|
| 1 | +# Azure Pipelines Matrix Generator |
| 2 | + |
| 3 | +* [Usage in a pipeline](#usage-in-a-pipeline) |
| 4 | +* [Matrix config file syntax](#matrix-config-file-syntax) |
| 5 | + * [Fields](#fields) |
| 6 | + * [matrix](#matrix) |
| 7 | + * [include](#include) |
| 8 | + * [exclude](#exclude) |
| 9 | + * [displayNames](#displaynames) |
| 10 | +* [Matrix Generation behavior](#matrix-generation-behavior) |
| 11 | + * [all](#all) |
| 12 | + * [sparse](#sparse) |
| 13 | + * [include/exclude](#includeexclude) |
| 14 | + * [displayNames](#displaynames-1) |
| 15 | + * [Filters](#filters) |
| 16 | + * [Under the hood](#under-the-hood) |
| 17 | +* [Testing](#testing) |
| 18 | + |
| 19 | + |
| 20 | +This directory contains scripts supporting dynamic, cross-product matrix generation for azure pipeline jobs. |
| 21 | +It aims to replicate the [cross-product matrix functionality in github actions](https://docs.github.com/free-pro-team@latest/actions/reference/workflow-syntax-for-github-actions#example-running-with-more-than-one-version-of-nodejs), |
| 22 | +but also adds some additional features like sparse matrix generation, cross-product includes and excludes, and programmable matrix filters. |
| 23 | + |
| 24 | +This functionality is made possible by the ability for the azure pipelines yaml to take a [dynamic variable as an input |
| 25 | +for a job matrix definition](https://docs.microsoft.com/azure/devops/pipelines/process/phases?view=azure-devops&tabs=yaml#multi-job-configuration) (see the code sample at the bottom of the linked section). |
| 26 | + |
| 27 | +## Usage in a pipeline |
| 28 | + |
| 29 | +In order to use these scripts in a pipeline, you must provide a config file and call the matrix creation script within a powershell job. |
| 30 | + |
| 31 | +For a single matrix, you can include the `eng/pipelines/templates/jobs/job-matrix.yml` template in a pipeline: |
| 32 | + |
| 33 | +``` |
| 34 | +jobs: |
| 35 | +- template: /eng/pipelines/templates/jobs/job-matrix.yml |
| 36 | + parameters: |
| 37 | + MatrixConfigs: |
| 38 | + - Name: base_product_matrix |
| 39 | + Path: /eng/pipelines/matrix.json |
| 40 | + Selection: sparse |
| 41 | + GenerateVMJobs: true |
| 42 | + - Name: sdk_specific_matrix |
| 43 | + Path: /sdk/foobar/matrix.json |
| 44 | + Selection: all |
| 45 | + GenerateContainerJobs: true |
| 46 | + steps: |
| 47 | + - pwsh: |
| 48 | + ... |
| 49 | +``` |
| 50 | + |
| 51 | +## Matrix config file syntax |
| 52 | + |
| 53 | +Matrix parameters can either be a list of strings, or a set of grouped strings (represented as a hash). The latter parameter |
| 54 | +type is useful for when 2 or more parameters need to be grouped together, but without generating more than one matrix permutation. |
| 55 | + |
| 56 | +``` |
| 57 | +"matrix": { |
| 58 | + "<parameter1 name>": [ <values...> ], |
| 59 | + "<parameter2 name>": [ <values...> ], |
| 60 | + "<parameter set>": { |
| 61 | + "<parameter set 1 name>": { |
| 62 | + "<parameter set 1 value 1": "value", |
| 63 | + "<parameter set 1 value 2": "<value>", |
| 64 | + }, |
| 65 | + "<parameter set 2 name>": { |
| 66 | + "<parameter set 2 value 1": "value", |
| 67 | + "<parameter set 2 value 2": "<value>", |
| 68 | + } |
| 69 | + } |
| 70 | +} |
| 71 | +"include": [ <matrix>, <matrix>, ... ], |
| 72 | +"exclude": [ <matrix>, <matrix>, ... ], |
| 73 | +"displayNames": { <parameter value>: <human readable override> } |
| 74 | +``` |
| 75 | + |
| 76 | +See `samples/matrix.json` for a full sample. |
| 77 | + |
| 78 | +### Fields |
| 79 | + |
| 80 | +#### matrix |
| 81 | + |
| 82 | +The `matrix` field defines the base cross-product matrix. The generated matrix can be full or sparse. |
| 83 | + |
| 84 | +Example: |
| 85 | +``` |
| 86 | +"matrix": { |
| 87 | + "operatingSystem": [ |
| 88 | + "windows-2019", |
| 89 | + "ubuntu-18.04", |
| 90 | + "macOS-10.15" |
| 91 | + ], |
| 92 | + "framework": [ |
| 93 | + "net461", |
