yandex-tank takes commits but has not tagged a release since 2020
Load and performance benchmark tool
At a glance
- What is it?
- A load testing tool that wraps four load generators written in four languages behind one configuration system, with results analysis pushed to a hosted service. Its newest tag is a Python 2 build from October 2020, while the package now requires Python 3.12 and caps numpy and pandas from below.
- Who is it for?
- This suits a team that already has load generators and wants one configuration file, one autostop rule and one results format across them. It does not suit anyone starting fresh, because the packaging is the problem rather than the idea.
- Can I use it commercially?
- Check first. The repository uses a licence we do not classify automatically, so read its LICENSE file before any commercial use.
- Is it still maintained?
- Yes. The repository last received commits 19 days ago.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.
Star history
- yandex/yandex-tank2,600 ★ · +1
| Repository | 2026-10-04 | 2026-11-15 |
|---|---|---|
| yandex/yandex-tank | 2.6k | 2.6k |
1 Stars in the last 42 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
The newest tag is a Python 2 build from October 2020
The README leads with a migration notice: the project has been moved to Python 3, and a link is offered to the latest stable release for Python 2. That link points at a tag named simply Python2, with no version number in it. The release list confirms how odd the naming became. There is Python2 from 2020-10-30, then v1.12.8 from 2020-01-21, then v1.12.3 from 2019-07-04. So the numbered scheme ran to 1.12.8 and the Python 2 line was tagged with a bare word after that. The Python 3 line has no tag in the list at all. The last commit to the branch is dated 2026-09-22, so work continues, but the newest published artifact anyone can install by version is almost six years old.
A Python 3.12 floor with ceilings on numpy and pandas
The package declares python_requires of 3.12 or newer, which is a recent interpreter floor for a tool whose README is still explaining a Python 2 to 3 move. Against that, the numeric stack is pinned from above: pandas at or below 2.0.3 and numpy at or below 1.26.4. Both are less-than-or-equal constraints on libraries that have moved several majors since. The combination leaves the resolver needing an old numpy and an old pandas that still ship wheels for a new Python, which is the opposite of the usual pattern where a new interpreter forces newer numeric libraries. Everything else in the requirement list uses a floor instead, including psutil, requests, PyYAML, cerberus for validation and influxdb.
retrying appears twice in the requirement list
The install requirements name the retrying library once with a floor of 1.3.4 and once more with no constraint at all, in the same list, a few entries apart. Two entries for one distribution means the effective constraint is decided by the resolver rather than by the author, and which of the two wins is not something the file states. It is the kind of duplicate that survives because a lower bound is harmless when the library is already present. The same list also carries pytest-runner as a runtime requirement, so a plugin for running tests is installed by anyone who installs the load tester. The test requirements are a separate list, and they pin zipp at exactly 0.5.1 while also asking for pytest 7.4.4 or newer.
The description names a generator the feature list does not
Four load generators are offered. Phantom is the default, written in C++ and described as very fast at 100,000 requests per second or more. JMeter is the extendable and widely known one, in Java. BFG is Python-based and lets you write load scenarios in Python. Pandora is a Golang generator marked experimental. The package description, however, says the tool uses other load generators such as JMeter, ab or phantom, and ab does not appear anywhere in the feature list. So the generator most people already have installed is named in the metadata and omitted from the documentation of what is supported. The value the description claims is a common configuration system over those generators plus analytics for their results, which is the real reason to use it rather than any one generator.
Results analysis is a hosted service, and monitoring goes over SSH
One of the main features is a performance analytics backend where test results are stored and analysed online, and the see also list points at that same service again. So the tool splits in two: generation runs on your machines, analysis runs on someone else's, and the two links to that host differ in scheme, one written as plain http and the other as https. The monitoring feature is described as customizable, extendable, and working over SSH, which means the tool opens SSH connections to the hosts it watches rather than polling an agent port. Two other capabilities are local and self-contained: several ammo formats including a plain URL list and an access log, and a test autostop plugin that ends a run once the results have become obvious, which is aimed at saving wall clock time on long tests.
Ten layers of packaging and CI tooling accumulated in one root
The root carries an Ant build file and a Debian packaging configuration side by side, plus a Travis configuration and a GitHub directory. Formatting is configured twice, once for flake8 and once for a YAPF style file. The changelog is generated from commit messages through its own configuration file rather than written by hand, and readthedocs has a configuration file. Tests run under pytest with a dedicated configuration file. There is no pyproject.toml: the build runs through setup.py and setup.cfg, and the version is imported from a module inside the package rather than read from a tag or a data file. So packaging metadata lives in three places and the version lives in Python source.
The licence is stated three ways and the metadata could not determine it
The repository's licence metadata could not be resolved, and a licence file sits at the root. The packaging script declares the licence as LGPLv2 and adds a classifier for the Lesser General Public License version 2 or later, which is a slightly wider grant than the bare string. Those two agree with each other. They do not establish what the root licence file contains, and the file itself is not reproduced anywhere in the documentation, so a reader cannot confirm that the text matches the declared identifier without opening it. Rather than pick one of the two statements, note that the declared string, the classifier and the file at the root are three separate places a licence can be recorded, and only two of them are visible here.
Editorial conclusion
This suits a team that already has load generators and wants one configuration file, one autostop rule and one results format across them. It does not suit anyone starting fresh, because the packaging is the problem rather than the idea. Three things to weigh. No release has been tagged since 2020 even though commits land regularly, so pinning a version means pinning a commit. The dependency set has a Python 3.12 floor alongside ceilings on numpy and pandas, which is a narrow window to resolve in. And the licence is stated three different ways in three different files, so confirm which one applies before you vendor it.
Frequently asked questions
Which load generators does yandex-tank support?
Four. Phantom in C++ as the default, described as very fast at 100,000 requests per second or more, JMeter, BFG for writing scenarios in Python, and an experimental Golang generator called pandora. The package description also names ab, which is not in the supported list.
Which Python versions does yandex-tank support now?
The packaging script requires 3.12 or newer. The README still carries a notice that the project moved to Python 3 and links to a Python 2 release, and that Python 2 tag is the newest one published.
Where are yandex-tank test results stored?
In a hosted performance analytics backend that the tool links as one of its main features, described as storing and analysing results online. A separate section links the same service as the related analytics server.
Does yandex-tank stop a test automatically?
There is a test autostop plugin that stops the run once the results have become obvious, which is intended to save time. Input is supplied through several ammo formats, including a plain URL list and an access log.
What licence is yandex-tank released under?
The packaging script declares LGPLv2 and carries a classifier for the Lesser General Public License version 2 or later, and a licence file sits at the repository root. The repository licence metadata could not be resolved, so the file itself is the place to confirm the terms.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/yandex-yandex-tank)Star history chart for your README
[](https://hysenlabs.com/projects/yandex-yandex-tank)Was this useful?