Intelligent scheduling device for equal-weight batch processing tasks and resource management method
By introducing stable sequence numbers and auxiliary sorting metrics into container scheduling, and constructing stable keys for equally weighted task sorting, the problems of scheduling inconsistency and queue jitter are solved, achieving stable task scheduling and interpretability, and reducing the risk of starvation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENZHEN XINGSHENG DIGITAL TECH CO LTD
- Filing Date
- 2026-04-13
- Publication Date
- 2026-05-15
AI Technical Summary
In container scheduling, the scheduling order of tasks with equal weights is unstable, leading to amplified task starvation and contention. Strategies are difficult to quantify, and existing implicit rules cause scheduling inconsistencies and queue jitter, making them difficult to reproduce and troubleshoot.
By introducing stable sequence numbers and auxiliary sorting indicators, a stable key is constructed to sort tasks with equal weights, generating a stable scheduling order. Scheduling audit information is recorded to support reproduction and troubleshooting. The stable sorting is achieved by using merge sort or Timsort algorithm.
It achieves a stable scheduling order for tasks with equal weights, reduces queue jitter and starvation risks, improves scheduling predictability and interpretability, and supports policy evaluation and troubleshooting.
Smart Images
Figure CN122044802A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of container cluster batch processing task scheduling and resource management technology, specifically to an intelligent scheduling device and resource management method for equally weighted batch processing tasks. Background Technology
[0002] In container scheduling mechanisms, a common practice is to configure weight parameters for tasks to be scheduled (such as Pods, Jobs, and Tasks) to express priority, queue quotas, or business importance. In each scheduling cycle, the scheduler sorts the set of tasks to be scheduled based on their weights and selects tasks to bind and distribute in sequence.
[0003] Typical application scenarios include, but are not limited to: Online services and offline tasks are co-located: The same cluster simultaneously hosts online services and batch processing tasks. Online services usually guarantee SLO through priority or weight, while batch processing tasks are queued in the remaining resources. When batch tasks have the same weight, the unstable order may cause some tasks to not get a chance for a long time or be frequently rolled back.
[0004] Multi-tenant queues and elastic quotas: In multi-tenant scenarios, scheduling devices often assign the same weight to tasks in the same queue to reflect fairness. When the number of tasks is large and resources are scarce, if the subset of tasks with equal weights lacks a stable order, unpredictable "queue jumping / rollback" phenomena will occur, causing fluctuations in the tenant's experience and making it difficult to audit.
[0005] Batch startup of pipeline / training jobs: In CI / CD, data processing or model training, a single submission may trigger a large number of equally weighted tasks to be queued concurrently; task order drift can lead to amplified contention with external dependencies (caching, data sharding, shared resources), resulting in queue jitter and decreased throughput.
[0006] In engineering practice, it is common for multiple tasks to have the same weight parameters, such as tasks within the same queue, batch requests from the same tenant, or tasks assigned the same weight under fair scheduling constraints. When there are many tasks with the same weight, existing devices often use implicit rules such as arbitrary order, traversal order, hash table iteration order, and concurrent queue dequeue order as the actual scheduling order for equal-weight scenarios. These implicit rules usually lack stability and predictability, leading to the following phenomena and shortcomings: Inconsistent scheduling order: Under the same input load and the same weight configuration, the order of tasks with equal weight may be different in different scheduling rounds or on different nodes, which is difficult to reproduce and troubleshoot.
[0007] Task starvation and contention amplification: In multiple scheduling cycles, the repeated rearrangement of the order of equal tasks can cause some tasks to remain at the back of the queue for a long time or be repeatedly "jumped in line", creating a starvation risk; at the same time, the unstable task order can cause resource contention and rollback jitter.
