Build a build matrix and see the exact jobs it expands to, with include and exclude applied, before a run turns three dimensions into thirty jobs.
6 runs
2 axes, expanded.
Values are comma separated. Anything YAML would read as a number or a boolean gets quoted automatically — 3.10 unquoted is the float 3.1, which is the reason Python version matrices so often run the wrong version.
strategy:
fail-fast: false
matrix:
os: [ubuntu-latest, macos-latest]
node: [20, 22, 24]Written out so the count is checkable rather than asserted.
fail-fast defaults to true on GitHub, which cancels every other combination as soon as one fails. That is right when you only need to know whether anything broke, and wrong when you need to know which combinations broke — the usual reason to turn it off. This tool defaults it to off because a matrix is normally there to find out exactly that.
GitHub Actions Matrix Generator runs entirely on your device — nothing is uploaded.
You’re on the Free plan. Compare plans
A matrix multiplies, and the multiplication is invisible in the YAML. Three operating systems, four language versions and two database versions is twenty-four jobs, and nobody intends twenty-four jobs. Here the full expansion is listed, so the number is a decision rather than a surprise in the billing.
`include` and `exclude` are the parts that behave unintuitively, and both are applied here so you can see the result. `exclude` removes combinations. `include` has two distinct behaviours: an entry matching an existing combination adds variables to it, and one that does not match adds a whole new job. The same keyword doing two different things is why matrices end up with jobs nobody expected.
`fail-fast` is explained because the default is on, which cancels every other job the moment one fails. For a matrix whose purpose is finding out which platforms fail, that is exactly wrong — you learn about the first and nothing about the rest.
`max-parallel` is the other lever. A twenty-four job matrix with no limit can saturate your runner allocation and make every other workflow in the repository queue behind it.
Nothing is executed and no GitHub API is called. This expands the matrix locally and emits the YAML.
The product of every dimension's length, minus exclusions, plus any include entry that matches nothing. Three by four by two is twenty-four. The multiplication is the part that is invisible in the YAML, which is what this tool makes visible.
`exclude` removes combinations. `include` either adds variables to a combination it matches, or adds an entirely new job when it matches none. That second behaviour is where unexpected jobs come from.
Usually yes for a compatibility matrix. The default cancels every remaining job when one fails, so you learn that one platform broke and nothing about the others. For a deployment pipeline, failing fast is the right default.
Caps how many matrix jobs run at once. Without it a large matrix can consume your whole runner allocation and queue every other workflow in the repository behind it.
In some positions, with limits — a matrix can be generated from a job output as JSON, but not every expression context is available where you might want it. The expansion here works on literal values, which is what most matrices use.