Large model container quick starting system and method based on object storage and medium

Abstract

The invention relates to a large model container quick starting system and method based on object storage and a medium, the system comprises a user plane component, a driving component, a persistent volume controller, a first control plane component and a second control plane component, the user plane component is used for monitoring a data volume resource creation event and a creation PVC request; the driving component is used for querying the size of a required data file in the object storage database; the persistent volume controller is used for initiating a PV resource creation request; the first control plane component is used for receiving a PV resource creation request and initiating a data volume creation request to the driving component based on the file list information; the driving component is used for loading required data files based on the file list and aggregating the required data files to form a file tree structure to obtain a read-only data volume; and the second control plane component is used for mounting the read-only data volume to the node of the large model container based on the PVC request, so that the rapid starting of the large model container is realized.

Description

Technical Field

[0001] This application relates to the field of computer technology, and particularly to a large model container fast startup system, method, and medium based on object storage. Background Art

[0002] When deploying and running large models in a Kubernetes environment, in addition to the need for computing resources such as CPUs (Central Processing Units) and GPUs (Graphics Processing Units), the primary task is to complete the loading of model data. For various current large models with high accuracy and the ability to handle complex tasks, the scale of their model data usually reaches the GB level, and the space required after complete loading often exceeds 100 GB. This poses high requirements for storage space and network bandwidth on the machine nodes within the Kubernetes cluster. If the model data needs to be downloaded again every time the model runs, a series of problems will occur, including slow startup of large model containers, a large amount of Kubernetes cluster network bandwidth being occupied, affecting the timely response of other containers to network requests, and the storage space of some cluster nodes being easily filled up, making it impossible to continue accepting container deployment and scheduling requests.

[0003] In view of the problem of slow startup of large model containers in the related art, no effective solution has been proposed yet. Summary of the Invention

[0004] Based on this, in view of the above technical problems, it is necessary to provide a large model container fast startup system, method, and medium based on object storage.

[0005] In a first aspect, an embodiment of this application provides a large model container fast startup system based on object storage, which is applied to a Kubernetes cluster. The system includes: a user plane component, a driver component, a persistent volume controller, a first control plane component, and a second control plane component.

[0006] The user plane component is configured to, when a data volume resource creation event is monitored, initiate a query request to the driver component according to the file list information of the required data files.

[0007] The driver component is configured to query in the object storage database based on the query request to obtain the size of the data files, where the object storage database stores large model training data.

[0008] The user plane component is configured to create a PVC request according to the file list information and size of the data files.

[0009] The persistent volume controller is configured to, when the creation of the PVC request is monitored, initiate a PV resource creation request.

[0010] The first control plane component is used to receive the PV resource creation request, extract the file list information, and initiate a data volume creation request to the drive component based on the file list information;

[0011] The drive component is used to load the required data files based on the file list and aggregate them into a file tree structure to obtain a read-only data volume;

[0012] The first control plane component is used to create a PV resource based on the PV resource creation request and bind the PV resource and the PVC request according to the PVC reference information in the request;

[0013] The second control plane component is used to mount the read-only data volume to the node of the large model container based on the PVC request, so that the large model container can load large model data based on the read-only data volume.

[0014] In one embodiment, the user plane component is further used to determine whether the data volume has been created when a data volume resource creation event is monitored. If so, the creation of the data volume is skipped.

[0015] In one embodiment, the user plane component is further used to initiate a query request to the drive component according to the file list of the required data files to determine whether there is a missing data file; if so, the creation of the data volume is terminated.

[0016] In one embodiment, the user plane component is further used to delete the PVC request when a data volume resource deletion event is monitored;

[0017] The persistent volume controller is further used to initiate a PV resource deletion request when the deletion of the PVC request is monitored;

[0018] The first control plane component is used to extract the file list information based on the PV resource deletion request; unbind the binding relationship between the PVC request and the PV resource, and initiate a data volume deletion request to the drive component based on the file list information;

[0019] The drive component is used to find and delete the specified data volume based on the file list information;

[0020] The first control plane component is further used to delete the PV resource after the data volume is deleted.

