Does Google Use Python? Where Python Fits at Google
Yes, Google uses Python, and it is one of the company's official languages. Google publishes its own Python style guide. Inside the company, Python is used for internal tooling, build and release scripts, data pipelines, machine learning research code, and parts of YouTube. What Python is not used for is the code that answers a search query or serves an ad in a few milliseconds. That code is written in C++, Java, or Go.
This page covers where Python fits and where it does not. If you want the full list of languages Google uses across all of its products, that question is answered in Which coding language is used by Google?.
Where Python is used at Google
Internal tools. Google has thousands of small programs that engineers use every day. Scripts check code before it is submitted, tools query logs, and dashboards summarize test results. Python is the default for that class of program, because a tool written in an afternoon is worth more than a faster tool written in a week.
Build and release scripts. The steps that compile, package, test, and deploy a service are a long list of small commands. Python is the language used to run those commands in order and check their results. Speed does not matter here. The compiler and the tests take the time, not the script that starts them.
Data pipelines. A pipeline is a program that reads a large data set, transforms it, and writes the result somewhere else. Google's pipeline frameworks expose Python interfaces. An analyst can describe a pipeline in Python while the slow parts of the work run in C++.
Machine learning research. TensorFlow and JAX, two machine learning libraries created at Google, are written to be used from Python. The Python code describes the model. The library sends the arithmetic to a C++ engine and then to GPUs or TPUs. Researchers therefore get Python's convenience without paying Python's speed cost on the arithmetic.
YouTube. Large parts of YouTube's application code were written in Python from the start. The site is the most cited example of Python at Google scale. Over time the most demanding paths were moved to compiled languages, but Python is still a major part of the codebase.
Where Python is not used, and why
Python is not used on latency-critical serving paths. A serving path is the code that runs between a user's request arriving and the response leaving. For search, ads, and Gmail, that code must finish in a few milliseconds while handling thousands of requests per second per machine. Those paths are C++, Java, or Go.
The reason is the interpreter. Python runs by reading each instruction and executing it at the moment it is reached. C++, Java, and Go are compiled to machine code ahead of time or shortly after start. For the same loop, the compiled version is commonly 10 to 100 times faster.
The standard Python interpreter also has a global interpreter lock. This is a rule that only one thread may run Python code at any moment. A Python process therefore cannot spread CPU work across the cores of a machine the way a Go or Java process can. A serving machine with 64 cores would waste most of them.
The general principle is that Python is chosen where engineer time is the scarce resource. A compiled language is chosen where machine time is. Google's backend stack is described further in What backend does Google use?.
Where Go fits between the two
Go was created at Google to solve a specific problem. Engineers wanted something close to Python's simplicity for writing network services, without Python's speed limit. Go compiles to a single binary, starts in milliseconds, and has built-in support for running many tasks at the same time. That is why it is the common choice for infrastructure services such as Kubernetes, the container scheduler that Google released as open source.
In practice Go is often the middle choice. A team writes a tool in Python and finds it is too slow or too hard to run as a service. The team then rewrites it in Go rather than in C++, because Go is far easier to write correctly and is fast enough for almost everything that is not search or ads.
Can you interview at Google in Python?
Yes. Google lets you code in any mainstream language during the interview. Python is one of the most common choices, because it is short to write on a whiteboard or in a shared editor. Interviewers do not penalize you for choosing it.
What they do expect is fluent, idiomatic Python. That means using a list, a dictionary, or a set for the right reason and knowing the cost of each operation. Appending to a list takes constant time. Inserting at the front takes time proportional to the list length. A dictionary lookup takes about constant time. The guide to that choice is In Python, when to use a Dictionary, List or Set?.
They also expect you to know the language's own tools. Generators are functions that produce one value at a time using yield instead of building a whole list. Using one shows that you understand memory. Comprehensions, sorting with a key function, and the collections module save lines and signal experience.
Whether Python is the best choice for your own interview, compared with Java or C++, is weighed in Which language is best for Google coding interview?.
How to Prepare
- Learn the cost of every Python container. Write down the time of insert, delete, lookup, and iterate for list, dict, set, and deque. Check each one against a small experiment.
- Practice the Google patterns in Python. Grokking the Google Coding Interview works through the question types Google asks most often, with Python solutions.
- Write generators until they are natural. Rewrite three solutions that build a list so that they yield values instead. Explain the memory difference out loud.
- Explain your choice of language in one sentence. "I use Python because it lets me show the algorithm with the least syntax" is a complete answer.
GET YOUR FREE
Coding Questions Catalog
$98
$197
$72
