add a target taint mark for the first privacy computing node. If the event that the status of the first privacy computing node changes to normal is detected before the first preset duration, at this time, there is no need to add the target taint mark.

For the above target taint mark, indicating that the corresponding node cannot run any service pod, the privacy computing node after adding the target taint mark can be called a privacy computing invalid node.

Afterwards, in step S230, in response to the above-mentioned addition of the target taint mark, all service pods on the first privacy computing node are scheduled to available privacy computing nodes under the same organization (corresponding to the same namespace). Wherein, the state of the available privacy computing node is normal (for example, the node state is Ready), and it does not carry the above-mentioned target taint mark. In this way, the pod service can be guaranteed to run normally and stably. In addition, the available privacy computing nodes that are scheduled to run the pod service may be referred to as second privacy computing nodes herein.

In one embodiment, according to the resources that pods need to occupy on the first privacy computing node, and the resource usage of the available privacy computing nodes in the first namespace, such as remaining memory, it can be determined that the pod on the first privacy computing node The pods of all services of are scheduled to one or more second privacy computing nodes. In one example, all pods can be scheduled to the same second privacy computing node; in another example, all pods can be divided into two groups and scheduled to two second privacy computing nodes respectively. In another implementation, for the case that the first privacy computing node includes multiple pods running several services, the multiple pods can be correspondingly scheduled to multiple second privacy computing nodes, so that by pairing the pods running several services Decentralized deployment of multiple pods can effectively guarantee the stability of the service.

It should be noted that, the execution subject of the above step S230 may be the default scheduler of kubernetes or a customized scheduler.

From the above, timely scheduling of pods serving on nodes with invalid privacy computing can be realized, ensuring the availability of services.

Above, the discovery of invalid nodes for privacy computing and the scheduling of pods served on them are introduced.

According to another embodiment, considering that there are various reasons for the privacy computing node to be in an invalid state, in some cases, for example, the node is invalid due to network jitter, when the network is restored, the privacy computing node can follow To restore the normal state, therefore, in the above method, it is also possible to design the removal of the target taint mark of the privacy computing node that has been restored to normal, and restore the privacy computing node to be available, so that it can be put into use again.

Therefore, after the above step S220, the method may further include: obtaining third status information of the first privacy computing node, which indicates that the node status is normal; In the case of the second preset duration, the target taint mark on the first privacy computing node is removed. Further, in a specific embodiment, the obtained third state information also includes a second time stamp, and after the above-mentioned second preset time period has elapsed since the second time stamp, the fourth state of the first privacy computing node is obtained information; if the fourth state information still indicates that the node state is normal, remove the target taint mark carried by the first privacy computing node; otherwise, do not remove it. In another specific embodiment, the third state information is obtained by listening to the second event that the state of the first privacy computing node changes, so that within the above-mentioned second preset time period since the second event , if the event that the state of the first privacy computing node changes to abnormal is not detected, the target taint mark is removed. In this way, the removal of the target taint mark of the first privacy computing node restored to a normal state can restore the first privacy computing node to be available.

According to yet another embodiment, after an invalid privacy computing node is found, it may also be cleaned up, so as to recycle the computing node. Specifically, the execution subject of the method may be called a privacy computing node controller. In one embodiment, the controller may provide an application programming interface (Application Programming Interface, API for short) to the outside world (for example, to the kubernetes default scheduler). ), used to support state information query and delete operations for privacy computing nodes. Therefore, in a specific embodiment, after performing the above step S220, the method may further include: receiving a first call to the first target API provided by the privacy computing node controller, and the call information includes a corresponding The identification parameter of the namespace of an organization, based on the first call, provides the state data of the privacy computing node under the organization. Exemplarily, the status data may include the duration of the abnormal and invalid status of the first privacy computing node. In another specific embodiment, after performing the above step S220, the method may further include: receiving a second call to the second target API provided by the privacy computing node controller, where the call information includes the first privacy computing The identification parameter of the node and the deletion operation for the first privacy computing node, based on the second call, delete the first privacy computing node from the kubernetes cluster.

In another embodiment, the privacy computing node controller can be designed with an automatic cleaning mechanism for privacy invalid computing nodes. After the above step S220, the method may further include: after adding the target taint mark to the first privacy computing node, if it is detected that the duration of the abnormal state of the node reaches the corresponding preset duration threshold, the first privacy computing node Removed from the kubernetes cluster. On the other hand, in a specific embodiment, a switch of the automatic cleaning function can also be designed to control the activation or deactivation of the automatic cleaning function.

In this way, the cleaning of invalid nodes for privacy calculation can be realized.

To sum up, using the management method of privacy computing nodes disclosed in the embodiment of this specification, the privacy computing nodes under the organization can be managed in a more fine-grained manner, and the invalid nodes can be found in time and the target taint mark can be added, so as to trigger the service on it in time. The rescheduling of pods ensures the availability and stability of services. Furthermore, it is also possible to clean up and recycle invalid nodes for privacy computing, and remove the target taint mark for private computing nodes that have returned to a normal state and are in a stable state, so that they can be put into use again and give full play to the use efficiency of private computing nodes.