[0021] In one embodiment, the user plane component is further configured to, when a data volume resource deletion event is monitored, determine whether the data volume exists. If not, terminate the deletion of the data volume. If so, determine whether the data volume is used by the large model container. If so, terminate the deletion of the data volume.

[0022] In one embodiment, the second control plane component is further configured to, in response to a data volume unloading request of the large model container, extract the file list information, and initiate a data volume unloading request to the drive component based on the file list information.

[0023] The drive component is further configured to, based on the file list information, find a specified data volume for unloading.

[0024] In one embodiment, a custom CSI plugin is constructed based on the user plane component, the drive component, the first control plane component, and the second control plane component, and the custom CSI plugin is integrated into the Kubernetes cluster.

[0025] In a second aspect, an embodiment of the present application further provides a method for quickly starting a large model container based on object storage, which is applied to the system as described in the first aspect above. The method includes:

[0026] When the user plane component monitors a data volume resource creation event, it initiates a query request to the drive component according to the file list information of the required data files.

[0027] The drive component queries in the object storage database based on the query request to obtain the size of the data files.

[0028] The user plane component creates a PVC request according to the file list information and size of the data files.

[0029] When the persistent volume controller monitors the creation of the PVC request, it initiates a PV resource creation request.

[0030] The first control plane component receives the PV resource creation request, extracts the file list information, and initiates a data volume creation request to the drive component based on the file list information.

[0031] The drive component loads the required data files based on the file list and aggregates them into a file tree structure to obtain a read-only data volume.

[0032] The first control plane component creates a PV resource based on the PV resource creation request, and binds the PV resource and the PVC request according to the PVC reference information in the request.

[0033] Based on the PVC request, the second control plane component mounts the read-only data volume to the node of the large model container, so that the large model container loads large model data based on the read-only data volume.

[0034] In a third aspect, an embodiment of the present application further provides a computer device, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to run the computer program to execute the method described in the second aspect above.

[0035] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium. A computer program is stored in the storage medium, and when the computer program is executed by a processor, the method described in the second aspect above is implemented.

[0036] The above-mentioned large model container fast startup system, method and medium based on object storage. The system includes: a user plane component, a driver component, a persistent volume controller, a first control plane component, and a second control plane component. The user plane component is used to initiate a query request to the driver component according to the file list information of the required data files when a data volume resource creation event is monitored; the driver component is used to query in the object storage database based on the query request to obtain the size of the data files, where the object storage database stores large model training data; the user plane component is used to create a PVC request according to the file list information and size of the data files; the persistent volume controller is used to initiate a PV resource creation request when the creation of the PVC request is monitored; the first control plane component is used to receive the PV resource creation request, extract the file list information, and initiate a data volume creation request to the driver component based on the file list information; the driver component is used to load the required data files based on the file list and aggregate them into a file tree structure to obtain a read-only data volume; the first control plane component is used to create a PV resource based on the PV resource creation request and bind the PV resource and the PVC request according to the PVC reference information in the request; the second control plane component is used to mount the read-only data volume to the node of the large model container based on the PVC request, so that the large model container loads large model data based on the read-only data volume, solving the problem of slow startup of the large model container and realizing the fast startup of the large model container.

