microsoft/openvmm
Publicmirrored from https://github.com/microsoft/openvmmAvailable
flowey/flowey_cli/src/pipeline_resolver/common_yaml.rs
370lines · modecode
| 1 | // Copyright (c) Microsoft Corporation. |
| 2 | // Licensed under the MIT License. |
| 3 | |
| 4 | //! Shared functionality for emitting a pipeline as ADO/GitHub YAML files |
| 5 | |
| 6 | use crate::cli::exec_snippet::FloweyPipelineStaticDb; |
| 7 | use crate::cli::exec_snippet::SerializedRequest; |
| 8 | use crate::cli::pipeline::CheckMode; |
| 9 | use crate::pipeline_resolver::generic::ResolvedPipelineJob; |
| 10 | use anyhow::Context; |
| 11 | use flowey_core::node::FlowArch; |
| 12 | use flowey_core::node::FlowPlatform; |
| 13 | use petgraph::visit::EdgeRef; |
| 14 | use serde::Serialize; |
| 15 | use serde_yaml::Value; |
| 16 | use std::collections::BTreeMap; |
| 17 | use std::collections::BTreeSet; |
| 18 | use std::io::Write; |
| 19 | use std::path::Path; |
| 20 | |
| 21 | /// The output of resolving a flow into a sequence of YAML steps (shared by ADO |
| 22 | /// and GitHub resolvers). |
| 23 | pub(crate) struct ResolvedFlowSteps { |
| 24 | /// The resolved YAML steps for this job. |
| 25 | pub steps: Vec<Value>, |
| 26 | /// The serialized request database, keyed by node module path. |
| 27 | pub request_db: BTreeMap<String, Vec<SerializedRequest>>, |
| 28 | /// The serialized config database, keyed by node module path. |
| 29 | pub config_db: BTreeMap<String, Vec<SerializedRequest>>, |
| 30 | } |
| 31 | |
| 32 | #[derive(Debug)] |
| 33 | pub(crate) enum FloweySource { |
| 34 | // bool indicates if this node should publish the flowey it bootstraps for |
| 35 | // other nodes to consume |
| 36 | Bootstrap(String, bool), |
| 37 | Consume(String), |
| 38 | } |
| 39 | |
| 40 | /// each job has one of three "roles" when it comes to bootstrapping flowey: |
| 41 | /// |
| 42 | /// 1. Build flowey |
| 43 | /// 2. Building _and_ publishing flowey |
| 44 | /// 3. Consuming a pre-built flowey |
| 45 | /// |
| 46 | /// We _could_ just have every bootstrap job also publish flowey, but this |
| 47 | /// will spam the artifact feed with artifacts no one will consume, which is |
| 48 | /// wasteful. |
| 49 | /// |
| 50 | /// META: why go through all this hassle anyways? i.e: why not just do |
| 51 | /// something dead simple like: |
| 52 | /// |
| 53 | /// - discover which platforms exist in the graph |
| 54 | /// - have the first jobs of every pipeline be standalone "bootstrap flowey" |
| 55 | /// jobs, which all subsequent jobs of a certain platform can take a dep on |
| 56 | /// |
| 57 | /// well... it turns out that provisioning job runners is _sloooooow_, |
| 58 | /// and having every single pipeline run these "bootstrap flowey" steps |
| 59 | /// gating the rest of the "interesting" stuff would really stink. |
| 60 | /// |
| 61 | /// i.e: it's better to do redundant flowey bootstraps if it means that we |
| 62 | /// can avoid the extra time it takes to tear down + re-provision a worker. |
| 63 | pub(crate) fn job_flowey_bootstrap_source( |
| 64 | graph: &petgraph::Graph<ResolvedPipelineJob, ()>, |
| 65 | order: &Vec<petgraph::prelude::NodeIndex>, |
| 66 | ) -> BTreeMap<petgraph::prelude::NodeIndex, FloweySource> { |
| 67 | let mut bootstrapped_flowey = BTreeMap::new(); |
| 68 | |
| 69 | // the first traversal builds a list of all ancestors of a give node |