| 94 | + "netcoreapp2.1", |
| 95 | + "net50" |
| 96 | + ], |
| 97 | + "additionalTestArguments": [ |
| 98 | + "", |
| 99 | + "/p:UseProjectReferenceToAzureClients=true", |
| 100 | + ] |
| 101 | +} |
| 102 | +``` |
| 103 | + |
| 104 | +#### include |
| 105 | + |
| 106 | +The `include` field defines any number of matrices to be appended to the base matrix after processing exclusions. |
| 107 | + |
| 108 | +#### exclude |
| 109 | + |
| 110 | +The `include` field defines any number of matrices to be removed from the base matrix. Exclude parameters can be a partial |
| 111 | +set, meaning as long as all exclude parameters match against a matrix entry (even if the matrix entry has additional parameters), |
| 112 | +then it will be excluded from the matrix. For example, the below entry will match the exclusion and be removed: |
| 113 | + |
| 114 | +``` |
| 115 | +matrix entry: |
| 116 | +{ |
| 117 | + "a": 1, |
| 118 | + "b": 2, |
| 119 | + "c": 3, |
| 120 | +} |
| 121 | +
|
| 122 | +"exclude": [ |
| 123 | + { |
| 124 | + "a": 1, |
| 125 | + "b": 2 |
| 126 | + } |
| 127 | +] |
| 128 | +``` |
| 129 | + |
| 130 | +#### displayNames |
| 131 | + |
| 132 | +Specify any overrides for the azure pipelines definition and UI that determines the matrix job name. If some parameter |
| 133 | +values are too long or unreadable for this purpose (e.g. a command line argument), then you can replace them with a more |
| 134 | +readable value here. For example: |
| 135 | + |
| 136 | +``` |
| 137 | +"displayNames": { |
| 138 | + "/p:UseProjectReferenceToAzureClients=true": "UseProjectRef" |
| 139 | +}, |
| 140 | +"matrix": { |
| 141 | + "additionalTestArguments": [ |
| 142 | + "/p:UseProjectReferenceToAzureClients=true" |
| 143 | + ] |
| 144 | +} |
| 145 | +``` |
| 146 | + |
| 147 | +## Matrix Generation behavior |
| 148 | + |
| 149 | +#### all |
| 150 | + |
| 151 | +`all` will output the full matrix, i.e. every possible permutation of all parameters given (p1.Length * p2.Length * ...). |
| 152 | + |
| 153 | +#### sparse |
| 154 | + |
| 155 | +`sparse` outputs the minimum number of parameter combinations while ensuring that all parameter values are present in at least one matrix job. |
| 156 | +Effectively this means the total length of a sparse matrix will be equal to the largest matrix dimension, i.e. `max(p1.Length, p2.Length, ...)`. |
| 157 | + |
| 158 | +To build a sparse matrix, a full matrix is generated, and then walked diagonally N times where N is the largest matrix dimension. |
| 159 | +This pattern works for any N-dimensional matrix, via an incrementing index (n, n, n, ...), (n+1, n+1, n+1, ...), etc. |
| 160 | +Index lookups against matrix dimensions are calculated modulus the dimension size, so a two-dimensional matrix of 4x2 might be walked like this: |
| 161 | + |
| 162 | +``` |
| 163 | +index: 0, 0: |
| 164 | +o . . . |
| 165 | +. . . . |
| 166 | +
|
| 167 | +index: 1, 1: |
| 168 | +. . . . |
| 169 | +. o . . |
| 170 | +
|
| 171 | +index: 2, 2 (modded to 2, 0): |
| 172 | +. . o . |
| 173 | +. . . . |
| 174 | +
|
| 175 | +index: 3, 3 (modded to 3, 1): |
| 176 | +. . . . |
| 177 | +. . . o |
| 178 | +``` |
| 179 | + |
| 180 | +#### include/exclude |
| 181 | + |
| 182 | +Include and exclude support additions and subtractions off the base matrix. Both include and exclude take an array of matrix values. |
| 183 | +Typically these values will be a single entry, but they also support the cross-product matrix definition syntax of the base matrix. |
| 184 | + |
| 185 | +Include and exclude are parsed fully. So if a sparse matrix is called for, a sparse version of the base matrix will be generated, but |
| 186 | +the full matrix of both include and exclude will be processed. |
| 187 | + |
| 188 | +Excludes are processed first, so includes can be used to add back any specific jobs to the matrix. |
| 189 | + |
| 190 | +#### displayNames |
| 191 | + |