[0037] Details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more concise and understandable. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The accompanying drawings described herein are used to provide a further understanding of the present application, and constitute a part of the present application. The schematic embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation of the present application. In the drawings:

[0039] Figure 1 is a schematic structural diagram of a large model container quick startup system based on object storage in an embodiment;

[0040] Figure 2 is a schematic diagram of the large model data volume creation process of a large model container quick startup system based on object storage in an embodiment;

[0041] Figure 3 is a schematic diagram of the large model data volume mounting process of a large model container quick startup system based on object storage in an embodiment;

[0042] Figure 4 is a schematic diagram of the large model data volume unloading process of a large model container quick startup system based on object storage in an embodiment;

[0043] Figure 5 is a schematic diagram of the large model data volume deletion process of a large model container quick startup system based on object storage in an embodiment;

[0044] Figure 6 is a schematic diagram of the structure of a computer device in an embodiment. Detailed implementation manners

[0045] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be described and explained below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application. Based on the embodiments provided by the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application. Based on the embodiments provided by the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0046] Obviously, the accompanying drawings in the following description are only some examples or embodiments of the present application. For those of ordinary skill in the art, the present application can also be applied to other similar scenarios based on these drawings without creative efforts. In addition, it can also be understood that although the efforts made in this development process may be complex and time-consuming, for those of ordinary skill in the art related to the content disclosed in the present application, some design, manufacturing or production changes based on the technical content disclosed in the present application are only conventional technical means and should not be understood as the content disclosed in the present application being insufficient.

[0047] Reference to "embodiment" in this application means that the specific features, structures, or characteristics described in connection with the embodiment may be included in at least one embodiment of this application. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment each time, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0048] Unless otherwise defined, the technical terms or scientific terms involved in this application shall have the ordinary meaning as understood by those of ordinary skill in the technical field to which this application pertains. The words "a", "an", "one kind", "the" and other similar words involved in this application do not indicate a limitation in quantity and may represent a singular or plural number. The terms "include", "comprise", "have" and any variations thereof involved in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may further include steps or units not listed, or may further include other steps or units inherent to these processes, methods, products or devices. The words "connect", "be connected", "couple" and other similar words involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The term "plurality" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects and indicates that three relationships may exist. For example, "A and / or B" may represent: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally indicates an "or" relationship between the associated objects before and after. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific order for the objects.

[0049] The following are the explanations of some terms that appear in some embodiments of this application:

[0050] PV (Persistent Volume): A resource used to provide volumes for containers to mount in a Kubernetes cluster.

[0051] PVC (Persistent Volume Claims): A resource configuration that dynamically creates PVs and reclaims idle resources according to the storage resources and characteristics required by users in a Kubernetes cluster.

[0052] CSI (Container Storage Interface): A standardized plug-in interface used in Kubernetes clusters, mainly to solve the compatibility problem between storage plug-in developers and the Kubernetes platform, enabling various different types of storage solutions to be integrated into Kubernetes.

[0053] Kubelet Node Controller: An important component in the Kubernetes cluster. It runs on each node and is responsible for interacting with the node's operating system and performing container lifecycle management, including creation, startup, stop, and deletion.

[0054] Storage Class: An object in Kubernetes used to define storage volume provisioning policies. Kubernetes cluster administrators can configure and name different storage classes to facilitate the use of different storage classes to create storage volumes according to different requirements during container deployment.

[0055] The embodiments of the present application provide a large model container fast startup system based on object storage, which is applied to a Kubernetes cluster, as Figure 1 shown in the structural schematic diagram of the large model container fast startup system based on object storage. The system includes a user plane component, a driver component, a persistent volume controller, a first control plane component, and a second control plane component.

[0056] The user plane component is used to initiate a query request to the driver component according to the file list information of the required data files when a data volume resource creation event is monitored; the driver component is used to query in the object storage database based on the query request to obtain the size of the data files.

[0057] Among them, the object storage database stores large model training data; the object storage database is deployed inside the Kubernetes cluster, and the large model pre-training data is manually stored in the object storage database by the operation and maintenance personnel, and interacts with the object storage database through the driver component.