| 70 | let mut ancestors = BTreeMap::< |
| 71 | petgraph::prelude::NodeIndex, |
| 72 | BTreeSet<(petgraph::prelude::NodeIndex, FlowPlatform, FlowArch)>, |
| 73 | >::new(); |
| 74 | for idx in order { |
| 75 | for ancestor_idx in graph |
| 76 | .edges_directed(*idx, petgraph::Direction::Incoming) |
| 77 | .map(|e| e.source()) |
| 78 | { |
| 79 | ancestors.entry(*idx).or_default().insert(( |
| 80 | ancestor_idx, |
| 81 | graph[ancestor_idx].platform, |
| 82 | graph[ancestor_idx].arch, |
| 83 | )); |
| 84 | |
| 85 | if let Some(set) = ancestors.get(&ancestor_idx).cloned() { |
| 86 | ancestors.get_mut(idx).unwrap().extend(&set); |
| 87 | } |
| 88 | } |
| 89 | } |
| 90 | |
| 91 | // the second traversal assigns roles to each node |
| 92 | let mut floweyno = 0; |
| 93 | 'outer: for idx in order { |
| 94 | let ancestors = ancestors.remove(idx).unwrap_or_default(); |
| 95 | |
| 96 | let mut elect_bootstrap = None; |
| 97 | |
| 98 | for (ancestor_idx, platform, arch) in ancestors { |
| 99 | if platform != graph[*idx].platform || arch != graph[*idx].arch { |
| 100 | continue; |
| 101 | } |
| 102 | |
| 103 | let role = |
| 104 | bootstrapped_flowey |
| 105 | .get_mut(&ancestor_idx) |
| 106 | .and_then(|existing| match existing { |
| 107 | FloweySource::Bootstrap(s, true) => Some(FloweySource::Consume(s.clone())), |
| 108 | FloweySource::Consume(s) => Some(FloweySource::Consume(s.clone())), |
| 109 | // there is an ancestor that is building, but not |
| 110 | // publishing. maybe they should get upgraded... |
| 111 | FloweySource::Bootstrap(_, false) => { |
| 112 | elect_bootstrap = Some(ancestor_idx); |
| 113 | None |
| 114 | } |
| 115 | }); |
| 116 | |
| 117 | if let Some(role) = role { |
| 118 | bootstrapped_flowey.insert(*idx, role); |
| 119 | continue 'outer; |
| 120 | } |
| 121 | } |
| 122 | |
| 123 | // if we got here, that means we couldn't find a valid ancestor. |
| 124 | // |
| 125 | // check if we can upgrade an existing ancestor vs. bootstrapping |
| 126 | // things ourselves |
| 127 | if let Some(elect_bootstrap) = elect_bootstrap { |
| 128 | let FloweySource::Bootstrap(s, publish) = |
| 129 | bootstrapped_flowey.get_mut(&elect_bootstrap).unwrap() |
| 130 | else { |
| 131 | unreachable!() |
| 132 | }; |
| 133 | |
| 134 | *publish = true; |
| 135 | let s = s.clone(); |
| 136 | |
| 137 | bootstrapped_flowey.insert(*idx, FloweySource::Consume(s)); |
| 138 | } else { |
| 139 | // Having this extra unique `floweyno` per bootstrap is |
| 140 | // necessary since GitHub doesn't let you double-publish an |
| 141 | // artifact with the same name |
| 142 | floweyno += 1; |
| 143 | let platform = graph[*idx].platform; |
| 144 | let arch = graph[*idx].arch; |
| 145 | bootstrapped_flowey.insert( |
| 146 | *idx, |
| 147 | FloweySource::Bootstrap( |
| 148 | format!("_internal-flowey-bootstrap-{arch}-{platform}-uid-{floweyno}"), |
| 149 | false, |
| 150 | ), |
| 151 | ); |
| 152 | } |
| 153 | } |
| 154 | |
| 155 | bootstrapped_flowey |
| 156 | } |
| 157 | |
| 158 | /// convert `pipeline` to YAML and `pipeline_static_db` to JSON. |
| 159 | /// if `check` is `Some`, then we will compare the generated YAML and JSON |
| 160 | /// against the contents of `check` and error if they don't match. |
| 161 | /// if `check` is `None`, then we will write the generated YAML and JSON to |
| 162 | /// `repo_root/pipeline_file.yaml` and `repo_root/pipeline_file.json` respectively. |
