Merge branch 'worktree-agent-ad8a36dfcd4176015' into spend-guardrails
41 files+1038−1350/41 viewed
| 1063 | 1063 | or a comment made with it runs nothing, so a workflow cannot set itself | |
| 1064 | 1064 | off. `workflow_dispatch` and [`repository_dispatch`](#repository-dispatch) | |
| 1065 | 1065 | are the exceptions, for a workflow that means to start another. | |
| 1066 | + | - It **never puts g1t to work**. A comment it posts that mentions | |
| 1067 | + | `@g1t` starts nothing, and it cannot assign an issue or a plan to g1t, | |
| 1068 | + | queue one for it, hand it work or ask it for a review. Otherwise a | |
| 1069 | + | workflow that asks g1t to fix a failing check would run again on g1t's | |
| 1070 | + | push, and ask again, without end. A step that should put g1t to work | |
| 1071 | + | uses a token of a person's own, stored as a | |
| 1072 | + | [secret](/guides/secrets-and-variables/). | |
| 1066 | 1073 | ||
| 1067 | 1074 | `permissions:` goes at the top of the workflow, for every job, or on a job, | |
| 1068 | 1075 | which then ignores the workflow's. Once either is written, every permission |
| 184 | 184 | ||
| 185 | 185 | Each push sends only what the one before did not. | |
| 186 | 186 | ||
| 187 | + | ### Pushes of many branches or tags | |
| 188 | + | ||
| 189 | + | Every branch and tag in a push is stored. Each one is also announced as a | |
| 190 | + | `git.push` event, which starts workflows, mirrors the repository and | |
| 191 | + | calls webhooks, except in a push of many: | |
| 192 | + | ||
| 193 | + | | A push of | What is announced | | |
| 194 | + | | --- | --- | | |
| 195 | + | | Up to 3 tags | Each tag | | |
| 196 | + | | More than 3 tags (`git push --tags`, say) | None of the tags | | |
| 197 | + | | Up to 1,000 branches | Each branch | | |
| 198 | + | | More than 1,000 branches | Only the default branch, if it moved | | |
| 199 | + | ||
| 200 | + | To have tags start workflows, push them 3 or fewer at a time. | |
| 201 | + | ||
| 187 | 202 | ### When the store is busy | |
| 188 | 203 | ||
| 189 | 204 | If Cloudflare Artifacts is rate limiting g1t or not answering, g1t tries |
| 48 | 48 | ||
| 49 | 49 | `data` holds what the event is about: ids and numbers to fetch the rest with | |
| 50 | 50 | the [API](/reference/api/). `actor` is null for something g1t did by | |
| 51 | − | itself. | |
| 51 | + | itself. An event whose `data` would be larger than about 96 KB has its | |
| 52 | + | long text, such as a comment's or release's body, shortened, and | |
| 53 | + | `data.truncated` is `true`: fetch the whole text from the API. | |
| 52 | 54 | ||
| 53 | 55 | With these headers: | |
| 54 | 56 | ||
| ⋯ | |||
| 64 | 66 | ||
| 65 | 67 | | Event | When | | |
| 66 | 68 | | --- | --- | | |
| 67 | − | | `git.push` | A branch moved. `data.ref`, `data.after`, `data.default_branch`. | | |
| 69 | + | | `git.push` | A branch moved. `data.ref`, `data.after`, `data.default_branch`. A push of more than 3 tags, or more than 1,000 branches, sends fewer; see [pushes of many branches or tags](/guides/git/#pushes-of-many-branches-or-tags). | | |
| 68 | 70 | | `repo.created`, `repo.forked` | A repository was made, or forked for a pull request. | | |
| 69 | 71 | | `repo.updated` | Its description, website, topics, protection or visibility changed. | | |
| 70 | 72 | | `repo.visibility_changed` | It was made public or private. | | |
| 414 | 414 | (`ops@g1t.sh`), in URLs, in package scopes (`@g1t/platform`) and in longer | |
| 415 | 415 | names (`@g1t-bot`) are ignored. Matching ignores case. Agents mentioning | |
| 416 | 416 | `@g1t` start nothing, so | |
| 417 | − | agents cannot set each other to work this way. | |
| 417 | + | agents cannot set each other to work this way. Neither does a comment made | |
| 418 | + | with a workflow job's own token (`G1T_TOKEN`), so a workflow cannot set | |
| 419 | + | off g1t whose push sets off the workflow again; see | |
| 420 | + | [the job's token](/guides/actions/#the-jobs-token). | |
| 418 | 421 | ||
| 419 | 422 | **Who can.** People with the Write [role](/guides/access-and-roles/) or higher on the | |
| 420 | 423 | repository, members or not. Anyone else who mentions it gets a short reply | |
| ⋯ | |||
| 428 | 431 | ||
| 429 | 432 | Each comment starts one run at most; to ask again, write a new comment. | |
| 430 | 433 | ||
| 434 | + | A mention sends g1t back to a pull request it made even after it stopped | |
| 435 | + | there, or used up the repository's revisions. At most 10 mentions in a day | |
| 436 | + | send it back to the same pull request; past that, g1t replies that it | |
| 437 | + | has reached the most it takes, and the next one works a day after the | |
| 438 | + | first of those 10. | |
| 439 | + | ||
| 431 | 440 | ## The label rule | |
| 432 | 441 | ||
| 433 | 442 | Under a project's **Settings → Agents**, someone with the Maintain role or | |
| 39 | 39 | pub mod security; | |
| 40 | 40 | pub mod teams; | |
| 41 | 41 | pub mod security_suite; | |
| 42 | + | pub mod subscribers; | |
| 42 | 43 | pub mod time; | |
| 43 | 44 | pub mod tokens; | |
| 44 | 45 | pub mod updates; |
| 1 | + | //! Which events each subscribing service's queue is sent. | |
| 2 | + | //! | |
| 3 | + | //! The events service passes each batch on to one queue per subscriber | |
| 4 | + | //! (`SUBSCRIBER_*` bindings). Each is sent only the types its consumer acts | |
| 5 | + | //! on, so an event costs a queue write for the services that read it, not | |
| 6 | + | //! for all of them. A consumer that starts acting on a new type adds it | |
| 7 | + | //! here in the same change, or it never hears of it; the tests below and | |
| 8 | + | //! beside each consumer check what they can. | |
| 9 | + | //! | |
| 10 | + | //! A pattern is an exact type (`git.push`), a family (`pull.*`, every type | |
| 11 | + | //! that starts `pull.`), or `*`, every type. A binding not named here is | |
| 12 | + | //! sent everything, so a new subscriber is never left out. | |
| 13 | + | ||
| 14 | + | /// What the shared helpers in `g1t_kit` act on (`rename`, `transfer`, | |
| 15 | + | /// `lifecycle`, `deleted`, `user_deleted`): rows that follow a workspace | |
| 16 | + | /// or repository, or go with it. Rare, so every subscriber is sent them. | |
| 17 | + | const LIFECYCLE: [&str; 8] = [ | |
| 18 | + | "workspace.renamed", | |
| 19 | + | "workspace.deleted", | |
| 20 | + | "repo.renamed", | |
| 21 | + | "repo.transferred", | |
| 22 | + | "repo.deleted", | |
| 23 | + | "repo.restored", | |
| 24 | + | "repo.purged", | |
| 25 | + | "user.deleted", | |
| 26 | + | ]; | |
| 27 | + | ||
| 28 | + | /// Each subscriber's binding and the types it is sent beside [`LIFECYCLE`]. | |
| 29 | + | pub const ROUTES: &[(&str, &[&str])] = &[ | |
| 30 | + | // Memory captured from merges and comments, mentions, pull request | |
| 31 | + | // bases, runs stopped in archived repositories (work/src/lib.rs queue). | |
| 32 | + | ( | |
| 33 | + | "SUBSCRIBER_WORK", | |
| 34 | + | &[ | |
| 35 | + | "git.push", | |
| 36 | + | "branch.renamed", | |
| 37 | + | "repo.created", | |
| 38 | + | "repo.archived", | |
| 39 | + | "repo.default_branch_changed", | |
| 40 | + | "pull.merged", | |
| 41 | + | "comment.created", | |