[0058] The user plane component is used to create a PVC request according to the file list information and size of the data file; the persistent volume controller is used to initiate a PV resource creation request when it monitors the creation of the PVC request; the first control plane component is used to receive the PV resource creation request, extract the file list information, and initiate a data volume creation request to the drive component based on the file list information; the drive component is used to load the required data files based on the file list and aggregate them into a file tree structure to obtain a read-only data volume; the first control plane component is used to create a PV resource based on the PV resource creation request and bind the PV resource and the PVC request according to the PVC reference information in the request; the second control plane component is used to mount the read-only data volume to the node of the large model container based on the PVC request, so that the large model container can load large model data based on the read-only data volume.

[0059] Among them, the main responsibility of the user plane component is to monitor the large model pre-training data volume resource events, and the event types are mainly divided into two categories: data volume creation events and data volume deletion events. One is the data volume creation event. The user declares the pre-training data files (including directory structures) required by the application in the large model pre-training data volume resource in a manual configuration manner. When these resources are created in the Kubernetes cluster, this event will be triggered; the other is the data volume deletion event. When the user manually finds and deletes the large model pre-training data volume resource with the corresponding name in the Kubernetes cluster, this event will be triggered. It is responsible for the life cycle management of the PVC and storage class of the large model data volume.

[0060] The main responsibility of the driver component is to operate the object storage database, interact with the object storage database, provide gRPC interfaces for other components to make remote calls, and run in the Kubernetes cluster in the form of a DaemonSet. Operating the object storage: Implement file search and aggregation to form a file tree structure (implement a custom read-only file system), and finally create a read-only volume that can be used externally; Provide remote call interfaces for the first control plane component: including CreateVolume (create volume), DeleteVolume (delete volume), ControllerPublishVolume (mount the volume to the node by the controller), ControllerUnpublishVolume (unmount the volume from the node by the controller), NodeStageVolume (prepare the volume by the node), and NodeUnstageVolume (cancel the volume preparation by the node), etc.; Provide remote call interfaces for the second control plane component: including NodePublishVolume (publish the volume to the container path by the node) and NodeUnpublishVolume (cancel the volume publication by the node), etc.; Provide remote call interfaces for the user plane component: Provide interfaces for file information retrieval, such as querying whether a file exists, obtaining the file size, volume reference information, etc.

[0061] The main responsibility of the first control plane component is to interface with the requests of the PersistentVolume Controller in the Kubernetes cluster, implement the business logic of underlying storage volume creation and deletion, PV creation and deletion, and PV and PVC binding and unbinding, and run in the Kubernetes cluster in the form of a DaemonSet.

[0062] The main responsibility of the second control plane component is to interface with the Kubelet node controller of the Kubernetes cluster nodes, and complete volume mounting before the large model container application runs on the allocated specific nodes and volume unmounting when it stops running.

[0063] The user plane component, the driver component, the first control plane component, and the second control plane component implement a custom CSI plugin according to the Kubernetes CSI (Container Storage Interface) specification to provide the ability of large model data volumes. The custom CSI plugin is integrated into the Kubernetes cluster.

[0064] The Persistent Volume Controller is a control plane component that is responsible for monitoring PVC request objects in the Kubernetes cluster, automatically matching appropriate PV resources to meet PVC requests, and managing the PV resource lifecycle.

[0065] The large model container fast startup system based on object storage provided by the embodiments of the present application combines with an object storage database, enabling, after the large model pre-training data is manually stored in the object storage by the operation and maintenance personnel, to quickly search, combine and form a read-only volume from the object storage database according to the large model pre-training data requirements, and finally mount it to the container of the large model application, avoiding the process of pulling the full amount of model data before the large model container starts, thus significantly accelerating the startup speed of the large model container.

[0066] In one embodiment, the user plane component is further configured to, when detecting a data volume resource creation event, determine whether the data volume has been created, and if so, skip the creation of the data volume.

[0067] In one embodiment, the user plane component is further configured to send a query request to the driver component according to the file list of the required data files, and determine whether there is a missing data file; if so, terminate the creation of the data volume.