Corresponding to the above-mentioned management method, the embodiment of this specification also discloses a management device for a privacy computing node. FIG. 3 shows a schematic structural diagram of a management device for a privacy computing node according to an embodiment. As shown in FIG. 3, the device 300 includes the following units:

The state obtaining unit 310 is configured to obtain first state information of a first private computing node in the kubernetes cluster, which indicates that the state of the node is abnormal; the first private computing node belongs to the first namespace corresponding to the first institution. The taint adding unit 320 is configured to add a target taint mark to the first privacy computing node in response to When the target taint mark is added, dispatching the pod served on the first privacy computing node to the second privacy computing node located in the first namespace in the kubernetes cluster; the first preset duration Set for the first organization; the second privacy computing node is in a normal state and does not have the target taint mark.

In an embodiment, the first status information further includes a first timestamp. The taint adding unit 320 is specifically configured to: obtain the second status information of the first privacy computing node after the first preset time period has elapsed since the first timestamp; when the second status information is still Adding the target taint mark to the first privacy computing node when it indicates that the state of the node is abnormal.

In one embodiment, the state acquisition unit 310 is specifically configured to: listen to a first event of a state change of the first privacy computing node, and the information of the first event includes the first state information; the taint addition unit 320 The specific configuration is: within the first preset period of time since the first event, if the event that the state of the first privacy computing node changes to normal is not detected, the first privacy A compute node adds the target taint flag.

In one embodiment, the device further includes a service scheduling unit 330 configured to, in response to the addition of the target taint mark, schedule the service pod on the first privacy computing node to the kubernetes cluster located at the A second privacy computing node under the first namespace.

In one embodiment, the apparatus 300 further includes: a state reacquisition unit 340, configured to obtain third state information of the first privacy computing node, which indicates that the node state is normal; a stain removal unit 350, configured to obtain the third state information based on If the third state information determines that the duration of the normal state of the node reaches a second preset duration, the target taint mark is removed.

In a specific embodiment, the third state information further includes a second time stamp; the stain removal unit 350 is specifically configured to: after the second preset time period elapses from the second time stamp, obtain the The fourth status information of the first privacy computing node; when the fourth status information still indicates that the node status is normal, remove the target taint mark.

In a specific embodiment, the state re-acquisition unit 340 is specifically configured to: listen to a second event in which the state of the first privacy computing node changes, and the information of the second event includes the third state information; The removal unit 350 is specifically configured to: remove the first privacy computing node if no event is detected that the state of the first privacy computing node changes to abnormal within the second preset time period since the second event. Target taint marker.

In one embodiment, the first privacy computing node includes a plurality of pods running several services; wherein, the service scheduling unit 330 is specifically configured to: correspondingly schedule the plurality of pods to a plurality of second privacy computing nodes.

In one embodiment, the kubernetes cluster includes a plurality of computing nodes; the apparatus 300 further includes: a privacy node filtering unit 360 configured to filter out a plurality of privacy nodes with privacy computing labels from the plurality of computing nodes; A computing node; the plurality of privacy computing nodes belong to different organizations, including the first privacy computing node and the second privacy computing node.

In one embodiment, the method is executed by a privacy computing node controller; wherein, the apparatus 300 further includes: an invalid node removal unit 370, and in a specific embodiment, the invalid node removal unit 370 is configured to respond to Deleting the first privacy computing node from the kubernetes cluster upon invoking the target API provided by the privacy computing node controller. In another specific embodiment, the invalid node clearing unit 370 is configured to: obtain the fifth status information of the first privacy computing node after a third preset period of time has elapsed since the first timestamp; If the fifth state information indicates that the state of the node is abnormal, the first privacy computing node is deleted from the kubernetes cluster.

To sum up, using the management device for privacy computing nodes disclosed in the embodiments of this specification can manage privacy computing nodes under organizations in a more fine-grained manner, discover invalid nodes among them in a timely manner and add target taint marks, thereby triggering timely access to the services on it. The rescheduling of pods ensures the availability and stability of services. Furthermore, it is also possible to clean up and recycle invalid nodes for privacy computing, and remove the target taint mark for private computing nodes that have returned to a normal state and are in a stable state, so that they can be put into use again and give full play to the use efficiency of private computing nodes.

According to another embodiment, there is also provided a computer-readable storage medium on which a computer program is stored. When the computer program is executed in a computer, the computer is instructed to execute the method described in conjunction with FIG. 2 .

According to yet another embodiment, there is also provided a computing device, including a memory and a processor, wherein executable code is stored in the memory, and when the processor executes the executable code, the implementation described in conjunction with FIG. 2 is implemented. method.

Those skilled in the art should be aware that, in the above one or more examples, the functions described in the present invention may be implemented by hardware, software, firmware or any combination thereof. When implemented in software, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium.

