Recently Amazon 再次开启大规模 Online Assessment 发放，本轮 OA 统一使用 HackerRank 平台。不少同学在做完之后的共同感受是：题目本身并不偏难，但一旦理解偏差，后续代码几乎无法挽回。今天来对 Amazon OA 的两道 Coding 题进行系统性复盘，帮助后续参加 OA 的同学提前建立正确预期。
Amazon OA basic information overview
- Test platform: HackerRank
- Number of questions: 2 Coding
- Overall difficulty: above average
- Main inspection capabilities:
- Fast abstraction of complex business rules
- Ability to use dynamic data structures
- Is the greedy strategy reasonable?
- Sensitivity to hidden constraints and boundary conditions
Question 1: Server allocation and cost calculation issues
Question background
There are several servers in the system, and each server has a certain number of idle instances. The server status is given in the form of an array, the array subscript represents the server number, and the value represents the current number of available instances.
There is now M Customers arrive one after another, and each customer needs to choose a server to rent an instance. The status of the server changes continuously as the client makes choices.
Allocation and Cost Rules
For each customer, the system will perform the following operations:
- Among all current servers, select the server with the largest number of idle instances
- After a successful selection, the number of idle instances of the server is reduced by 1
- This selection will generate a cost, which is calculated as:
- Cost = the minimum number of idle instances in all current servers before selection + the maximum number of idle instances
最终要求输出：在 M The cumulative sum of all costs after each customer has been allocated.
Analysis of the essence of problem solving
This is a very typical Amazon-style question. The essence is not to test complex algorithms, but to test whether the candidate can:
- Quickly identify that this is a problem of dynamically maintaining maximum and minimum values
- In multiple update scenarios, avoid using inefficient full array scans
During ProgramHelp’s coaching process, the main reasons for losing points on this question focus on three aspects:
- Didn't realize that cost must be calculated "before selection"
- Each round traverses the array to find the maximum/minimum value, resulting in excessive time complexity.
- Not properly handling the server status after the number of instances is reduced to 0
Correct solution direction
The standard solution to this type of problem is to use a heap structure to maintain server state:
- Use the maximum heap to dynamically maintain the server with the most idle instances currently
- At the same time maintain the current minimum idle instance value (can be through the minimum heap or additional counting structure)
Each round of operations only requires:
- Take the maximum value
- Record the current maximum and minimum
- Update server status and re-enter the heap
As long as the data structure is chosen correctly, the overall implementation is stable and controllable.
Question 2: Log Parts and Warehouses Allocation Maximization Problem
Question background
系统中存在多个 log delivery，每个 log 对应一定数量的零件（parts）。现在有 K Warehouses, among which K Is an even number.
The storage rules are as follows:
- Each warehouse can only store data from Same log Parts
- Parts of the same log can be distributed and stored in multiple warehouses
- Parts that are allowed to have partial logs are ultimately not stored in any warehouse.
Sorting and target constraints
When all warehouses have completed storage:
- Sort by the number of parts stored in each warehouse
- Top ranking
K/2The warehouse is considered the "most stored half" - After ranking
K/2The warehouse is considered the "least half of storage"
The question requires output:
The maximum possible value of the sum of the number of parts in the second half (k/2 warehouses with the least storage).
Core difficulties in problem solving
The difficulty of this question is not in the implementation, but in the understanding of the goal.
Many students will subconsciously want to "store as many parts as possible", but the real optimization goal is not the total storage amount, but:
After sorting, try to raise the sum of the second half of the warehouse as much as possible.
这意味着：不能让前半部分 warehouse 吃掉过多零件，分配策略必须尽量均衡，同时满足单 warehouse 只能来自同一 log 的限制。
为什么很多考生在应对 Amazon OA 会选择专业助攻？
在 ProgramHelp 的真实案例中，很多同学的问题并不是“完全不会写”，而是：OA 时间不足，来不及试错，题目规则复杂，容易在细节处踩坑，平台限制严格，一次失误直接淘汰。
Amazon OA 的本质是筛人而不是教学。在这种情况下，选择稳定、经验成熟的面试支持，往往比单纯“再刷几道题”更有效。