[0068] In one embodiment, the user plane component is further configured to, when detecting a data volume resource deletion event, delete the PVC request; the persistent volume controller is further configured to, when detecting the deletion of the PVC request, initiate a PV resource deletion request; the first control plane component is configured to extract the file list information based on the PV resource deletion request; and unbind the binding relationship between the PVC request and the PV resource, and send a data volume deletion request to the driver component based on the file list information; the driver component is configured to search for and delete the specified data volume based on the file list information; the first control plane component is further configured to delete the PV resource after the data volume is deleted.

[0069] In one embodiment, the user plane component is further configured to, when detecting a data volume resource deletion event, determine whether the data volume exists, and if not, terminate the deletion of the data volume; if so, determine whether the data volume is used by the large model container, and if so, terminate the deletion of the data volume.

[0070] In one embodiment, the second control plane component is further configured to, in response to the data volume unloading request of the large model container, extract the file list information, and send a data volume unloading request to the driver component based on the file list information; the driver component is further configured to search for and unload the specified data volume based on the file list information.

[0071] In one of the embodiments, a custom CSI plug-in is constructed based on the user plane component, the driver component, the first control plane component, and the second control plane component, and the custom CSI plug-in is integrated into a Kubernetes cluster.

[0072] By using customized CSI plug-ins and object storage, the system can flexibly adapt to different storage requirements and is easy to expand, enhancing the flexibility and scalability of the system.

[0073] The present application also provides a method for quickly starting a large model container based on object storage, which is applied to Figure 1 The large model container rapid startup system based on object storage shown in the figure includes the following steps:

[0074] Step 201: When the user plane component monitors a data volume resource creation event, it initiates a query request to the driver component based on the file list information of the required data file.

[0075] Step 202: The driver component queries the object storage database based on the query request to obtain the size of the data file;

[0076] Step 203: The user plane component creates a PVC request based on the file list information and size of the data file.

[0077] Step 204: Upon receiving the PVC creation request, the persistent volume controller initiates a PV resource creation request.

[0078] Step 205: The first control plane component receives the PV resource creation request, extracts the file list information, and initiates a data volume creation request to the driver component based on the file list information.

[0079] Step 206: The driver component loads the required data files based on the file list and aggregates them into a file tree structure to obtain a read-only data volume.

[0080] Step 207: The first control plane component creates a PV resource based on the PV resource creation request, and binds the PV resource to the PVC request according to the PVC reference information in the request.

[0081] In step 208 , the second control plane component mounts the read-only data volume to the node of the large model container based on the PVC request, so that the large model container loads the large model data based on the read-only data volume.

[0082] In a specific embodiment, a method for creating a large model data volume is provided, which is applied to Figure 1The large model container fast startup system based on object storage as shown, the large model data volume creation process is as follows Figure 2 shown, including the following steps:

[0083] Step 301, the user plane component completes the establishment of a long connection through the API Server (Application Programming Interface Server) interface of the Kubernetes cluster, and continuously listens for events of large model data volume resources;

[0084] Step 302, when the user plane component receives an event to create a large model data volume, it checks whether the same volume has been created. If so, it skips the creation process and directly reuses the existing data volume; if not, it enters the complete data volume creation process;

[0085] Step 303, the user plane component obtains all the pre-trained model data files required for the large model data volume resource, and sends a file existence query request to the drive component. If there are missing parts of the files, the volume creation process will be terminated;

[0086] Step 304, after the user plane component completes the file retrieval, it sends a request to the drive component again to obtain the space size of all files;

[0087] Step 305, the user plane component creates a specific Storage Class (storage class) according to the file information list, and then creates a PVC according to the file space size information and this Storage Class (storage class);

[0088] Step 306, the persistent volume controller of the Kubernetes cluster listens for the event of PVC creation, and then initiates a PV creation request;

[0089] Step 307, the first control plane component receives the PV creation request, extracts the file list information parameters, and then entrusts the drive component to create the volume;

[0090] Step 308, the drive component will load all the required files, merge them into a file tree structure and finally provide a read-only type of volume, and finally complete this entrusted process;

[0091] Step 309, the first control plane component starts to create a PV, and according to the PVC reference information in the request, completes the relationship binding between the PV and the PVC;

[0092] Step 310, when the large model application needs to be containerized and deployed in the Kubernetes cluster, only the previously created PVC needs to be configured, and then during the actual container creation stage, the Kubelet node controller calls the second control plane component to mount the volume, so as to quickly complete the quick mounting of the large model pre-training data and shorten the container startup time.