| 42 | + | ], | |
| 43 | + | ), | |
| 44 | + | // Pull requests g1t is seeing through, mentions, queued issues and the | |
| 45 | + | // merge queue (runner/src/index.ts queue). | |
| 46 | + | ( | |
| 47 | + | "SUBSCRIBER_RUNNER", | |
| 48 | + | &[ | |
| 49 | + | "pull.opened", | |
| 50 | + | "pull.ready", | |
| 51 | + | "pull.updated", | |
| 52 | + | "pull.mergeability", | |
| 53 | + | "pull.mergecheck", | |
| 54 | + | "pull.merge_requested", | |
| 55 | + | "pull.merged", | |
| 56 | + | "pull.closed", | |
| 57 | + | "checks.completed", | |
| 58 | + | "review.completed", | |
| 59 | + | "queue.changed", | |
| 60 | + | "comment.created", | |
| 61 | + | "issue.opened", | |
| 62 | + | "issue.updated", | |
| 63 | + | "issue.closed", | |
| 64 | + | "agent.asked", | |
| 65 | + | "repo.archived", | |
| 66 | + | ], | |
| 67 | + | ), | |
| 68 | + | // Push mirrors and write-back to linked trackers (integrations/src). | |
| 69 | + | ("SUBSCRIBER_INTEGRATIONS", &["git.push", "issue.closed", "pull.opened"]), | |
| 70 | + | // A hook may ask for any type, or for every one. | |
| 71 | + | ("SUBSCRIBER_WEBHOOKS", &["*"]), | |
| 72 | + | // Every event a workflow can run on (g1t_actions::events::github_events), | |
| 73 | + | // and runs stopped in archived repositories. | |
| 74 | + | ( | |
| 75 | + | "SUBSCRIBER_ACTIONS", | |
| 76 | + | &[ | |
| 77 | + | "git.push", | |
| 78 | + | "pull.*", | |
| 79 | + | "issue.*", | |
| 80 | + | "comment.*", | |
| 81 | + | "release.*", | |
| 82 | + | "deployment.created", | |
| 83 | + | "deployment_status.created", | |
| 84 | + | "review.completed", | |
| 85 | + | "workflow.completed", | |
| 86 | + | "repo.archived", | |
| 87 | + | ], | |
| 88 | + | ), | |
| 89 | + | // Previews and production deploys (deployments/src/index.ts queue). | |
| 90 | + | ( | |
| 91 | + | "SUBSCRIBER_DEPLOYMENTS", | |
| 92 | + | &[ | |
| 93 | + | "git.push", | |
| 94 | + | "branch.renamed", | |
| 95 | + | "repo.default_branch_changed", | |
| 96 | + | "pull.opened", | |
| 97 | + | "pull.ready", | |
| 98 | + | "pull.updated", | |
| 99 | + | "pull.reopened", | |
| 100 | + | "pull.closed", | |
| 101 | + | "pull.merged", | |
| 102 | + | "workspace.deleting", | |
| 103 | + | "workspace.restored", | |
| 104 | + | ], | |
| 105 | + | ), | |
| 106 | + | // Projects follow their repositories, and wake on activity | |
| 107 | + | // (projects/src/index.ts, activity.ts). | |
| 108 | + | ( | |
| 109 | + | "SUBSCRIBER_PROJECTS", | |
| 110 | + | &[ | |
| 111 | + | "git.push", | |
| 112 | + | "repo.*", | |
| 113 | + | "issue.opened", | |
| 114 | + | "issue.closed", | |
| 115 | + | "issue.reopened", | |
| 116 | + | "pull.opened", | |
| 117 | + | "pull.updated", | |
| 118 | + | "pull.merged", | |
| 119 | + | "pull.closed", | |
| 120 | + | "review.completed", | |
| 121 | + | "comment.created", | |
| 122 | + | "deployment.succeeded", | |
| 123 | + | "deployment.failed", | |
| 124 | + | "package.published", | |
| 125 | + | "package.deleted", | |
| 126 | + | "package.version_deleted", | |
| 127 | + | "workspace.deleting", | |
| 128 | + | "workspace.restored", | |
| 129 | + | ], | |
| 130 | + | ), | |
| 131 | + | // Pull request working copies, workspaces' repositories, contributors | |
| 132 | + | // shown as ghost (repos/src/lib.rs handle_event). | |
| 133 | + | ( | |
| 134 | + | "SUBSCRIBER_REPOS", | |
| 135 | + | &[ | |
| 136 | + | "pull.merged", | |
| 137 | + | "pull.closed", | |
| 138 | + | "pull.reopened", | |
| 139 | + | "workspace.deleting", | |
| 140 | + | "workspace.restored", | |
| 141 | + | "user.deleting", | |
| 142 | + | "user.restored", | |
| 143 | + | ], | |
| 144 | + | ), | |
| 145 | + | // Usage rows that follow a renamed workspace or repository. | |
| 146 | + | ("SUBSCRIBER_BILLING", &[]), | |
| 147 | + | // Secret and dependency scans, and version update pull requests | |
| 148 | + | // (security/src/lib.rs on_event). | |
| 149 | + | ( | |
| 150 | + | "SUBSCRIBER_SECURITY", | |
| 151 | + | &[ | |
| 152 | + | "git.push", | |
| 153 | + | "pull.opened", | |
| 154 | + | "pull.updated", | |
| 155 | + | "pull.ready", | |
| 156 | + | "pull.merged", | |
| 157 | + | "pull.closed", | |
| 158 | + | "checks.completed", | |
| 159 | + | "comment.created", | |
| 160 | + | "repo.created", | |
| 161 | + | "repo.visibility_changed", | |
| 162 | + | ], | |
| 163 | + | ), | |
| 164 | + | // What agents are told about a repository (context/src/index.ts). | |
| 165 | + | ( | |
| 166 | + | "SUBSCRIBER_CONTEXT", | |
| 167 | + | &[ | |
| 168 | + | "git.push", | |
| 169 | + | "issue.opened", | |
| 170 | + | "issue.updated", | |
| 171 | + | "issue.closed", | |
| 172 | + | "pull.ready", | |
| 173 | + | "pull.merged", | |
| 174 | + | "memory.changed", | |
| 175 | + | ], | |
| 176 | + | ), | |
| 177 | + | // The site-wide search index (search/src/index.rs). | |
| 178 | + | ("SUBSCRIBER_SEARCH", &["git.push", "repo.*", "issue.*", "pull.*", "user.*", "workspace.*"]), | |
| 179 | + | // Composer packages read from repositories, and packages that follow | |
| 180 | + | // their workspace, team or repository (packages/src/lib.rs on_event). | |
| 181 | + | ( | |
| 182 | + | "SUBSCRIBER_PACKAGES", | |
| 183 | + | &[ | |
| 184 | + | "git.push", | |
| 185 | + | "repo.visibility_changed", | |
| 186 | + | "workspace.deleting", | |
| 187 | + | "workspace.restored", | |
| 188 | + | "team.edited", | |
| 189 | + | "team.deleted", | |
| 190 | + | ], | |
| 191 | + | ), | |
| 192 | + | ]; | |
| 193 | + | ||
| 194 | + | /// Whether `pattern` (see the module's notes) matches the type `kind`. | |
| 195 | + | fn matches(pattern: &str, kind: &str) -> bool { | |
| 196 | + | match pattern.strip_suffix('*') { | |
| 197 | + | Some(prefix) => kind.starts_with(prefix), | |
| 198 | + | None => pattern == kind, | |
| 199 | + | } | |
| 200 | + | } | |
| 201 | + | ||
| 202 | + | /// Whether the subscriber behind `binding` is sent events of type `kind`. | |
| 203 | + | pub fn routed(binding: &str, kind: &str) -> bool { | |
| 204 | + | match ROUTES.iter().find(|(name, _)| *name == binding) { | |
| 205 | + | Some((_, patterns)) => { | |
| 206 | + | LIFECYCLE.contains(&kind) || patterns.iter().any(|pattern| matches(pattern, kind)) | |
| 207 | + | } | |
| 208 | + | None => true, | |
| 209 | + | } | |
| 210 | + | } | |
| 211 | + | ||
| 212 | + | #[cfg(test)] | |
| 213 | + | mod tests { | |
| 214 | + | use super::*; | |
| 215 | + | ||
| 216 | + | /// Types published on the bus that hooks are not offered. | |
| 217 | + | const UNOFFERED: [&str; 18] = [ | |
| 218 | + | "pull.mergecheck", | |
| 219 | + | "pull.mergeability", | |
| 220 | + | "deployment.review_requested", | |
| 221 | + | "memory.changed", | |
| 222 | + | "abuse.flagged", | |
| 223 | + | "invite.created", | |
| 224 | + | "invite.redeemed", | |
| 225 | + | "waitlist.requested", | |
| 226 | + | "user.updated", | |
| 227 | + | "user.deleting", | |
| 228 | + | "user.restored", | |
| 229 | + | "user.deleted", | |
| 230 | + | "user.email_added", | |
| 231 | + | "workspace.updated", | |
| 232 | + | "workspace.renamed", | |
| 233 | + | "workspace.deleting", | |
| 234 | + | "workspace.restored", | |
| 235 | + | "workspace.deleted", | |
| 236 | + | ]; | |
| 237 | + | ||
| 238 | + | fn published(kind: &str) -> bool { | |