[0008] Difficulty in parameter tuning of strategies: Due to the instability of the actual order of tasks with equal weights, the effectiveness of weighted strategies and queue strategies is difficult to quantify and evaluate, affecting the operability of scheduling. Summary of the Invention
[0009] To address the aforementioned problems, this invention provides an intelligent scheduling resource management method for batch processing tasks with equal weights, applicable to an intelligent scheduling device for batch processing tasks in container clusters. The method includes the following steps: Step 1: Obtain the set of tasks currently in the waiting-to-be-scheduled state, extract the corresponding weight parameters for each task, and obtain the monotonically increasing stable sequence number assigned to each task at the moment of enqueueing. The stable sequence number remains fixed throughout the entire lifecycle of the task. Step 2: Perform preliminary sorting of the task set based on the weight parameters, identify the subset of tasks with the same weight parameters, schedule non-equal tasks according to the weight sorting results, and enter the stable sorting processing branch for the subset of equal tasks. Step 3: For the subset of tasks with equal rights, obtain at least one auxiliary ranking index, wherein the auxiliary ranking index includes at least the stable sequence number, which is used to characterize the relative order of tasks in the scheduling device; Step 4: Construct a stable key based on the auxiliary sorting index, and perform stable sorting on the subset of equal-weighted tasks using the stable key as the comparison key to generate a stable scheduling order; the stable sorting satisfies the following: when the stable keys of two tasks are equal, the relative order of the inputs of the two tasks in the output sequence remains unchanged; Step 5: Perform scheduling operations on the subset of equal-weight tasks according to the stable scheduling order; when a task scheduling fails and a rollback is triggered, do not reset the stable sequence number of the task, and keep the stable key of the task unchanged under the same policy version; at the same time, record the scheduling-related audit information to support the reproduction of the scheduling order and troubleshooting.
[0010] Preferably, the auxiliary sorting indicators also include one or more of the following: task queue identifier, enqueue timestamp, waiting time, last selected scheduling round, and position within the queue.
[0011] Preferably, the stable key is a composite sorting key, whose field set and field comparison order are fixed in advance, and the substitution rules for missing key fields are pre-defined to ensure that the stable keys generated by different scheduling instances under the same input state are consistent.
[0012] Preferably, the stable sequence number is allocated using a monotonically increasing counter within the queue. When multiple queues or multiple scheduling instances are concurrently enqueued, a globally unique sorting anchor point is formed by combining the queue identifier with the stable sequence number.
[0013] Preferably, the audit information includes one or more of the following: scheduling round, strategy version number, current processing weight value, size of the equalized task subset, summary of equalized sorting results, summary of filtering conditions, task ID that was attempted to be scheduled in this round, scheduling result, and reason for scheduling failure.
[0014] Preferably, the stable sorting is implemented using merge sort or Timsort algorithm.
[0015] Preferably, the stable sorting is achieved by maintaining a corresponding stable priority queue for each weight value. The elements in the stable priority queue are arranged in order of stability key, and the scheduling always starts from the element with the smallest stability key in the queue.
[0016] Preferably, when the scheduling strategy or weight configuration changes, the strategy version number is increased and the changed strategy version number is written into the audit chain; for the equal-weighted task subset under the same strategy version number, the output result of stable sorting can be reproduced.
[0017] Preferably, it also includes a quantitative evaluation step of the scheduling order stability, specifically: calculating the reordering rate of the equal task subset in two adjacent scheduling rounds. Equal rights hunger risk index Retry rate One or more of the following are used to evaluate scheduling stability based on the calculation results; Among them, rearrangement rate The calculation formula is: ; For the first An ordered list of equal-weighted task subsets in round-robin scheduling. For the first An ordered list of equal-weighted task subsets in round-robin scheduling; Equal rights hunger risk index The calculation formula is: ; For the task Waiting time For the subset of equal rights tasks The average waiting time; Retry rate The calculation formula is: ; This is the number of times a rollback and retries will occur. This represents the number of times the relative order of tasks has been advanced after a rollback.
[0018] The present invention also provides an intelligent scheduling device for equal-weighted batch processing tasks, used to implement the aforementioned intelligent scheduling resource management method for equal-weighted batch processing tasks.
[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Introduce a stable sorting key for task subsets with the same weight to ensure a consistent scheduling order under the same input conditions, which facilitates reproduction and troubleshooting.
[0020] 2. The equalization task is no longer repeatedly rearranged due to changes in the iteration order, reducing the chain reaction of queue jitter-back-reordering and lowering the probability of starvation.
[0021] 3. Stable order and versioned effective rules make scheduling results interpretable, can form audit logs, and support strategy evaluation.
[0022] 4. Without changing the weight semantics, a stable sorting rule is introduced only in branches with the same weight, which facilitates incremental access in the existing scheduler. Attached Figure Description
[0023] Figure 1 This is the overall flowchart of the present invention. Detailed Implementation
[0024] Example 1: Reference Figure 1 The intelligent scheduling resource management method for equal-weighted batch processing tasks provided by this invention is applied in an intelligent scheduling device for equal-weighted batch processing tasks. After sorting the task set according to weight parameters, a subset of tasks with the same weight parameters is identified; at least one auxiliary sorting index is obtained for the task subset, and a stable sorting is performed based on the auxiliary sorting index to generate a stable scheduling order; the scheduler performs scheduling operations on the tasks according to the stable scheduling order, thereby ensuring the consistency and predictability of the order under equal weight conditions.