[0093] In a specific embodiment, the application Figure 1 The large model container quick start system based on object storage shown in the figure has a large model data volume mounting process as shown in the figure. Figure 3 As shown, the following steps are included:

[0094] Step 401: reference the PVC of the data volume in the large model container and deploy the container;

[0095] Step 402: The Kubelet node controller initiates a volume mount request.

[0096] Step 403: The second control plane component receives the request parameters and remotely calls the driver component.

[0097] Step 404: The driver component finds the specified volume and returns the volume;

[0098] Step 405: The second control plane component mounts the volume to the path specified by the node;

[0099] In step 406, the Kubelet node controller updates the container's mount point and starts the container.

[0100] At this point, the large model container starts quickly to process the business.

[0101] In a specific embodiment, the application Figure 1 The large model container quick start system based on object storage shown in the figure, the large model data volume unloading process is as follows Figure 4 As shown, the following steps are included:

[0102] Step 501: The large model application container stops running;

[0103] Step 502: The Kubelet node controller detects that the container is stopped, cleans up the volume mount point of the container on the node, and initiates a volume unmount request.

[0104] Step 503: The second control plane component receives the request parameters and remotely calls the driver component.

[0105] Step 504: The driver component finds the specified volume and completes the volume unmounting.

[0106] In step 505, the Kubelet node controller completely terminates the container, and the life cycle of the large model application container ends.

[0107] In a specific embodiment, the application Figure 1 The large model container quick start system based on object storage is shown in FIG. Figure 5 As shown, the following steps are included:

[0108] Step 601, when the large model application container is terminated, the Kubelet node controller on the corresponding Kubernetes cluster node monitors this event, and the Kubelet node controller will call the second control plane component to remove the large model data volume on the container mount point;

[0109] Step 602, when there is no reference to the large model data volume in the Kubernetes cluster, it can be manually deleted by the user;

[0110] Step 603, when the user plane component monitors the deletion event of the large model data volume, it will first check whether the volume exists. If not, it will directly terminate the deletion process; then it will check whether the volume is still in use through the interface of the drive component. If so, it will directly terminate the deletion process;

[0111] Step 604, after the user plane component completes the check, it submits a deletion request for the PVC;

[0112] Step 605, the persistent volume controller of the Kubernetes cluster monitors the event of PVC deletion and then initiates a PV deletion request;

[0113] Step 606, the first control plane component receives the PV deletion request, extracts the request parameters, first unbinds the binding relationship between the PVC and the PV, and then entrusts the drive component to complete the deletion of the volume;

[0114] Step 607, the first control plane component officially enters the deletion submission of the PV;

[0115] Step 608, after the PV is deleted, the user plane component officially submits the deletion of the Storage Class (storage class) and the final cleaning and recycling process of the PVC.

[0116] In one embodiment, a computer device is provided. The computer device can be a terminal, and its internal structure diagram can be as Figure 6 shown. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner. The wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication), or other technologies. When the computer program is executed by the processor, it implements a fast startup system for large model containers based on object storage.

[0117] Those skilled in the art can understand that Figure 6 the structure shown in Figure 6 is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0118] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in any of the above-mentioned embodiments of the large model container fast startup system based on object storage are implemented.

[0119] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above-mentioned embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to a memory, storage, database or other medium used in the various embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0120] The technical features of the above-mentioned embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered that the scope described in this specification is covered.