| 239 | + | crate::webhooks::EVENT_TYPES.contains(&kind) || UNOFFERED.contains(&kind) | |
| 240 | + | } | |
| 241 | + | ||
| 242 | + | #[test] | |
| 243 | + | fn every_exact_route_names_a_type_that_is_published() { | |
| 244 | + | for (binding, patterns) in ROUTES { | |
| 245 | + | for pattern in patterns.iter().chain(LIFECYCLE.iter()) { | |
| 246 | + | if pattern.ends_with('*') { | |
| 247 | + | let prefix = pattern.trim_end_matches('*'); | |
| 248 | + | assert!( | |
| 249 | + | prefix.is_empty() | |
| 250 | + | || crate::webhooks::EVENT_TYPES.iter().chain(UNOFFERED.iter()).any(|kind| kind.starts_with(prefix)), | |
| 251 | + | "{binding}: {pattern} matches nothing" | |
| 252 | + | ); | |
| 253 | + | } else { | |
| 254 | + | assert!(published(pattern), "{binding}: {pattern} is not a type anything publishes"); | |
| 255 | + | } | |
| 256 | + | } | |
| 257 | + | } | |
| 258 | + | } | |
| 259 | + | ||
| 260 | + | #[test] | |
| 261 | + | fn every_route_is_a_subscriber_binding_named_once() { | |
| 262 | + | let mut names: Vec<&str> = ROUTES.iter().map(|(name, _)| *name).collect(); | |
| 263 | + | assert!(names.iter().all(|name| name.starts_with("SUBSCRIBER_"))); | |
| 264 | + | let count = names.len(); | |
| 265 | + | names.sort_unstable(); | |
| 266 | + | names.dedup(); | |
| 267 | + | assert_eq!(names.len(), count); | |
| 268 | + | assert_eq!(count, 13); | |
| 269 | + | } | |
| 270 | + | ||
| 271 | + | #[test] | |
| 272 | + | fn subscribers_hear_what_they_act_on_and_not_the_rest() { | |
| 273 | + | assert!(routed("SUBSCRIBER_WEBHOOKS", "secret_scanning_alert.created")); | |
| 274 | + | assert!(routed("SUBSCRIBER_WEBHOOKS", "anything.new")); | |
| 275 | + | assert!(routed("SUBSCRIBER_ACTIONS", "pull.labeled")); | |
| 276 | + | assert!(routed("SUBSCRIBER_ACTIONS", "release.published")); | |
| 277 | + | assert!(!routed("SUBSCRIBER_ACTIONS", "check_run.completed")); | |
| 278 | + | assert!(routed("SUBSCRIBER_RUNNER", "comment.created")); | |
| 279 | + | assert!(!routed("SUBSCRIBER_RUNNER", "status.created")); | |
| 280 | + | assert!(routed("SUBSCRIBER_BILLING", "workspace.renamed")); | |
| 281 | + | assert!(!routed("SUBSCRIBER_BILLING", "git.push")); | |
| 282 | + | assert!(routed("SUBSCRIBER_SEARCH", "user.updated")); | |
| 283 | + | assert!(routed("SUBSCRIBER_REPOS", "user.deleting")); | |
| 284 | + | assert!(!routed("SUBSCRIBER_REPOS", "git.push")); | |
| 285 | + | // Every subscriber follows what moves or removes a repository. | |
| 286 | + | for (binding, _) in ROUTES { | |
| 287 | + | for kind in LIFECYCLE { | |
| 288 | + | assert!(routed(binding, kind), "{binding} misses {kind}"); | |
| 289 | + | } | |
| 290 | + | } | |
| 291 | + | } | |
| 292 | + | ||
| 293 | + | #[test] | |
| 294 | + | fn a_subscriber_not_in_the_table_hears_everything() { | |
| 295 | + | assert!(routed("SUBSCRIBER_NEW", "git.push")); | |
| 296 | + | assert!(routed("SUBSCRIBER_NEW", "status.created")); | |
| 297 | + | } | |
| 298 | + | ||
| 299 | + | #[test] | |
| 300 | + | fn a_family_matches_only_its_own_types() { | |
| 301 | + | assert!(matches("pull.*", "pull.opened")); | |
| 302 | + | assert!(!matches("pull.*", "pulls.opened")); | |
| 303 | + | assert!(!matches("pull.*", "pull")); | |
| 304 | + | assert!(matches("*", "x")); | |
| 305 | + | assert!(matches("git.push", "git.push")); | |
| 306 | + | assert!(!matches("git.push", "git.pushed")); | |
| 307 | + | } | |
| 308 | + | } |
| 58 | 58 | "d1": { "database": "g1t-events", "migrations": "migrations" }, | |
| 59 | 59 | "stage": "core", | |
| 60 | 60 | "secrets": [], | |
| 61 | + | "setup": [ | |
| 62 | + | "The dead-letter queue every queue consumer sends what it gave up on to, before any unit that names it deploys: npx wrangler queues create g1t-events-dlq" | |
| 63 | + | ], | |
| 61 | 64 | "self_host": "run" | |
| 62 | 65 | }, | |
| 63 | 66 | "identity": { |
| 611 | 611 | `resources.kv`, then the ids in the configs. | |
| 612 | 612 | - Queues: `npx wrangler queues create <queue>` for each queue in the | |
| 613 | 613 | manifest: `g1t-events`, `g1t-events-<service>` for every subscriber, | |
| 614 | − | `g1t-search-jobs`, `g1t-context-jobs`. | |
| 614 | + | `g1t-search-jobs`, `g1t-context-jobs`, and the dead-letter queue | |
| 615 | + | `g1t-events-dlq` (below). | |
| 615 | 616 | - R2: `npx wrangler r2 bucket create g1t-screenshots`, | |
| 616 | 617 | `npx wrangler r2 bucket create g1t-actions-cache` (with its two | |
| 617 | 618 | lifecycle rules, above), and `npx wrangler r2 bucket create g1t-git-packs`, | |
| ⋯ | |||
| 631 | 632 | that does not exist yet may be refused; deploy that service first with | |
| 632 | 633 | `--only`. | |
| 633 | 634 | ||
| 635 | + | ## The dead-letter queue | |
| 636 | + | ||
| 637 | + | Every queue consumer (`g1t-events`, each `g1t-events-<service>`, | |
| 638 | + | `g1t-search-jobs`, `g1t-context-jobs`) names `max_retries` and the | |
| 639 | + | dead-letter queue `g1t-events-dlq`, so a message that keeps failing stops | |
| 640 | + | after its retries instead of being retried for ever. A consumer bound to a | |
| 641 | + | queue that does not exist fails to deploy, so create it once per account, | |
| 642 | + | before the first deploy that names it: | |
| 643 | + | ||
| 644 | + | ```sh | |
| 645 | + | npx wrangler queues create g1t-events-dlq | |
| 646 | + | ``` | |
| 647 | + | ||
| 648 | + | Nothing consumes it: read what landed there with `npx wrangler queues | |
| 649 | + | info g1t-events-dlq`, fix the cause, and replay by hand if needed. The | |
| 650 | + | manifest lists each unit's dead-letter queues under `queues.dead_letter`. | |
| 651 | + | ||
| 634 | 652 | ## OIDC tokens for workflow jobs | |
| 635 | 653 | ||
| 636 | 654 | The API is the issuer of workflow jobs' OIDC tokens, | |
| 471 | 471 | } | |
| 472 | 472 | const out = stack.units.map(({ config, ...unit }) => ({ | |
| 473 | 473 | ...unit, | |
| 474 | − | queues: { produces: (config?.queues?.producers ?? []).map((q) => q.queue), consumes: (config?.queues?.consumers ?? []).map((q) => q.queue) }, | |
| 474 | + | queues: { | |
| 475 | + | produces: (config?.queues?.producers ?? []).map((q) => q.queue), | |
| 476 | + | consumes: (config?.queues?.consumers ?? []).map((q) => q.queue), | |
| 477 | + | dead_letter: [...new Set((config?.queues?.consumers ?? []).map((q) => q.dead_letter_queue).filter(Boolean))], | |
| 478 | + | }, | |
| 475 | 479 | kv: (config?.kv_namespaces ?? []).map((kv) => stack.resources.kv?.[kv.id] ?? kv.id), | |
| 476 | 480 | r2: (config?.r2_buckets ?? []).map((b) => b.bucket_name), | |
| 477 | 481 | vectorize: (config?.vectorize ?? []).map((v) => v.index_name), |
| 33 | 33 | echo "== Dispatch namespace and event queue" | |
| 34 | 34 | w dispatch-namespace list 2>/dev/null | grep -q "$NAMESPACE" || w dispatch-namespace create "$NAMESPACE" | |
| 35 | 35 | w queues list 2>/dev/null | grep -q g1t-events-deployments || w queues create g1t-events-deployments | |
| 36 | + | # Where every consumer sends what it gave up on (docs/DEPLOYING.md). | |
| 37 | + | w queues list 2>/dev/null | grep -q g1t-events-dlq || w queues create g1t-events-dlq | |
| 36 | 38 | ||
| 37 | 39 | echo "== DNS record and the service's token" | |
| 38 | 40 | if [ -n "${CLOUDFLARE_API_KEY:-}" ] && [ -n "${CLOUDFLARE_EMAIL:-}" ]; then |