The specific embodiments described above have further described the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above descriptions are only specific embodiments of the present invention and are not intended to limit the scope of the present invention. Protection scope, any modification, equivalent replacement, improvement, etc. made on the basis of the technical solution of the present invention shall be included in the protection scope of the present invention.

Claims (12)

1. A method of managing a private computing node, comprising:

acquiring first state information of a first private computing node in a kubernets cluster, wherein the first state information indicates that the state of the node is abnormal; the first privacy computing node belongs to a first namespace corresponding to a first organization;

adding a target taint mark to the first privacy calculation node under the condition that the duration of the abnormal node state is determined to reach a first preset duration based on the first state information; the first preset duration is set for the first mechanism;

in response to the addition of the target taint marker, scheduling a pod served on the first private computing node to a second private computing node in the kubernets cluster that is located under the first namespace; the second private computing node is in a normal state and does not have the target taint mark.

2. The method of claim 1, wherein the first state information further comprises a first timestamp;

wherein, in a case that it is determined based on the first state information that the duration of the node state anomaly reaches a first preset duration, adding a target taint mark to the first privacy computing node includes:

acquiring second state information of the first privacy computation node after the first preset time length from the first timestamp;

and under the condition that the second state information still indicates that the node state is abnormal, adding the target taint mark for the first privacy computing node.

3. The method of claim 1, wherein obtaining first state information for a first private computing node in a kubernets cluster comprises:

monitoring a first event of state change of the first privacy calculation node, wherein the information of the first event comprises the first state information;

under the condition that the duration of the abnormal node state reaches a first preset duration based on the first state information, adding a target taint mark to the first privacy calculation node, wherein the target taint mark comprises:

and adding the target taint mark to the first privacy computing node when the event that the state of the first privacy computing node is changed to be normal is not monitored in the first preset time period after the first event.

4. The method of claim 1, wherein after adding a target taint mark for the first private computing node, the method further comprises:

acquiring third state information of the first privacy calculation node, wherein the third state information indicates that the state of the node is normal;

and removing the target stain mark under the condition that the normal duration of the node state is determined to reach a second preset duration based on the third state information.

5. The method of claim 4, wherein the third state information further includes a second timestamp;

wherein, removing the target stain mark when it is determined that the duration of the node state normality reaches a second preset duration based on the third state information includes:

acquiring fourth state information of the first privacy calculation node after the second preset time from the second timestamp;

and under the condition that the fourth state information still indicates that the node state is normal, removing the target taint mark.

6. The method of claim 4, wherein obtaining third state information of the first private computing node comprises:

monitoring a second event of state change of the first privacy calculation node, wherein the information of the second event comprises the third state information;

wherein, removing the target stain mark when it is determined that the duration of the node state normality reaches the second preset duration based on the third state information includes:

and removing the target taint mark when the event that the state of the first privacy computation node is changed to be abnormal is not monitored within the second preset time period after the second event.

7. The method of claim 1, wherein the kubernets cluster includes a plurality of computing nodes; before obtaining first state information of a first private computing node in a kubernets cluster, the method further comprises:

filtering out a plurality of privacy compute nodes with privacy compute tags from the plurality of compute nodes; the plurality of privacy computing nodes belong to different namespaces corresponding to different organizations, and comprise the first privacy computing node and the second privacy computing node.

8. The method of claim 1, the method performed by a private computing node controller; wherein, after scheduling a pod served on the first private computing node to a second private computing node in the kubernets cluster that is located under the first namespace, the method further comprises:

deleting the first private computing node from the kubernets cluster in response to a call to a target API provided by the private computing node controller.

9. The method of claim 1, wherein the first state information further comprises a first timestamp; after scheduling a pod served on the first private computing node to a second private computing node in the kubernets cluster that is located under the first namespace, the method further comprises:

acquiring fifth state information of the first privacy computation node after a third preset time from the first timestamp;

deleting the first privacy computing node from the kubernets cluster if the fifth state information indicates that the node state is abnormal.

10. An apparatus for managing private computing nodes, comprising:

the state acquisition unit is configured to acquire first state information of a first private computing node in the kubernets cluster, and the first state information indicates that the state of the node is abnormal; the first privacy computation node belongs to a first namespace corresponding to a first mechanism;

a taint adding unit, configured to add a target taint mark to the first privacy computing node when it is determined that the duration of the node state anomaly reaches a first preset duration based on the first state information, and to schedule a pod served on the first privacy computing node to a second privacy computing node located under the first namespace in the kubernets cluster in response to the addition of the target taint mark; the first preset duration is set for the first mechanism; the second private computing node is in a normal state and does not have the target taint mark.

11. A computer-readable storage medium, on which a computer program is stored, wherein the computer program causes a computer to carry out the method of any one of claims 1-9, when the computer program is carried out in the computer.

12. A computing device comprising a memory and a processor, wherein the memory has stored therein executable code that when executed by the processor implements the method of any of claims 1-9.

Images