[0025] For ease of understanding, some technical terms are defined as follows (the full English abbreviations appearing in the text are given): 1. Task: A scheduling object in a pending scheduling state, which can be a Pod, Job, Task or its equivalent entity.
[0026] 2. Weight Parameter: A numerical parameter used to express task priority or scheduling preference. The weights of the parameters are denoted as... , No. The weights of the parameters are denoted as... .
[0027] 3. Equal-Weight Subset: Satisfies A subset of the task set.
[0028] 4. Auxiliary Metric: A set of metrics used to distinguish the relative order of tasks when they have the same weight. It includes at least the stable sequence number enqueue_seq, as well as the entry time enqueue_ts, waiting time wait_time, last selected round last_selected_round, or queue position queue_pos.
[0029] 5. Stable Sort: When sorting an input sequence, if two elements have the same sort key, the relative order of the elements in the output remains unchanged.
[0030] 6. Stable key: A composite key used in the equalization task to generate a deterministic relative order. It must contain at least enqueue_seq and may optionally contain queue_id, policy_revision, etc.
[0031] The device of the present invention includes at least the following modules: 1. Task Collection Module: Used to collect a set of tasks in the pending scheduling state from the pending scheduling cache / queue, and to extract the task metadata in a standardized manner.
[0032] 2. Weighted Sorting and Equal-Weight Grouping Module: This module performs a preliminary sorting of the task set based on weight parameters and identifies subsets of tasks with equal weights. This provides clear boundaries for handling affirmative action branches.
[0033] 3. Stable sequence number allocation module: It is used to assign a monotonically increasing stable sequence number enqueue_seq to a task at the moment of task enqueueing, and ensure that the sequence number remains unchanged or maintains traceable monotonicity in backtracking / retrying / migration scenarios, thereby providing an anchor point for deterministic ordering.
[0034] 4. Auxiliary indicator extraction and stable key construction module: Used to extract from task objects and scheduling state storage: Auxiliary indicators such as queue_id, enqueue_ts, wait_time, last_selected_round, and policy_revision are used, and a stable key is constructed to make the relative order of the equitable task subset machine-readable and interpretable.
[0035] 5. Stable Sorting Module: Used to perform stable sorting on a subset of equal-weight tasks based on the stable_key and output a stable scheduling order. When there are concurrent threads or different running instances, the stable sorting module ensures that the output order is consistent under the same input state through the deterministic constraint of the stable key.
[0036] 6. Scheduling Execution and Rollback Module: Used for scheduling execution and rollback. The scheduling and binding are attempted sequentially; when a scheduling failure triggers a rollback, the stable_key is kept unchanged and the reason for failure is written, so that the failure retry does not change the source of the equalization order.
[0037] 7. Audit and Reproduction Module: Used to record policy_revision / schedule_round, summary of equitable task subsets and summary of sorting results, and supports reproducing the equitable order and failure reasons within a certain round or window based on the audit records.
[0038] This invention, through a combination of "weighted sorting + equal-weight grouping + auxiliary indicator extraction + stable sorting + audit reproducibility," enables the scheduler to obtain a deterministic order without altering the semantics of the weights. Its working process includes at least the following: 1. Task collection and standardization: The task collection module obtains a set of tasks from the scheduling cache and generates a normalized task_meta(task_id, queue_id, weight, enqueue_ts, enqueue_seq, policy_revision) for each task; where enqueue_seq is generated at the moment of enqueueing and remains unchanged, serving as an anchor point for stable sorting.
[0039] 2. Preliminary weighting and equal weighting identification: The weighted sorting module performs an initial sorting of tasks according to their weights, and the equal-weight grouping module identifies the results. For non-equal tasks, the scheduler processes them in weighted order; for subsets of equal tasks, it enters the stable sorting branch.
[0040] 3. Extraction of auxiliary indicators and construction of stable bonds: The auxiliary indicator extraction module extracts enqueue_seq and optional indicators, such as queue_id / enqueue_ts / wait_time / last_selected_round, from the subset of tasks with equal weighting, and constructs a stable key. When a task fails to be scheduled in multiple rounds and is rolled back, the stable key remains unchanged, thereby avoiding order drift caused by rollback.