| 1114 | 1114 | mod tests { | |
| 1115 | 1115 | use super::*; | |
| 1116 | 1116 | use g1t_contracts::work::{Pull, g1t_author}; | |
| 1117 | + | ||
| 1118 | + | #[test] | |
| 1119 | + | fn every_event_a_workflow_runs_on_is_sent_to_this_queue() { | |
| 1120 | + | for kind in g1t_contracts::webhooks::EVENT_TYPES { | |
| 1121 | + | if !github_events(kind).is_empty() { | |
| 1122 | + | assert!( | |
| 1123 | + | g1t_contracts::subscribers::routed("SUBSCRIBER_ACTIONS", kind), | |
| 1124 | + | "{kind} starts workflows but is not routed to actions (g1t_contracts::subscribers)" | |
| 1125 | + | ); | |
| 1126 | + | } | |
| 1127 | + | } | |
| 1128 | + | } | |
| 1117 | 1129 | use g1t_contracts::{Membership, PrincipalKind}; | |
| 1118 | 1130 | ||
| 1119 | 1131 | fn repo() -> Repo { |
| 37 | 37 | "r2_buckets": [{ "binding": "ACTIONS_CACHE", "bucket_name": "g1t-actions-cache" }], | |
| 38 | 38 | // Every event on the bus: what starts workflows, and pushes that change them. | |
| 39 | 39 | "queues": { | |
| 40 | − | "consumers": [{ "queue": "g1t-events-actions", "max_batch_size": 10, "max_batch_timeout": 1, "max_retries": 3 }] | |
| 40 | + | "consumers": [{ "queue": "g1t-events-actions", "max_batch_size": 10, "max_batch_timeout": 1, "max_retries": 3, "dead_letter_queue": "g1t-events-dlq" }] | |
| 41 | 41 | }, | |
| 42 | 42 | // Schedules, jobs waiting for room, and jobs whose runner went quiet. | |
| 43 | 43 | "triggers": { "crons": ["* * * * *"] }, |
| 33 | 33 | // Events from the bus: a workspace renamed moves its billing to the | |
| 34 | 34 | // new slug (src/rename.rs). | |
| 35 | 35 | "queues": { | |
| 36 | − | "consumers": [{ "queue": "g1t-events-billing", "max_batch_size": 20, "max_batch_timeout": 1 }] | |
| 36 | + | "consumers": [{ "queue": "g1t-events-billing", "max_batch_size": 20, "max_batch_timeout": 1, "max_retries": 3, "dead_letter_queue": "g1t-events-dlq" }] | |
| 37 | 37 | }, | |
| 38 | 38 | "vars": { | |
| 39 | 39 | // What is added to a model run's cost, in percent: g1t's overhead for |
| 37 | 37 | // jobs, one project of a backfill each, so no request does them all. | |
| 38 | 38 | "producers": [{ "binding": "JOBS", "queue": "g1t-context-jobs" }], | |
| 39 | 39 | "consumers": [ | |
| 40 | − | { "queue": "g1t-events-context", "max_batch_size": 20, "max_batch_timeout": 2 }, | |
| 41 | − | { "queue": "g1t-context-jobs", "max_batch_size": 1, "max_batch_timeout": 1, "max_retries": 2 } | |
| 40 | + | { "queue": "g1t-events-context", "max_batch_size": 20, "max_batch_timeout": 2, "max_retries": 3, "dead_letter_queue": "g1t-events-dlq" }, | |
| 41 | + | { "queue": "g1t-context-jobs", "max_batch_size": 1, "max_batch_timeout": 1, "max_retries": 2, "dead_letter_queue": "g1t-events-dlq" } | |
| 42 | 42 | ] | |
| 43 | 43 | }, | |
| 44 | 44 | "observability": { "enabled": true } |
| 34 | 34 | // Pull requests opened, pushed to, closed and merged; pushes to the | |
| 35 | 35 | // default branch. | |
| 36 | 36 | "queues": { | |
| 37 | − | "consumers": [{ "queue": "g1t-events-deployments", "max_batch_size": 20, "max_batch_timeout": 1 }] | |
| 37 | + | "consumers": [{ "queue": "g1t-events-deployments", "max_batch_size": 20, "max_batch_timeout": 1, "max_retries": 3, "dead_letter_queue": "g1t-events-dlq" }] | |
| 38 | 38 | }, | |
| 39 | 39 | // Builds that died, usage counts, idle previews, apps of ended plans, | |
| 40 | 40 | // and charging a month that is over. |
| 1 | + | -- Which subscriber queues an event already reached, written only when | |
| 2 | + | -- passing a batch on failed part way: its retry skips the queues that | |
| 3 | + | -- have it. Rows older than a day are removed by the daily sweep. | |
| 4 | + | CREATE TABLE fanout_sent ( | |
| 5 | + | event_id TEXT PRIMARY KEY, | |
| 6 | + | -- JSON array of SUBSCRIBER_* binding names. | |
| 7 | + | bindings TEXT NOT NULL, | |
| 8 | + | created_at TEXT NOT NULL | |
| 9 | + | ); | |
| 10 | + | CREATE INDEX fanout_sent_by_age ON fanout_sent (created_at); |
| 1 | + | //! Fitting events into queue messages, and batches into `sendBatch` calls. | |
| 2 | + | //! | |
| 3 | + | //! A queue takes up to 100 messages and 256 KB in one `sendBatch`, and up | |
| 4 | + | //! to 128 KB in one message. Batches are cut by count and by size, with | |
| 5 | + | //! room to spare for how the runtime encodes them. An event too large for | |
| 6 | + | //! one message has its long text shortened and is marked `truncated`: | |
| 7 | + | //! whoever needs the whole text reads it from the service that owns it. | |
| 8 | + | ||
| 9 | + | use g1t_contracts::events::Event; | |
| 10 | + | use serde_json::Value; | |
| 11 | + | ||
| 12 | + | /// The most messages in one `sendBatch`. | |
| 13 | + | pub const MAX_BATCH_MESSAGES: usize = 100; | |
| 14 | + | /// The most bytes of JSON in one `sendBatch`, under the queue's 256 KB. | |
| 15 | + | pub const MAX_BATCH_BYTES: usize = 200 * 1024; | |
| 16 | + | /// The most bytes of JSON in one message, under the queue's 128 KB. | |
| 17 | + | pub const MAX_MESSAGE_BYTES: usize = 96 * 1024; | |
| 18 | + | /// The key set in `data` on an event whose long text was shortened. | |
| 19 | + | pub const TRUNCATED: &str = "truncated"; | |
| 20 | + | /// How long text is cut to, longest first, until the event fits. Ids and | |
| 21 | + | /// names are shorter than the last, so they are never cut. | |
| 22 | + | const CUTS: [usize; 3] = [4096, 512, 128]; | |
| 23 | + | ||
| 24 | + | /// An event's size as JSON. | |
| 25 | + | pub fn size(event: &Event) -> usize { | |
| 26 | + | serde_json::to_string(event).map_or(usize::MAX, |json| json.len()) | |
| 27 | + | } | |
| 28 | + | ||
| 29 | + | /// Shortens `event`'s long strings until it fits in one message, and marks | |
| 30 | + | /// it truncated if anything was cut. Ids, numbers and short fields stay, so | |
| 31 | + | /// every consumer still knows what the event is about. | |
| 32 | + | pub fn fit(event: &mut Event) { | |
| 33 | + | if size(event) <= MAX_MESSAGE_BYTES { | |
| 34 | + | return; | |
| 35 | + | } | |
| 36 | + | for cut in CUTS { | |
| 37 | + | shorten(&mut event.data, cut); | |
| 38 | + | mark(&mut event.data); | |
| 39 | + | if size(event) <= MAX_MESSAGE_BYTES { | |
| 40 | + | return; | |
| 41 | + | } | |
| 42 | + | } | |
| 43 | + | // Still too large (a long list, say): only the top level's short | |
| 44 | + | // values are kept. | |
| 45 | + | let kept = match &event.data { | |
| 46 | + | Value::Object(map) => map | |
| 47 | + | .iter() | |
| 48 | + | .filter(|(_, value)| matches!(value, Value::String(_) | Value::Number(_) | Value::Bool(_) | Value::Null)) | |
| 49 | + | .map(|(key, value)| (key.clone(), value.clone())) | |
| 50 | + | .collect(), | |
| 51 | + | _ => serde_json::Map::new(), | |
| 52 | + | }; | |
| 53 | + | event.data = Value::Object(kept); | |
| 54 | + | mark(&mut event.data); | |
| 55 | + | } | |
| 56 | + | ||
| 57 | + | fn shorten(value: &mut Value, cut: usize) { | |
| 58 | + | match value { | |
| 59 | + | Value::String(text) if text.len() > cut => { | |
| 60 | + | let end = (0..=cut).rev().find(|end| text.is_char_boundary(*end)).unwrap_or(0); | |
| 61 | + | text.truncate(end); | |
| 62 | + | } | |