| 163 | fn check_or_write_generated_yaml_and_json<T>( |
| 164 | pipeline: &T, |
| 165 | pipeline_static_db: &FloweyPipelineStaticDb, |
| 166 | mode: CheckMode, |
| 167 | repo_root: &Path, |
| 168 | pipeline_file: &Path, |
| 169 | ado_post_process_yaml_cb: Option<Box<dyn FnOnce(Value) -> Value>>, |
| 170 | ) -> anyhow::Result<()> |
| 171 | where |
| 172 | T: Serialize, |
| 173 | { |
| 174 | let generated_yaml = |
| 175 | serde_yaml::to_value(pipeline).context("while serializing pipeline yaml")?; |
| 176 | let generated_yaml = if let Some(ado_post_process_yaml_cb) = ado_post_process_yaml_cb { |
| 177 | ado_post_process_yaml_cb(generated_yaml) |
| 178 | } else { |
| 179 | generated_yaml |
| 180 | }; |
| 181 | |
| 182 | let generated_yaml = |
| 183 | serde_yaml::to_string(&generated_yaml).context("while emitting pipeline yaml")?; |
| 184 | let generated_yaml = format!( |
| 185 | r#" |
| 186 | ############################## |
| 187 | # THIS FILE IS AUTOGENERATED # |
| 188 | # DO NOT MANUALLY EDIT # |
| 189 | ############################## |
| 190 | {generated_yaml}"# |
| 191 | ); |
| 192 | let generated_yaml = generated_yaml.trim_start(); |
| 193 | |
| 194 | let generated_json = |
| 195 | serde_json::to_string_pretty(pipeline_static_db).context("while emitting pipeline json")?; |
| 196 | |
| 197 | match mode { |
| 198 | CheckMode::Runtime(ref check_file) | CheckMode::Check(ref check_file) => { |
| 199 | let existing_yaml = fs_err::read_to_string(check_file) |
| 200 | .context("cannot check pipeline that doesn't exist!")?; |
| 201 | |
| 202 | let yaml_out_of_date = existing_yaml != generated_yaml; |
| 203 | |
| 204 | if yaml_out_of_date { |
| 205 | println!( |
| 206 | "generated yaml {}:\n==========\n{generated_yaml}", |
| 207 | generated_yaml.len() |
| 208 | ); |
| 209 | println!( |
| 210 | "existing yaml {}:\n==========\n{existing_yaml}", |
| 211 | existing_yaml.len() |
| 212 | ); |
| 213 | } |
| 214 | |
| 215 | if yaml_out_of_date { |
| 216 | anyhow::bail!("checked in pipeline YAML is out of date! run `cargo xflowey regen`") |
| 217 | } |
| 218 | |
| 219 | // Only write the JSON if we're in runtime mode, not in check mode |
| 220 | if let CheckMode::Runtime(_) = mode { |
| 221 | let mut f = fs_err::File::create(check_file.with_extension("json"))?; |
| 222 | f.write_all(generated_json.as_bytes()) |
| 223 | .context("while emitting pipeline database json")?; |
| 224 | } |
| 225 | |
| 226 | Ok(()) |
| 227 | } |
| 228 | CheckMode::None => { |
| 229 | let out_yaml_path = repo_root.join(pipeline_file); |
| 230 | |
| 231 | let mut f = fs_err::File::create(out_yaml_path)?; |
| 232 | f.write_all(generated_yaml.as_bytes()) |
| 233 | .context("while emitting pipeline yaml")?; |
| 234 | |
| 235 | Ok(()) |
| 236 | } |
| 237 | } |
| 238 | } |
| 239 | |
| 240 | /// See [`check_or_write_generated_yaml_and_json`] |
| 241 | pub(crate) fn check_generated_yaml_and_json<T>( |
| 242 | pipeline: &T, |
| 243 | pipeline_static_db: &FloweyPipelineStaticDb, |
| 244 | check: CheckMode, |
| 245 | repo_root: &Path, |
| 246 | pipeline_file: &Path, |
| 247 | ado_post_process_yaml_cb: Option<Box<dyn FnOnce(Value) -> Value>>, |
| 248 | ) -> anyhow::Result<()> |
| 249 | where |
| 250 | T: Serialize, |
| 251 | { |
| 252 | check_or_write_generated_yaml_and_json( |
| 253 | pipeline, |
| 254 | pipeline_static_db, |
| 255 | check, |