| 192 | +In the matrix job output that azure pipelines consumes, the format is a dictionary of dictionaries. For example: |
| 193 | + |
| 194 | +``` |
| 195 | +{ |
| 196 | + "net461_macOS1015": { |
| 197 | + "framework": "net461", |
| 198 | + "operatingSystem": "macOS-10.15" |
| 199 | + }, |
| 200 | + "net50_ubuntu1804": { |
| 201 | + "framework": "net50", |
| 202 | + "operatingSystem": "ubuntu-18.04" |
| 203 | + }, |
| 204 | + "netcoreapp21_windows2019": { |
| 205 | + "framework": "netcoreapp2.1", |
| 206 | + "operatingSystem": "windows-2019" |
| 207 | + }, |
| 208 | + "UseProjectRef_net461_windows2019": { |
| 209 | + "additionalTestArguments": "/p:UseProjectReferenceToAzureClients=true", |
| 210 | + "framework": "net461", |
| 211 | + "operatingSystem": "windows-2019" |
| 212 | + } |
| 213 | +} |
| 214 | +``` |
| 215 | + |
| 216 | +The top level keys are used as job names, meaning they get displayed in the azure pipelines UI when running the pipeline. |
| 217 | + |
| 218 | +The logic for generating display names works like this: |
| 219 | + |
| 220 | +- Join parameter values by "_" |
| 221 | + a. If the parameter value exists as a key in `displayNames` in the matrix config, replace it with that value. |
| 222 | + b. For each name value, strip all non-alphanumeric characters (excluding "_"). |
| 223 | + c. If the name is greater than 100 characters, truncate it. |
| 224 | + |
| 225 | +#### Filters |
| 226 | + |
| 227 | +Filters can be passed to the matrix as an array of strings, each matching the format of <key>=<regex>. When a matrix entry |
| 228 | +does not contain the specified key, it will default to a value of empty string for regex parsing. This can be used to specify |
| 229 | +filters for keys that don't exist or keys that optionally exist and match a regex, as seen in the below example. |
| 230 | + |
| 231 | +Display name filters can also be passed as a single regex string that runs against the [generated display name](#displaynames) of the matrix job. |
| 232 | +The intent of display name filters is to be defined primarily as a top level variable at template queue time in the azure pipelines UI. |
| 233 | + |
| 234 | +For example, the below command will filter for matrix entries with "windows" in the job display name, no matrix variable |
| 235 | +named "ExcludedKey", a framework variable containing either "461" or "5.0", and an optional key "SupportedClouds" that, if exists, must contain "Public": |
| 236 | + |
| 237 | +``` |
| 238 | +./Create-JobMatrix.ps1 ` |
| 239 | + -ConfigPath samples/matrix.json ` |
| 240 | + -Selection all ` |
| 241 | + -DisplayNameFilter ".*windows.*" ` |
| 242 | + -Filters @("ExcludedKey=^$", "framework=(461|5\.0)", "SupportedClouds=^$|.*Public.*") |
| 243 | +``` |
| 244 | + |
| 245 | +#### Under the hood |
| 246 | + |
| 247 | +The script generates an N-dimensional matrix with dimensions equal to the parameter array lengths. For example, |
| 248 | +the below config would generate a 2x2x1x1x1 matrix (five-dimensional): |
| 249 | + |
| 250 | +``` |
| 251 | +"matrix": { |
| 252 | + "framework": [ "net461", "netcoreapp2.1" ], |
| 253 | + "additionalTestArguments": [ "", "/p:SuperTest=true" ] |
| 254 | + "pool": [ "ubuntu-18.04" ], |
| 255 | + "container": [ "ubuntu-18.04" ], |
| 256 | + "testMode": [ "Record" ] |
| 257 | +} |
| 258 | +``` |
| 259 | + |
| 260 | +The matrix is stored as a one-dimensional array, with a row-major indexing scheme (e.g. `(2, 1, 0, 1, 0)`). |
| 261 | + |
| 262 | +## Testing |
| 263 | + |
| 264 | +The matrix functions can be tested using [pester](https://pester.dev/): |
| 265 | + |
| 266 | +``` |
| 267 | +$ Invoke-Pester |
| 268 | +
|
| 269 | +Starting discovery in 1 files. |
| 270 | +Discovery finished in 384ms. |
| 271 | +[+] /home/ben/sdk/azure-sdk-for-net/eng/scripts/job-matrix/job-matrix-functions.tests.ps1 4.09s (1.52s|2.22s) |
| 272 | +Tests completed in 4.12s |
| 273 | +Tests Passed: 120, Failed: 0, Skipped: 4 NotRun: 0 |
| 274 | +``` |
0 commit comments