| 63 | + | Value::Array(items) => items.iter_mut().for_each(|item| shorten(item, cut)), | |
| 64 | + | Value::Object(map) => map.values_mut().for_each(|item| shorten(item, cut)), | |
| 65 | + | _ => {} | |
| 66 | + | } | |
| 67 | + | } | |
| 68 | + | ||
| 69 | + | fn mark(data: &mut Value) { | |
| 70 | + | if let Value::Object(map) = data { | |
| 71 | + | map.insert(TRUNCATED.to_owned(), Value::Bool(true)); | |
| 72 | + | } | |
| 73 | + | } | |
| 74 | + | ||
| 75 | + | /// `events` cut into `sendBatch` calls, in order: no more than | |
| 76 | + | /// [`MAX_BATCH_MESSAGES`] and [`MAX_BATCH_BYTES`] each. Events are expected | |
| 77 | + | /// to have been [`fit`]. | |
| 78 | + | pub fn chunks<'a>(events: &[&'a Event]) -> Vec<Vec<&'a Event>> { | |
| 79 | + | let mut chunks: Vec<Vec<&Event>> = Vec::new(); | |
| 80 | + | let mut bytes = 0; | |
| 81 | + | for event in events { | |
| 82 | + | let event_bytes = size(event); | |
| 83 | + | let full = chunks | |
| 84 | + | .last() | |
| 85 | + | .is_none_or(|chunk| chunk.len() >= MAX_BATCH_MESSAGES || bytes + event_bytes > MAX_BATCH_BYTES); | |
| 86 | + | if full { | |
| 87 | + | chunks.push(Vec::new()); | |
| 88 | + | bytes = 0; | |
| 89 | + | } | |
| 90 | + | chunks.last_mut().expect("just pushed").push(event); | |
| 91 | + | bytes += event_bytes; | |
| 92 | + | } | |
| 93 | + | chunks | |
| 94 | + | } | |
| 95 | + | ||
| 96 | + | #[cfg(test)] | |
| 97 | + | mod tests { | |
| 98 | + | use super::*; | |
| 99 | + | use serde_json::json; | |
| 100 | + | ||
| 101 | + | fn event(data: Value) -> Event { | |
| 102 | + | Event { | |
| 103 | + | id: "evt_1".into(), | |
| 104 | + | kind: "comment.edited".into(), | |
| 105 | + | source: "work".into(), | |
| 106 | + | time: "2026-10-08T00:00:00Z".into(), | |
| 107 | + | repo_id: Some("rep_1".into()), | |
| 108 | + | actor: Some("usr_1".into()), | |
| 109 | + | data, | |
| 110 | + | } | |
| 111 | + | } | |
| 112 | + | ||
| 113 | + | #[test] | |
| 114 | + | fn a_small_event_is_left_as_it_is() { | |
| 115 | + | let mut small = event(json!({ "number": 4, "changes": { "body": { "from": "hello" } } })); | |
| 116 | + | fit(&mut small); | |
| 117 | + | assert_eq!(small.data, json!({ "number": 4, "changes": { "body": { "from": "hello" } } })); | |
| 118 | + | } | |
| 119 | + | ||
| 120 | + | #[test] | |
| 121 | + | fn a_large_event_keeps_its_shape_and_ids_with_its_text_shortened() { | |
| 122 | + | let mut large = event(json!({ | |
| 123 | + | "commentId": "cmt_1", | |
| 124 | + | "number": 4, | |
| 125 | + | "comment": { "body": "é".repeat(100_000), "author": "ana" }, | |
| 126 | + | })); | |
| 127 | + | fit(&mut large); | |
| 128 | + | assert!(size(&large) <= MAX_MESSAGE_BYTES); | |
| 129 | + | assert_eq!(large.data[TRUNCATED], true); | |
| 130 | + | assert_eq!(large.data["commentId"], "cmt_1"); | |
| 131 | + | assert_eq!(large.data["number"], 4); | |
| 132 | + | // Still an object, so whoever reads its fields still can. | |
| 133 | + | assert_eq!(large.data["comment"]["author"], "ana"); | |
| 134 | + | assert!(large.data["comment"]["body"].as_str().unwrap().len() <= 4096); | |
| 135 | + | } | |
| 136 | + | ||
| 137 | + | #[test] | |
| 138 | + | fn an_event_too_large_even_shortened_keeps_its_top_level_values() { | |
| 139 | + | let many: Vec<Value> = (0..20_000).map(|n| json!({ "n": n })).collect(); | |
| 140 | + | let mut huge = event(json!({ "pullId": "pul_1", "files": many })); | |
| 141 | + | fit(&mut huge); | |
| 142 | + | assert!(size(&huge) <= MAX_MESSAGE_BYTES); | |
| 143 | + | assert_eq!(huge.data, json!({ "pullId": "pul_1", "truncated": true })); | |
| 144 | + | } | |
| 145 | + | ||
| 146 | + | #[test] | |
| 147 | + | fn batches_are_cut_by_count_and_by_size() { | |
| 148 | + | let small: Vec<Event> = (0..250).map(|_| event(json!({ "number": 1 }))).collect(); | |
| 149 | + | let refs: Vec<&Event> = small.iter().collect(); | |
| 150 | + | let counted: Vec<usize> = chunks(&refs).iter().map(Vec::len).collect(); | |
| 151 | + | assert_eq!(counted, [100, 100, 50]); | |
| 152 | + | ||
| 153 | + | let big: Vec<Event> = (0..5).map(|_| event(json!({ "text": "x".repeat(80 * 1024) }))).collect(); | |
| 154 | + | let refs: Vec<&Event> = big.iter().collect(); | |
| 155 | + | let cut = chunks(&refs); | |
| 156 | + | assert_eq!(cut.iter().map(Vec::len).collect::<Vec<_>>(), [2, 2, 1]); | |
| 157 | + | for chunk in cut { | |
| 158 | + | assert!(chunk.iter().map(|event| size(event)).sum::<usize>() <= MAX_BATCH_BYTES); | |
| 159 | + | } | |
| 160 | + | assert!(chunks(&[]).is_empty()); | |
| 161 | + | } | |
| 162 | + | } |
| 2 | 2 | //! and its durable log. | |
| 3 | 3 | //! | |
| 4 | 4 | //! Publishing puts events on a queue and returns. The queue consumer writes | |
| 5 | − | //! them to the log and passes each batch on to every subscriber's own | |
| 6 | − | //! queue, so a slow or failing subscriber holds up nobody else. | |
| 5 | + | //! them to the log and passes each batch on to each subscriber's own | |
| 6 | + | //! queue, so a slow or failing subscriber holds up nobody else. Each | |
| 7 | + | //! subscriber is sent only the types it acts on | |
| 8 | + | //! (`g1t_contracts::subscribers`), in batches the queues take (fanout.rs). | |
| 7 | 9 | //! | |
| 8 | 10 | //! It keeps two more things beside the log: the audit log (audit.rs) and | |
| 9 | 11 | //! each person's inbox (inbox.rs), written as events arrive, with who | |
| ⋯ | |||
| 14 | 16 | //! arguments. | |
| 15 | 17 | ||
| 16 | 18 | mod audit; | |
| 19 | + | mod fanout; | |
| 17 | 20 | mod inbox; | |
| 18 | 21 | mod subscriptions; | |
| 19 | 22 | ||
| ⋯ | |||
| 22 | 25 | use g1t_contracts::time::rfc3339; | |
| 23 | 26 | use g1t_kit::{args, js, now_ms, reply, rpc_method}; | |
| 24 | 27 | use serde::Deserialize; | |
| 28 | + | use std::collections::{HashMap, HashSet}; | |
| 25 | 29 | use worker::js_sys::{Array, Object}; | |
| 26 | 30 | use worker::wasm_bindgen::{JsCast, JsValue}; | |
| 27 | 31 | use worker::{Context, D1Database, Env, MessageBatch, Request, Response, Result, event}; | |
| ⋯ | |||
| 39 | 43 | "check_suite.completed", | |
| 40 | 44 | "check_suite.rerequested", | |
| 41 | 45 | ]; | |
| 42 | − | /// Every binding whose name starts with this is a queue that receives all | |
| 43 | − | /// events: one per subscribing service. | |
| 46 | + | /// Every binding whose name starts with this is a queue that receives | |
| 47 | + | /// events: one per subscribing service, sent the types it routes. | |
| 44 | 48 | const SUBSCRIBER_PREFIX: &str = "SUBSCRIBER_"; | |
| 49 | + | /// How long a record of which queues a batch reached is kept for its retries. | |
| 50 | + | const FANOUT_KEEP_MS: u64 = 24 * 60 * 60 * 1000; | |
| 45 | 51 | ||
| 46 | 52 | #[derive(Deserialize)] | |
| 47 | 53 | struct EventRow { | |
| ⋯ | |||
| 74 | 80 | value.as_deref().map_or(JsValue::NULL, JsValue::from) | |
| 75 | 81 | } | |
| 76 | 82 | ||
| 77 | − | /// Sends `events` to a queue binding, one message each. | |
| 78 | − | async fn send(queue: &JsValue, events: &[Event]) -> Result<()> { | |
| 79 | − | let messages = Array::new(); | |
| 80 | − | for event in events { | |
| 81 | − | let message = Object::new(); | |
| 82 | − | js::set(&message, "body", &js::to_js(event)?); | |
| 83 | − | messages.push(&message); | |
| 83 | + | #[derive(Deserialize)] | |