[0041] 4. Stable sorting and sequential output: The stable sorting module performs stable sorting using stable_key as the comparison key and outputs a stable scheduling order. When different running instances or concurrent threads exist, the monotonicity and uniqueness of enqueue_seq ensure that the order can be reproduced under the same input conditions.
[0042] 5. Scheduling Execution and Auditing: The scheduling and execution module is based on The scheduling is attempted sequentially; the audit and reproduction module writes a summary of the sorting results to policy_revision / schedule_round, making the source of the scheduling order explainable and traceable.
[0043] To ensure reproducibility and consistency across instances, this invention must at least meet the following requirements: 1. Consistency constraint of stable sequence number: enqueue_seq is generated and remains unchanged at the moment the task is enqueued; when multiple queues or multiple instances are enqueued concurrently, a globally unique anchor point can be formed by "monotonic sequence number in queue + queue_id" to avoid order drift caused by concurrency.
[0044] 2. Deterministic constraints of stable keys: The set of fields of stable_key and the comparison order are fixed, and the rules for replacing values when key fields are missing are clearly defined, so that different instances under the same input state obtain the same stable_key.
[0045] 3. No queue interruption constraint during rollback: For tasks that roll back due to scheduling failure, enqueue_seq is not reset, and stable_key under the same policy_revision remains unchanged; when the policy changes, the effective boundary is clarified by increasing policy_revision, so that the "order change" has an explainable reason.
[0046] 4. Audit logs are reproducible: Each round of scheduling outputs a subset of tasks for equalization. The system can generate a summary, such as a hash digest of the task ID list, and record the set of filtering conditions and failure reasons for this round, so that the "input-sorting-failure-rollback" chain can be reviewed during offline troubleshooting.
[0047] To facilitate engineering evaluation of the "observable benefits of stable reordering," in addition to the rearrangement rate U, this invention can also define the following quantitative indicators: 1. Equal rights hunger risk index : ; in For the task Waiting time For the subset of equal rights tasks Average waiting time; when The larger the value, the more pronounced the long-tail waiting and hunger risk.
[0048] 2. Retry rate on failure : ; in This is the number of times a rollback and retries will occur. This represents the number of times the relative order is advanced after the rollback; this is true when the "rollback does not interrupt" constraint is satisfied. It can approach 0.
[0049] To ensure that auditing and reproduction have verifiable data standards, audit records should include at least the following set of fields (example): schedule_round (scheduling round), policy_revision (policy version), weight (currently processed weight value), group_size (size of the weighted task subset), order_digest (summary of weighted sorting results (e.g., task ID sequence hash)), filter_digest (summary of filtering conditions (e.g., resource snapshot / constraint set hash)), selected_task_id (task ID that was attempted to be scheduled in this round), and result (scheduling result (success / failure / rollback)).
[0050] The method of the present invention includes the following steps: 1. Obtain the set of tasks currently in the pending scheduling state, and obtain the corresponding weight parameters for each task.
[0051] 2. Based on the weight parameters, perform a preliminary sorting of the task set and identify a subset of tasks with equal weight parameters.
[0052] 3. For the subset of tasks with the same weight parameters, obtain at least one auxiliary ranking index, which is used to characterize the relative order of tasks in the scheduling device.
[0053] 4. Based on the auxiliary sorting index, perform a stable sorting on the task subset to generate a stable scheduling order.
[0054] 5. Perform scheduling operations on the task subset according to the stable scheduling order.
[0055] The key constraints and outputs of the above steps include at least: Step 2 determines the boundary of the equalization branch and outputs the subset of equalization tasks; Step 3 obtains at least a stable sequence number enqueue_seq and guarantees its monotonicity and determinism; Step 4 outputs... It can also generate a sorting result summary (order_digest); Step 5 records the scheduling result (result) and the failure reason (fail_reason) and associates them with policy_revision / schedule_round to write to the audit record to support reproduction and troubleshooting.
[0056] Among these, the auxiliary sorting metrics include at least a stable sequence number `enqueue_seq`, used to ensure a deterministic order in equal-weight scenarios. As an example, the scheduler assigns a monotonically increasing sequence number `enqueue_seq` to each task when it enters the scheduling queue, and defines the sorting key for the tasks using a stable key. : ; When multiple auxiliary metrics exist, the following composite key can be used to express stronger predictability and jitter resistance: ; For internal order semantics used to distinguish different queues / tenants; To provide a time sequence reference under extreme concurrency; To provide determinism when timestamp resolution is insufficient or concurrent enqueues occur simultaneously.