| 256 | repo_root, |
| 257 | pipeline_file, |
| 258 | ado_post_process_yaml_cb, |
| 259 | ) |
| 260 | } |
| 261 | |
| 262 | /// See [`check_or_write_generated_yaml_and_json`] |
| 263 | pub(crate) fn write_generated_yaml_and_json<T>( |
| 264 | pipeline: &T, |
| 265 | pipeline_static_db: &FloweyPipelineStaticDb, |
| 266 | repo_root: &Path, |
| 267 | pipeline_file: &Path, |
| 268 | ado_post_process_yaml_cb: Option<Box<dyn FnOnce(Value) -> Value>>, |
| 269 | ) -> anyhow::Result<()> |
| 270 | where |
| 271 | T: Serialize, |
| 272 | { |
| 273 | check_or_write_generated_yaml_and_json( |
| 274 | pipeline, |
| 275 | pipeline_static_db, |
| 276 | CheckMode::None, |
| 277 | repo_root, |
| 278 | pipeline_file, |
| 279 | ado_post_process_yaml_cb, |
| 280 | ) |
| 281 | } |
| 282 | |
| 283 | /// Merges a list of bash commands into a single YAML step. |
| 284 | pub(crate) struct BashCommands { |
| 285 | commands: Vec<String>, |
| 286 | label: Option<String>, |
| 287 | can_merge: bool, |
| 288 | github: bool, |
| 289 | } |
| 290 | |
| 291 | impl BashCommands { |
| 292 | pub fn new_github() -> Self { |
| 293 | Self { |
| 294 | commands: Vec::new(), |
| 295 | label: None, |
| 296 | can_merge: true, |
| 297 | github: true, |
| 298 | } |
| 299 | } |
| 300 | |
| 301 | pub fn new_ado() -> Self { |
| 302 | Self { |
| 303 | commands: Vec::new(), |
| 304 | label: None, |
| 305 | can_merge: true, |
| 306 | github: false, |
| 307 | } |
| 308 | } |
| 309 | |
| 310 | #[must_use] |
| 311 | pub fn push( |
| 312 | &mut self, |
| 313 | label: Option<String>, |
| 314 | can_merge: bool, |
| 315 | mut cmd: String, |
| 316 | ) -> Option<Value> { |
| 317 | let val = if !can_merge && !self.can_merge { |
| 318 | self.flush() |
| 319 | } else { |
| 320 | None |
| 321 | }; |
| 322 | if !can_merge || self.label.is_none() { |
| 323 | self.label = label; |
| 324 | } |
| 325 | cmd.truncate(cmd.trim_end().len()); |
| 326 | self.commands.push(cmd); |
| 327 | self.can_merge &= can_merge; |
| 328 | val |
| 329 | } |
| 330 | |
| 331 | pub fn push_minor(&mut self, cmd: String) { |
| 332 | assert!(self.push(None, true, cmd).is_none()); |
| 333 | } |
| 334 | |
| 335 | #[must_use] |
| 336 | pub fn flush(&mut self) -> Option<Value> { |
| 337 | if self.commands.is_empty() { |
| 338 | return None; |
| 339 | } |
| 340 | let label = if self.commands.len() == 1 || !self.can_merge { |
| 341 | self.label.take() |
| 342 | } else { |
| 343 | None |
| 344 | }; |
| 345 | let label = label.unwrap_or_else(|| "🦀 flowey rust steps".into()); |
| 346 | let map = if self.github { |
| 347 | let commands = self.commands.join("\n"); |
| 348 | serde_yaml::Mapping::from_iter([ |
| 349 | ("name".into(), label.into()), |
| 350 | ("run".into(), commands.into()), |
| 351 | ("shell".into(), "bash".into()), |
| 352 | ]) |
| 353 | } else { |
| 354 | let commands = if self.commands.len() == 1 { |
| 355 | self.commands.drain(..).next().unwrap() |
| 356 | } else { |
| 357 | // ADO doesn't automatically fail on error on multi-line scripts. |
| 358 | self.commands.insert(0, "set -e".into()); |
| 359 | self.commands.join("\n") |
| 360 | }; |
| 361 | serde_yaml::Mapping::from_iter([ |
| 362 | ("bash".into(), commands.into()), |
| 363 | ("displayName".into(), label.into()), |
| 364 | ]) |
| 365 | }; |
| 366 | self.commands.clear(); |
| 367 | self.can_merge = true; |
| 368 | Some(map.into()) |
| 369 | } |
| 370 | } |
| 371 | |