| 84 | + | struct FanoutRow { | |
| 85 | + | event_id: String, | |
| 86 | + | /// JSON array of binding names. | |
| 87 | + | bindings: String, | |
| 88 | + | } | |
| 89 | + | ||
| 90 | + | /// Sends `events` to a queue binding, one message each, in as many | |
| 91 | + | /// `sendBatch` calls as the queue's limits need (fanout.rs). Events are | |
| 92 | + | /// expected to have been fit to a message. | |
| 93 | + | async fn send(queue: &JsValue, events: &[&Event]) -> Result<()> { | |
| 94 | + | for chunk in fanout::chunks(events) { | |
| 95 | + | let messages = Array::new(); | |
| 96 | + | for event in chunk { | |
| 97 | + | let message = Object::new(); | |
| 98 | + | js::set(&message, "body", &js::to_js(event)?); | |
| 99 | + | messages.push(&message); | |
| 100 | + | } | |
| 101 | + | js::call(queue, "sendBatch", &[messages.into()]).await?; | |
| 84 | 102 | } | |
| 85 | − | js::call(queue, "sendBatch", &[messages.into()]).await?; | |
| 86 | 103 | Ok(()) | |
| 87 | 104 | } | |
| 88 | 105 | ||
| ⋯ | |||
| 101 | 118 | let events: Vec<Event> = a | |
| 102 | 119 | .events | |
| 103 | 120 | .into_iter() | |
| 104 | − | .map(|event| Event { | |
| 105 | − | id: new_id("evt", now), | |
| 106 | − | kind: event.kind, | |
| 107 | − | source: event.source, | |
| 108 | − | time: rfc3339(now), | |
| 109 | − | repo_id: event.repo_id, | |
| 110 | − | actor: event.actor, | |
| 111 | − | data: event.data, | |
| 121 | + | .map(|event| { | |
| 122 | + | let mut event = Event { | |
| 123 | + | id: new_id("evt", now), | |
| 124 | + | kind: event.kind, | |
| 125 | + | source: event.source, | |
| 126 | + | time: rfc3339(now), | |
| 127 | + | repo_id: event.repo_id, | |
| 128 | + | actor: event.actor, | |
| 129 | + | data: event.data, | |
| 130 | + | }; | |
| 131 | + | // Too large for one message: its long text is shortened. | |
| 132 | + | fanout::fit(&mut event); | |
| 133 | + | event | |
| 112 | 134 | }) | |
| 113 | 135 | .collect(); | |
| 136 | + | let events: Vec<&Event> = events.iter().collect(); | |
| 114 | 137 | send(&js::binding(&self.env, "BUS")?, &events).await | |
| 115 | 138 | } | |
| 116 | 139 | ||
| ⋯ | |||
| 138 | 161 | Ok(rows.into_iter().map(Event::from).collect()) | |
| 139 | 162 | } | |
| 140 | 163 | ||
| 141 | − | /// Writes a batch from the bus to the log, hands it to every | |
| 142 | − | /// subscriber, then tells the people it concerns (inbox.rs). | |
| 164 | + | /// Writes a batch from the bus to the log, notes who now follows what | |
| 165 | + | /// (inbox.rs), hands each subscriber the events it routes, then tells | |
| 166 | + | /// the people it concerns. | |
| 167 | + | /// | |
| 168 | + | /// Everything before the hand-off is safe to repeat. A retried batch | |
| 169 | + | /// skips the queues that already have it: when a hand-off fails part | |
| 170 | + | /// way, which queues each event reached is written down first. | |
| 143 | 171 | async fn deliver(&self, events: &[Event]) -> Result<()> { | |
| 144 | − | let mut statements = Vec::with_capacity(events.len()); | |
| 172 | + | let mut statements = Vec::with_capacity(events.len() + 1); | |
| 145 | 173 | for event in events { | |
| 146 | 174 | statements.push( | |
| 147 | 175 | self.db | |
| ⋯ | |||
| 161 | 189 | ])?, | |
| 162 | 190 | ); | |
| 163 | 191 | } | |
| 164 | − | self.db.batch(statements).await?; | |
| 192 | + | let ids: Vec<&str> = events.iter().map(|event| event.id.as_str()).collect(); | |
| 193 | + | // Read in the same round trip: nearly always nothing. | |
| 194 | + | statements.push( | |
| 195 | + | self.db | |
| 196 | + | .prepare("SELECT event_id, bindings FROM fanout_sent WHERE event_id IN (SELECT value FROM json_each(?))") | |
| 197 | + | .bind(&[serde_json::to_string(&ids)?.into()])?, | |
| 198 | + | ); | |
| 199 | + | let results = self.db.batch(statements).await?; | |
| 200 | + | let mut sent: HashMap<String, HashSet<String>> = HashMap::new(); | |
| 201 | + | if let Some(rows) = results.last() { | |
| 202 | + | for row in rows.results::<FanoutRow>()? { | |
| 203 | + | let bindings: Vec<String> = serde_json::from_str(&row.bindings).unwrap_or_default(); | |
| 204 | + | sent.insert(row.event_id, bindings.into_iter().collect()); | |
| 205 | + | } | |
| 206 | + | } | |
| 165 | 207 | audit::follow_renames(&self.db, events).await?; | |
| 208 | + | // Before the hand-off, so its failing does not send the batch to | |
| 209 | + | // every queue again. | |
| 210 | + | inbox::follow(&self.db, events).await?; | |
| 166 | 211 | ||
| 167 | 212 | let bindings: &JsValue = self.env.as_ref(); | |
| 213 | + | let mut reached: Vec<String> = Vec::new(); | |
| 214 | + | let mut failed = None; | |
| 168 | 215 | for name in Object::keys(bindings.unchecked_ref::<Object>()).iter() { | |
| 169 | 216 | let Some(name) = name | |
| 170 | 217 | .as_string() | |
| ⋯ | |||
| 172 | 219 | else { | |
| 173 | 220 | continue; | |
| 174 | 221 | }; | |
| 175 | − | send(&js::binding(&self.env, &name)?, events).await?; | |
| 222 | + | let routed: Vec<&Event> = events | |
| 223 | + | .iter() | |
| 224 | + | .filter(|event| g1t_contracts::subscribers::routed(&name, &event.kind)) | |
| 225 | + | .filter(|event| !sent.get(&event.id).is_some_and(|reached| reached.contains(&name))) | |
| 226 | + | .collect(); | |
| 227 | + | if routed.is_empty() { | |
| 228 | + | continue; | |
| 229 | + | } | |
| 230 | + | match send(&js::binding(&self.env, &name)?, &routed).await { | |
| 231 | + | Ok(()) => reached.push(name), | |
| 232 | + | Err(error) => { | |
| 233 | + | worker::console_error!("events: passing {} events to {name} failed: {error}", routed.len()); | |
| 234 | + | failed = Some(error); | |
| 235 | + | } | |
| 236 | + | } | |
| 237 | + | } | |
| 238 | + | if let Some(error) = failed { | |
| 239 | + | self.note_reached(events, &sent, &reached).await?; | |
| 240 | + | return Err(error); | |
| 176 | 241 | } | |
| 177 | − | inbox::follow(&self.db, events).await?; | |
| 178 | 242 | let (work, repos, identity) = ( | |
| 179 | 243 | self.env.service("WORK")?, | |
| 180 | 244 | self.env.service("REPOS")?, | |
| ⋯ | |||
| 188 | 252 | inbox::deliver(&self.db, &sources, events).await; | |
| 189 | 253 | Ok(()) | |
| 190 | 254 | } | |
| 255 | + | ||
| 256 | + | /// Writes down which queues each event of a batch has reached, with | |
| 257 | + | /// those it had reached before, for the batch's retry. | |
| 258 | + | async fn note_reached( | |
| 259 | + | &self, | |
| 260 | + | events: &[Event], | |
| 261 | + | sent: &HashMap<String, HashSet<String>>, | |
| 262 | + | reached: &[String], | |
| 263 | + | ) -> Result<()> { | |
| 264 | + | let now = rfc3339(now_ms()); | |
| 265 | + | let mut statements = Vec::with_capacity(events.len()); | |
| 266 | + | for event in events { | |
| 267 | + | let mut bindings: Vec<&str> = reached.iter().map(String::as_str).collect(); | |
| 268 | + | if let Some(before) = sent.get(&event.id) { | |
| 269 | + | bindings.extend(before.iter().map(String::as_str)); | |
| 270 | + | } | |
| 271 | + | bindings.sort_unstable(); | |
| 272 | + | bindings.dedup(); | |
| 273 | + | statements.push( | |
| 274 | + | self.db | |
| 275 | + | .prepare( | |
| 276 | + | "INSERT INTO fanout_sent (event_id, bindings, created_at) VALUES (?, ?, ?) | |