[0057] To prevent the order of equalization tasks from drifting due to changes in external state during multiple rounds of scheduling, this invention adds a recording mechanism for the policy version (policy_revision) and the scheduling round (schedule_round) to the stable key: when the policy or weight configuration changes, the policy_revision is increased and written into the audit chain; the sorting results of equalization tasks under the same policy_revision can be reproduced, thus ensuring that the scheduling order remains stable when the policy remains unchanged and has an interpretable effective boundary when the policy changes.
[0058] To facilitate the quantification of "order stability", this invention can define the reordering rate of equal task subsets in adjacent rounds of scheduling. : ; in Indicates the first An ordered list of equal-weighted task subsets in round-robin scheduling; Indicates the first An ordered list of equal-weighted task subsets in round-robin scheduling; when The closer the order is to 0, the more stable the sequence. By introducing a stable sorting rule, this invention can improve the stability of the order in equitable scenarios. A significant decrease, thereby reducing the risk of hunger and competition for tremors.
[0059] Stable sorting algorithm implementation (example): To enable a readily implementable method for "stable sorting," this invention may employ one or a combination of the following implementation paths: 1. Implementation based on a stable sorting algorithm: After constructing a stable_key for the subset of tasks with equal weight, a stable sorting algorithm is used to sort the list. The stable sorting algorithm can be merge sort, Timsort, or an equivalent implementation, satisfying the following condition: when the stable_keys of two tasks are equal, the relative order of their inputs is maintained in the output. The complexity of this implementation is O(n log n). It allows for direct reuse of general sorting libraries while ensuring stability.
[0060] 2. Implementation based on a stable priority queue: A stable queue structure is maintained for each weight value. Queue elements are ordered by stable_key, and during scheduling, the smallest stable_key is always popped. This implementation avoids sorting a subset of all equally weighted tasks in each round and maintains order consistency during high-concurrency enqueueing.
[0061] 3. Deterministic constraints on stable key allocation: enqueue_seq can be allocated by a monotonically increasing counter; when multiple queues are concurrently enqueued, (queue_id, enqueue_seq) can be used as a globally unique anchor point; when there are simultaneous enqueuings due to insufficient timestamp resolution, enqueue_seq can be used as the final decision key to eliminate uncertainty.
[0062] 4. Maintaining the order of rollback and retry: When a task fails to be scheduled due to insufficient resources or unmet constraints, the enqueue_seq is not reset during rollback, and the stable_key under the same policy_revision remains unchanged, thereby avoiding "failure retry causing queue jumping".
[0063] In summary, when a task enters the scheduling queue, the scheduler records its weight and assigns a monotonically increasing `enqueue_seq`. During each scheduling round, the scheduler obtains the set of tasks to be scheduled, initially groups them by weight, and identifies a subset of tasks with equal weight. The `enqueue_seq` (optionally `queue_id` / `enqueue_ts`) is extracted from the subset of tasks with equal weight, and a stable sort is performed using the composite key to generate an ordered list. The scheduler selects tasks sequentially according to the sorting results for binding / assignment. Failed tasks are rolled back according to fixed rules, but their stable keys remain unchanged. The `policy_revision` / `schedule_round` and a summary of the sorting results are written to the audit log for reproduction and troubleshooting.
[0064] Example 2: This example provides an alternative to Example 1: 1. enqueue_seq can be allocated by centralized storage or allocated locally within the queue and combined with queue_id to ensure global uniqueness, as long as it satisfies monotonicity and determinism.
[0065] 2. Stable sorting can be implemented by a stable sorting algorithm or by a stable priority queue (a heap / skip list with keys ordered), as long as the constraint of "maintaining the original relative order when keys are equal" is satisfied.
[0066] 3. In addition to the stable sequence number, the auxiliary indicator can also introduce the wait time (wait_time) as an optional key to achieve controllable aging improvement while maintaining stability. However, it still maintains a stable relative order when the weights are the same and the indicators are equal.
