Interview experience: WeChat, NetEase Games, Jinshan Xishanju, Renren

[[144779]]

WeChat Interview Experience

Backend development engineer

one side

At the beginning, I was given 6 algorithm questions to do. It took about 40 minutes to finish them. They were all basic algorithms, such as linked lists and binary trees. There was no better solution for the maximum common substring of two strings. I was asked about key/value cache and how it was implemented. Where is the bottleneck and how to ensure the load. Later, I was asked about the low-coupling architecture of load balancing in the background of Tencent Interactive Entertainment. I didn’t explain it very well here. I was asked about disk efficiency optimization, how to optimize it, how to measure disk data, what commands (top, iostat), how to view iostat parameters, read and write queues, the ratio of waiting time to service time, and the idle ratio. Coding standards (naming standards, fixed bracket style, function comments and key comments, exception descriptions, class access control standards). I was asked about what should be paid attention to in network programming (TCP and UDP trade-offs), and how UDP achieves reliable transmission. Have you done any tests and the appropriateness of using scp as a comparison. I was asked about the commands used for network debugging, such as tcpdump, lsof, netstat. I was asked about the compiler, how it was implemented, and whether the solution to the Chinese encoding problem was considered. The implementation process of the compiler. Asked how to check the running status of a process, top, /proc/pid/, asked how to count the usage frequency of a function, using valgrind. ==. A stressful interview. .

Second side

The second interview was with the director. I introduced myself, then asked about the best project I had done, talked about the compiler, and finally asked me to introduce a project. I talked about the UDP file upload service. I was asked about the specific content of UDP reliable transmission, and was curious why the speed could be twice that of SCP. I was asked why I didn't write anything in undergraduate studies, and I didn't answer well. I was asked what technical books I read, and I talked about the detailed explanation of the Linux kernel. I talked about memory and introduced it. Finally, I was asked if memory was used in the project. In fact, there was no use, but I could talk about it better, such as my understanding of dynamic memory and how much the actual overhead of thread creation is in terms of memory. I talked about mmap in files and direct I/O. I was asked about three advantages and the biggest disadvantage. .

Three sides

The third interview was with the HR, and I introduced myself. In the future, I should make my self-introduction more substantial. They asked me how many offers I had received, my future work intentions, whether I was willing to stay in Guangzhou, and also asked me which project I thought was the best and why it was the best. Well, I need to think more about this.

Successfully got the offer~

NetEase Games Interview Experience

Game development engineer

written examination

I have to mention this. The total number of questions is three hours, and the questions are not easy. The first 30 basic questions are one point each, and they are not easy to do. The last 6 algorithm questions involve flexible array traversal, non-recursive algorithm implementation, BNF, linked lists, graphs, Bloom filters, etc. Most of them are written in a short question, which is difficult~ (actually, thinking back, it is because I am not good at applying knowledge flexibly)

one side

I started with a matrix multiplication test. After repeated checks, I found that it was just matrix multiplication. I might have overthought it. First, I introduced myself. I introduced C++, the difference between C++ and C, and gave an example of C++ polymorphism. The test was relatively lively, but I successfully completed it. I introduced the TCP protocol. I asked about epoll, the characteristics of epoll, and the difference between LT and ET. I was asked about the memory layout of the process. Since I was well prepared, I was able to answer all of them. I was asked about the BFS algorithm and its extension, and the segment tree problem. The last one was an algorithm for finding the point closest to a given point in the fastest time. I talked about the idea of ​​preprocessing with rounding, but there would be problems with data storage. If the problem domain is infinite, how to design such a data structure to store it? I was asked about the implementation of the compiler.

Second side

At the beginning, we had to count 24 points with four playing cards. Why couldn’t I figure it out at that time? = =~, and then we gave an algorithm to solve it. The algorithm for evenly placing points on the table, and finally the question of how to randomly place a point in a sector. The idea of ​​randomizing by arc and radius was mentioned, but this would concentrate the points at the top of the sector. It was mentioned that we could divide them by the area of ​​the top and bottom, but this would not be able to handle continuous situations and the same problem would occur. Finally, the idea of ​​continuous random density function was mentioned, but the final correct answer was not given. Give a random shuffling algorithm, how to prove that the shuffling is random. I also asked about the compiler and asked me to introduce it again. Finally, I asked why there was no HR interview, and he said that he could arrange one for me if necessary. . I consciously got out~