| 277 | + | ON CONFLICT (event_id) DO UPDATE SET bindings = excluded.bindings", | |
| 278 | + | ) | |
| 279 | + | .bind(&[event.id.as_str().into(), serde_json::to_string(&bindings)?.into(), now.as_str().into()])?, | |
| 280 | + | ); | |
| 281 | + | } | |
| 282 | + | self.db.batch(statements).await?; | |
| 283 | + | Ok(()) | |
| 284 | + | } | |
| 191 | 285 | } | |
| 192 | 286 | ||
| 193 | 287 | #[event(fetch)] | |
| ⋯ | |||
| 234 | 328 | } | |
| 235 | 329 | ||
| 236 | 330 | /// Events from the bus. A batch that fails is retried whole, which is safe | |
| 237 | − | /// because both the log and subscribers ignore an event they have seen. | |
| 331 | + | /// because the log ignores an event it has seen, and queues that already | |
| 332 | + | /// have it are skipped (`deliver`). After its retries it goes to the | |
| 333 | + | /// dead-letter queue (wrangler.jsonc). | |
| 238 | 334 | #[event(queue)] | |
| 239 | 335 | async fn queue(batch: MessageBatch<Event>, env: Env, _ctx: Context) -> Result<()> { | |
| 240 | 336 | let events = Events { | |
| ⋯ | |||
| 269 | 365 | let min_days = days("AUDIT_MIN_DAYS", audit::DEFAULT_MIN_DAYS).min(max_days); | |
| 270 | 366 | let Ok(db) = env.d1("DB") else { return }; | |
| 271 | 367 | let now = now_ms(); | |
| 368 | + | let cutoff = rfc3339(now.saturating_sub(FANOUT_KEEP_MS)); | |
| 369 | + | let purged = match db.prepare("DELETE FROM fanout_sent WHERE created_at < ?").bind(&[cutoff.into()]) { | |
| 370 | + | Ok(statement) => statement.run().await.map(|_| ()), | |
| 371 | + | Err(error) => Err(error), | |
| 372 | + | }; | |
| 373 | + | if let Err(error) = purged { | |
| 374 | + | worker::console_error!("could not remove old fan-out records: {error}"); | |
| 375 | + | } | |
| 272 | 376 | match inbox::purge(&db, now).await { | |
| 273 | 377 | Ok(removed) if removed > 0 => worker::console_log!("removed {removed} old inbox items"), | |
| 274 | 378 | Ok(_) => {} | |
| 22 | 22 | { "binding": "BUS", "queue": "g1t-events" }, | |
| 23 | 23 | // One queue per subscribing service, so each consumes, retries | |
| 24 | 24 | // and scales on its own. Bindings named SUBSCRIBER_* receive | |
| 25 | − | // every event. | |
| 25 | + | // the event types g1t_contracts::subscribers routes to them. | |
| 26 | 26 | { "binding": "SUBSCRIBER_WORK", "queue": "g1t-events-work" }, | |
| 27 | 27 | { "binding": "SUBSCRIBER_RUNNER", "queue": "g1t-events-runner" }, | |
| 28 | 28 | { "binding": "SUBSCRIBER_INTEGRATIONS", "queue": "g1t-events-integrations" }, | |
| ⋯ | |||
| 37 | 37 | { "binding": "SUBSCRIBER_SEARCH", "queue": "g1t-events-search" }, | |
| 38 | 38 | { "binding": "SUBSCRIBER_PACKAGES", "queue": "g1t-events-packages" } | |
| 39 | 39 | ], | |
| 40 | − | "consumers": [{ "queue": "g1t-events", "max_batch_size": 100, "max_batch_timeout": 1 }] | |
| 40 | + | "consumers": [{ "queue": "g1t-events", "max_batch_size": 100, "max_batch_timeout": 1, "max_retries": 3, "dead_letter_queue": "g1t-events-dlq" }] | |
| 41 | 41 | }, | |
| 42 | 42 | // Each workspace's audit log keeps as many days as billing says (7 free, | |
| 43 | 43 | // 90 on the plan, or what staff set), asked for over this binding. | |
| 1073 | 1073 | } | |
| 1074 | 1074 | ||
| 1075 | 1075 | /// For the queue: pushes on g1t that GitHub follows. | |
| 1076 | + | /// Whether `event` should push its repository out to GitHub, given the | |
| 1077 | + | /// repositories `pushed` out already for this batch: a sync sends every | |
| 1078 | + | /// ref, so a push of many refs, or many pushes in one batch, is one sync. | |
| 1079 | + | pub fn first_push_in_batch(pushed: &mut std::collections::HashSet<String>, event: &Event) -> bool { | |
| 1080 | + | match event.repo_id.as_deref() { | |
| 1081 | + | Some(repo_id) if event.kind == "git.push" => pushed.insert(repo_id.to_owned()), | |
| 1082 | + | _ => true, | |
| 1083 | + | } | |
| 1084 | + | } | |
| 1085 | + | ||
| 1076 | 1086 | pub async fn on_event(env: &Env, event: &Event) -> Result<()> { | |
| 1077 | 1087 | if event.kind != "git.push" || !AppConfig::from_env(env).configured() { | |
| 1078 | 1088 | return Ok(()); | |
| ⋯ | |||
| 1085 | 1095 | use super::*; | |
| 1086 | 1096 | ||
| 1087 | 1097 | #[test] | |
| 1098 | + | fn a_batch_pushes_each_repository_out_once() { | |
| 1099 | + | let event = |kind: &str, repo: &str| Event { | |
| 1100 | + | id: "evt_1".into(), | |
| 1101 | + | kind: kind.into(), | |
| 1102 | + | source: "repos".into(), | |
| 1103 | + | time: "2026-10-08T00:00:00Z".into(), | |
| 1104 | + | repo_id: Some(repo.into()), | |
| 1105 | + | actor: None, | |
| 1106 | + | data: serde_json::json!({}), | |
| 1107 | + | }; | |
| 1108 | + | let mut pushed = std::collections::HashSet::new(); | |
| 1109 | + | assert!(first_push_in_batch(&mut pushed, &event("git.push", "rep_1"))); | |
| 1110 | + | assert!(!first_push_in_batch(&mut pushed, &event("git.push", "rep_1"))); | |
| 1111 | + | assert!(first_push_in_batch(&mut pushed, &event("git.push", "rep_2"))); | |
| 1112 | + | assert!(first_push_in_batch(&mut pushed, &event("issue.opened", "rep_1"))); | |
| 1113 | + | } | |
| 1114 | + | ||
| 1115 | + | #[test] | |
| 1088 | 1116 | fn times_are_read_as_github_writes_them() { | |
| 1089 | 1117 | assert_eq!(parse_time("1970-01-01T00:00:00Z"), Some(0)); | |
| 1090 | 1118 | assert_eq!(parse_time("2016-07-11T22:14:10Z"), Some(1_468_275_250_000)); | |
| 1580 | 1580 | #[event(queue)] | |
| 1581 | 1581 | async fn queue(batch: MessageBatch<Event>, env: Env, _ctx: Context) -> Result<()> { | |
| 1582 | 1582 | let service = Integrations::new(&env)?; | |
| 1583 | + | let mut pushed = std::collections::HashSet::new(); | |
| 1583 | 1584 | for message in batch.messages()? { | |
| 1584 | 1585 | // A workspace renamed: its rows move to the slug it has now. | |
| 1585 | 1586 | if g1t_kit::rename::on_event(&env, &env.d1("DB")?, message.body(), rename::STATEMENTS).await? { | |
| ⋯ | |||
| 1598 | 1599 | } | |
| 1599 | 1600 | // A repository purged: what was kept for it goes. | |
| 1600 | 1601 | rename::on_purged(&env.d1("DB")?, message.body()).await?; | |
| 1601 | − | github::on_event(&env, message.body()).await?; | |
| 1602 | + | if github::first_push_in_batch(&mut pushed, message.body()) { | |
| 1603 | + | github::on_event(&env, message.body()).await?; | |
| 1604 | + | } | |
| 1602 | 1605 | service.on_event(message.body()).await?; | |
| 1603 | 1606 | message.ack(); | |
| 1604 | 1607 | } | |
| 28 | 28 | ], | |
| 29 | 29 | // Work starting on, and finishing, what an outside system is waiting on. | |
| 30 | 30 | "queues": { | |
| 31 | − | "consumers": [{ "queue": "g1t-events-integrations", "max_batch_size": 20, "max_batch_timeout": 1 }] | |
| 31 | + | "consumers": [{ "queue": "g1t-events-integrations", "max_batch_size": 20, "max_batch_timeout": 1, "max_retries": 3, "dead_letter_queue": "g1t-events-dlq" }] | |
| 32 | 32 | }, | |
| 33 | 33 | "vars": { | |
| 34 | 34 | "API_URL": "https://api.g1t.sh", |
| 47 | 47 | use g1t_contracts::repos::{GetArgs, Repo, RepoPath}; | |
| 48 | 48 | use g1t_contracts::{FailureCode, Outcome, User, Viewer, new_id}; | |
| 49 | 49 | use g1t_kit::{args, now_ms, reply, rpc_method}; | |
| 50 | − | use worker::{Context, Env, Fetcher, MessageBatch, Request, Response, Result, ScheduleContext, ScheduledEvent, event}; | |