[0067] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0068] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A resource management method for intelligent scheduling of batch processing tasks with equal weights, applied to an intelligent scheduling device for batch processing tasks in container clusters, characterized in that, Includes the following steps: Step 1: Obtain the set of tasks currently in the waiting-to-be-scheduled state, extract the corresponding weight parameters for each task, and obtain the monotonically increasing stable sequence number assigned to each task at the moment of enqueueing. The stable sequence number remains fixed throughout the entire lifecycle of the task. Step 2: Perform preliminary sorting of the task set based on the weight parameters, identify the subset of tasks with the same weight parameters, schedule non-equal tasks according to the weight sorting results, and enter the stable sorting processing branch for the subset of equal tasks. Step 3: For the subset of tasks with equal rights, obtain at least one auxiliary ranking index, wherein the auxiliary ranking index includes at least the stable sequence number, which is used to characterize the relative order of tasks in the scheduling device; Step 4: Construct a stable key based on the auxiliary sorting index, and perform stable sorting on the subset of equal-weighted tasks using the stable key as the comparison key to generate a stable scheduling order; the stable sorting satisfies the following: when the stable keys of two tasks are equal, the relative order of the inputs of the two tasks in the output sequence remains unchanged; Step 5: Perform scheduling operations on the subset of equal-weight tasks according to the stable scheduling order; when a task scheduling fails and a rollback is triggered, do not reset the stable sequence number of the task, and keep the stable key of the task unchanged under the same policy version; at the same time, record the scheduling-related audit information to support the reproduction of the scheduling order and troubleshooting.
2. The intelligent scheduling and resource management method for equal-weighted batch processing tasks according to claim 1, characterized in that, The auxiliary sorting indicators also include one or more of the following: task queue identifier, enqueue timestamp, waiting time, last selected scheduling round, and position within the queue.
3. The intelligent scheduling and resource management method for equal-weighted batch processing tasks according to claim 1, characterized in that, The stable key is a composite sorting key, whose field set and field comparison order are fixed in advance, and the substitution rules for missing key fields are pre-defined to ensure that the stable keys generated by different scheduling instances under the same input state are consistent.
4. The intelligent scheduling and resource management method for equal-weighted batch processing tasks according to claim 1, characterized in that, The stable sequence number is allocated using a monotonically increasing counter within the queue. When multiple queues or multiple scheduling instances are concurrently enqueued, a globally unique sorting anchor point is formed by combining the queue identifier with the stable sequence number.
5. The intelligent scheduling and resource management method for equal-weighted batch processing tasks according to claim 1, characterized in that, The audit information includes one or more of the following: scheduling round, strategy version number, current weight value, size of the equalized task subset, summary of equalized sorting results, summary of filtering conditions, task ID that was attempted to be scheduled in this round, scheduling result, and reason for scheduling failure.
6. The intelligent scheduling and resource management method for equal-weighted batch processing tasks according to claim 1, characterized in that, The stable sorting is implemented using either merge sort or Timsort algorithm.
7. The intelligent scheduling and resource management method for equal-weighted batch processing tasks according to claim 1, characterized in that, The stable sorting is achieved by maintaining a corresponding stable priority queue for each weight value. The elements in the stable priority queue are arranged in order of stability key. During scheduling, the element with the smallest stability key in the queue is always popped first for scheduling.
8. The intelligent scheduling and resource management method for equal-weighted batch processing tasks according to claim 1, characterized in that, When the scheduling strategy or weight configuration changes, the strategy version number is increased and the changed strategy version number is written into the audit chain; for the equal weighted subset of tasks under the same strategy version number, the output results of stable sorting can be reproduced.
9. The intelligent scheduling and resource management method for equal-weighted batch processing tasks according to claim 1, characterized in that, It also includes a quantitative evaluation step for the stability of the scheduling order, specifically: calculating the reordering rate of equal-weighted task subsets in two adjacent scheduling rounds. Equal rights hunger risk index Retry rate One or more of the following are used to evaluate scheduling stability based on the calculation results; Among them, rearrangement rate The calculation formula is: ; For the first An ordered list of equal-weighted task subsets in round-robin scheduling. For the first An ordered list of equal-weighted task subsets in round-robin scheduling; Equal rights hunger risk index The calculation formula is: ; For the task Waiting time, For the subset of equal rights tasks The average waiting time; Retry rate The calculation formula is: ; This is the number of times a rollback and retries will occur. This represents the number of times the relative order of tasks has been advanced after a rollback.
10. An intelligent scheduling device for batch processing tasks with equal weights, characterized in that, This method is used to implement the intelligent scheduling and resource management method for equal-weighted batch processing tasks as described in any one of claims 1 to 9.