Successfully received the offer

Jinshan Xishanju Noodles

Backend development engineer

one side

The interviewer is from Hunan. He asked various C++ questions at the beginning, such as the differences between new, delete, malloc, and free, and the undefined behavior of the compiler when calling new a[5] and then delete a, and the underlying reasons. Then he asked about memory allocation, saying that there was 1G of memory in the heap, but the memory was enough, but why the memory allocation error occurred. He asked about the construction order of C++ classes. He asked about the TCP packet sticking phenomenon. One is the Nagle algorithm, and the other is the problem of unbounded packets of application layer data, which cannot be solved because TCP cannot control the application layer. It is only responsible for transmitting in byte streams. He asked about the intelligence of trains transporting coal and pulling lights... I just joked about it...

Second side

He was a director who was working on a 3D mobile game. He asked me about the compiler at the beginning. I said I didn’t understand the principles of compilation. He then asked me how I did it. I talked for a while, and he finally said it was a good idea to practice first and then look at the theory. He said he liked people who were devoted to research = =~… Then he asked about the issues that needed to be paid attention to in multithreading and the purpose of FTP.

Three sides

When I met CTO Chen Feizhou, I didn’t recognize him. . . First, he asked me a question about the compiler, how to distinguish between function declaration and function definition, and how the compiler can tell the difference. Then he asked me about the homework I did for my internship at Tencent, and talked about the concurrency strategy (network interruption for multiple CPU cores requires special network cards, special platforms and higher versions of Linux support). He asked me if I had any code accumulation, and talked about his habit of accumulating code. He asked me what kind of work I would like to do in the future, and told me about how he accumulated code between different product departments and completed it efficiently and optimally. He talked about choices, and talked about how he identified with the company’s values. He talked about the game industry, and why Unreal Engine III is not affordable (the overall industrial environment for art is not good, and the art resources for the same rendering are 4 times that of Blizzard art). He talked about Yunfeng, Lei Jun and Xiaomi, and looked at the Xiaomi 3S engineering machine on the table = =~.

Got the Kingsoft offer~ This offer is really hard to give up. The interviewers in several rounds all left a good impression. The whole Xishanju environment is also very good, with a self-service canteen restaurant and various game consoles next to the workroom. It should be very comfortable to live in Zhuhai with a good salary...

After I declined the offer, the interviewer gave me some constructive advice on how to choose between WeChat and NetEase Games, for which I was very grateful.

Everyone's Face

Backend development engineer

one side

At the beginning, I wrote a de-duplication algorithm to convert aabbbccddff into abcdef. I thought it was complicated. In fact, I just need to move the latter to the front (am I too weak...). Then I asked about Tencent internship homework, disk efficiency, concurrency trade-offs, and the effect of UDP reliable transmission. Finally, I was asked about the algorithm to find the largest element in the stack in 0(1) time. I didn't think of this, but considering the characteristics of stack pop and stack, it is easy to solve.

Second side

The first thing I learned was the algorithm. Draw an irregular polygon. Now each side needs a piece of wood of equal length. Given a long enough piece of wood, you need to use the minimum physical strength to draw these side length wood (physical strength is equal to the length of the wood). In fact, this is the Huffman algorithm. It took me many times to figure this out. Fortunately, the algorithm was implemented quickly in the end. Then I asked about Linux to check the memory, check the remaining disk capacity, and how to output the directory through ls.

Three sides

I asked about key/value storage, what are the advantages and disadvantages, and in what scenarios it is used. I asked how to ensure reliable mapping in a distributed key/value storage, and the answer is that a central agent is responsible for mapping, so what are the disadvantages. Then I asked, if different routines need to update a larger value to a certain key, how to ensure that the final value is the largest without using semaphores, synchronization locks, etc. (synchronization, set get loops are clever). Later, I asked about finding the kth largest number in massive data (1. In fact, you can directly use bucket division, even if it is 64 bits, at most three traversals are enough. 2. Bit filtering can also be used, but there are too many disk operations. 3. k heap filtering, but if k is too large, memory will be frequently swapped. 4. Distribute to multiple machines, multiple machines simultaneously sort the external memory, and have another machine for multi-way merging)

Successfully got the offer~