[0121] The above-mentioned embodiments only represent several implementation manners of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of this application, several modifications and improvements can still be made, and these all belong to the protection scope of this application. Therefore, the protection scope of this application patent should be subject to the appended claims.

Claims

1. A fast startup system for large model containers based on object storage, characterized in that Applied to a Kubernetes cluster, the system includes: a user plane component, a driver component, a persistent volume controller, a first control plane component, and a second control plane component. The user plane component is used to initiate a query request to the driver component according to the file list information of the required data files when a data volume resource creation event is monitored. The driver component is used to query in the object storage database based on the query request to obtain the size of the data file, where the object storage database stores large model training data. The user plane component is used to create a PVC request according to the file list information and size of the data file. The persistent volume controller is used to initiate a PV resource creation request when the PVC request creation is monitored. The first control plane component is used to receive the PV resource creation request, extract the file list information, and initiate a data volume creation request to the driver component based on the file list information. The driver component is used to load the required data files based on the file list and aggregate them into a file tree structure to obtain a read-only data volume. The first control plane component is used to create a PV resource based on the PV resource creation request and bind the PV resource and the PVC request according to the PVC reference information in the request. The second control plane component is used to mount the read-only data volume to the node of the large model container based on the PVC request, so that the large model container can load large model data based on the read-only data volume.

2. The system according to claim 1, wherein the user plane component is further used to, when a data volume resource creation event is monitored, determine whether the data volume has been created. If so, skip the creation of the data volume.

3. The system according to claim 1, wherein the user plane component is further used to initiate a query request to the driver component according to the file list of the required data files, and determine whether the data file is missing. If so, terminate the creation of the data volume.

4. The system according to claim 1, wherein the user plane component is further used to delete the PVC request when a data volume resource deletion event is monitored. The persistent volume controller is further used to initiate a PV resource deletion request when the PVC request deletion is monitored. The first control plane component is used to extract the file list information based on the PV resource deletion request; unbind the binding relationship between the PVC request and the PV resource, and initiate a data volume deletion request to the driver component based on the file list information. The driver component is used to find and delete the specified data volume based on the file list information. The first control plane component is further used to delete the PV resource after the data volume is deleted.

5. The system according to claim 4, wherein The user plane component is further configured to, when a data volume resource deletion event is monitored, determine whether the data volume exists. If not, terminate the deletion of the data volume; if so, determine whether the data volume is used by the large model container. If so, terminate the deletion of the data volume.

6. The system according to claim 1, wherein The second control plane component is further configured to, in response to a data volume unloading request of the large model container, extract the file list information and initiate a data volume unloading request to the drive component based on the file list information; The drive component is further configured to, based on the file list information, find a specified data volume for unloading.

7. The system according to claim 1, characterized in that Based on the user plane component, the drive component, the first control plane component, and the second control plane component, a custom CSI plugin is constructed, and the custom CSI plugin is integrated into the Kubernetes cluster.

8. A method for quickly starting a large model container based on object storage, characterized in that, Applied to the system according to any one of claims 1 to 7, the method includes: When the user plane component monitors a data volume resource creation event, it initiates a query request to the drive component according to the file list information of the required data files; The drive component queries in the object storage database based on the query request to obtain the size of the data files; The user plane component creates a PVC request according to the file list information and size of the data files; When the persistent volume controller monitors the creation of the PVC request, it initiates a PV resource creation request; The first control plane component receives the PV resource creation request, extracts the file list information, and initiates a data volume creation request to the drive component based on the file list information; The drive component loads the required data files based on the file list and aggregates them into a file tree structure to obtain a read-only data volume; The first control plane component creates a PV resource based on the PV resource creation request and binds the PV resource and the PVC request according to the PVC reference information in the request; The second control plane component mounts the read-only data volume to the node of the large model container based on the PVC request, so that the large model container loads large model data based on the read-only data volume.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor implements the method according to claim 8 when executing the computer program.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program, when executed by the processor, implements the method according to claim 8.

Images