| 50 | + | use worker::{Context, Env, Fetcher, MessageBatch, MessageExt, Request, Response, Result, ScheduleContext, ScheduledEvent, event}; | |
| 51 | 51 | ||
| 52 | 52 | use access::{Action, LinkedTo, Target}; | |
| 53 | 53 | use db::{Db, PackageRow}; | |
| ⋯ | |||
| 944 | 944 | #[event(queue)] | |
| 945 | 945 | async fn queue(batch: MessageBatch<Event>, env: Env, _ctx: Context) -> Result<()> { | |
| 946 | 946 | let packages = Packages::from_env(&env)?; | |
| 947 | + | // Each event is acknowledged or retried on its own, so one that fails | |
| 948 | + | // does not run the rest of its batch again. | |
| 947 | 949 | for message in batch.messages()? { | |
| 948 | − | packages.on_event(&env, message.body()).await?; | |
| 950 | + | match packages.on_event(&env, message.body()).await { | |
| 951 | + | Ok(_) => message.ack(), | |
| 952 | + | Err(error) => { | |
| 953 | + | worker::console_error!("packages: event {} failed: {error}", message.body().id); | |
| 954 | + | message.retry(); | |
| 955 | + | } | |
| 956 | + | } | |
| 949 | 957 | } | |
| 950 | 958 | Ok(()) | |
| 951 | 959 | } | |
| 70 | 70 | // Workspaces renamed or deleted, repositories that change visibility, | |
| 71 | 71 | // are renamed, move or are purged. | |
| 72 | 72 | "queues": { | |
| 73 | − | "consumers": [{ "queue": "g1t-events-packages", "max_batch_size": 20, "max_batch_timeout": 1 }] | |
| 73 | + | "consumers": [{ "queue": "g1t-events-packages", "max_batch_size": 20, "max_batch_timeout": 1, "max_retries": 3, "dead_letter_queue": "g1t-events-dlq" }] | |
| 74 | 74 | }, | |
| 75 | 75 | "observability": { "enabled": true } | |
| 76 | 76 | } |
| 26 | 26 | ], | |
| 27 | 27 | // A repository created is given its own project. | |
| 28 | 28 | "queues": { | |
| 29 | − | "consumers": [{ "queue": "g1t-events-projects", "max_batch_size": 20, "max_batch_timeout": 1 }] | |
| 29 | + | "consumers": [{ "queue": "g1t-events-projects", "max_batch_size": 20, "max_batch_timeout": 1, "max_retries": 3, "dead_letter_queue": "g1t-events-dlq" }] | |
| 30 | 30 | }, | |
| 31 | 31 | "observability": { "enabled": true } | |
| 32 | 32 | } |
| 379 | 379 | use super::*; | |
| 380 | 380 | ||
| 381 | 381 | #[test] | |
| 382 | + | fn every_event_that_settles_a_copy_is_sent_to_this_queue() { | |
| 383 | + | let users = ["user.deleting", "user.restored"]; | |
| 384 | + | for kind in g1t_contracts::webhooks::EVENT_TYPES.iter().chain(users.iter()) { | |
| 385 | + | let acts = pull_change(kind).is_some() | |
| 386 | + | || crate::stats::shown_differently(kind, &serde_json::json!({ "username": "ana" })).is_some(); | |
| 387 | + | if acts { | |
| 388 | + | assert!(g1t_contracts::subscribers::routed("SUBSCRIBER_REPOS", kind), "{kind} is not routed to repos"); | |
| 389 | + | } | |
| 390 | + | } | |
| 391 | + | for kind in ["workspace.renamed", "workspace.deleting", "workspace.restored", "workspace.deleted"] { | |
| 392 | + | assert!(g1t_contracts::subscribers::routed("SUBSCRIBER_REPOS", kind)); | |
| 393 | + | } | |
| 394 | + | } | |
| 395 | + | ||
| 396 | + | #[test] | |
| 382 | 397 | fn a_copy_is_removed_days_after_its_pull_request_settles() { | |
| 383 | 398 | let settled = 1_791_936_000_000; // 2026-10-14T00:00:00Z | |
| 384 | 399 | assert_eq!(retire_after(settled, 7), "2026-10-21T00:00:00.000Z"); |
| 61 | 61 | ||
| 62 | 62 | use serde::Serialize; | |
| 63 | 63 | use worker::{ | |
| 64 | − | Context, Env, Fetcher, MessageBatch, Method, Request, Response, Result, ScheduleContext, ScheduledEvent, | |
| 64 | + | Context, Env, Fetcher, MessageBatch, MessageExt, Method, Request, Response, Result, ScheduleContext, ScheduledEvent, | |
| 65 | 65 | event, | |
| 66 | 66 | }; | |
| 67 | 67 | ||
| ⋯ | |||
| 2188 | 2188 | // not fit a repo per pull request, so the front end reports pushes | |
| 2189 | 2189 | // itself: one event for each branch that moved. | |
| 2190 | 2190 | let stored = self.store.open(&store_key(&repo)).await?; | |
| 2191 | − | for pushed in &pushed { | |
| 2191 | + | let announced = announced_refs(&pushed, &repo.default_branch); | |
| 2192 | + | if announced.len() < pushed.len() { | |
| 2193 | + | worker::console_log!( | |
| 2194 | + | "push to {}: {} of {} refs announced", | |
| 2195 | + | repo.name, | |
| 2196 | + | announced.len(), | |
| 2197 | + | pushed.len() | |
| 2198 | + | ); | |
| 2199 | + | } | |
| 2200 | + | for pushed in announced { | |
| 2192 | 2201 | // The store can refuse one ref and accept another, so each | |
| 2193 | 2202 | // branch is checked against where it actually is. A tag the | |
| 2194 | 2203 | // store cannot read back is taken as pushed. | |
| ⋯ | |||
| 2219 | 2228 | } | |
| 2220 | 2229 | } | |
| 2221 | 2230 | ||
| 2231 | + | /// The most tags one push announces: a push of more (`git push --tags` | |
| 2232 | + | /// into a new repository, say) announces none of them, so it starts no | |
| 2233 | + | /// workflows, mirror syncs or package reads, one per tag. | |
| 2234 | + | const MAX_PUSH_TAG_EVENTS: usize = 3; | |
| 2235 | + | /// The most branches one push announces. A push of more announces only | |
| 2236 | + | /// the default branch, if it moved, which the rest of g1t reads from. | |
| 2237 | + | const MAX_PUSH_BRANCH_EVENTS: usize = 1000; | |
| 2238 | + | ||
| 2239 | + | /// The refs of a push that are announced with a `git.push` event each. | |
| 2240 | + | /// Every ref is stored whatever this says; only the events are capped. | |
| 2241 | + | fn announced_refs<'a>(pushed: &'a [git_http::Pushed], default_branch: &str) -> Vec<&'a git_http::Pushed> { | |
| 2242 | + | let (branches, tags): (Vec<&git_http::Pushed>, Vec<&git_http::Pushed>) = | |
| 2243 | + | pushed.iter().partition(|pushed| pushed.branch().is_some()); | |
| 2244 | + | let mut announced = if branches.len() > MAX_PUSH_BRANCH_EVENTS { | |
| 2245 | + | branches.into_iter().filter(|pushed| pushed.branch() == Some(default_branch)).collect() | |
| 2246 | + | } else { | |
| 2247 | + | branches | |
| 2248 | + | }; | |
| 2249 | + | if tags.len() <= MAX_PUSH_TAG_EVENTS { | |
| 2250 | + | announced.extend(tags); | |
| 2251 | + | } | |
| 2252 | + | announced | |
| 2253 | + | } | |
| 2254 | + | ||
| 2222 | 2255 | /// A push the store accepted, to be recorded once git has its answer. | |
| 2223 | 2256 | struct PushDone { | |
| 2224 | 2257 | repo: Repo, | |
| ⋯ | |||
| 2668 | 2701 | handled | |
| 2669 | 2702 | } | |
| 2670 | 2703 | ||
| 2704 | + | /// Each event is acknowledged or retried on its own, so one that fails is | |
| 2705 | + | /// tried again without the others before and after it running twice. | |
| 2671 | 2706 | async fn handle_events(batch: &MessageBatch<Event>, env: &Env, registry: &Registry, identity: &Fetcher) -> Result<()> { | |
| 2672 | 2707 | for message in batch.messages()? { | |
| 2673 | − | let event = message.body(); | |
| 2674 | − | // A pull request merged, closed or reopened: its working copy is | |
| 2675 | − | // kept or let go (forks.rs). | |
| 2676 | − | if let Some(change) = forks::pull_change(&event.kind) { | |
| 2677 | − | let Some(pull_id) = forks::pull_id_of(&event.data) else { | |
| 2678 | − | worker::console_error!("{} {} names no pull request", event.kind, event.id); | |
| 2679 | − | continue; | |
| 2680 | − | }; | |
| 2681 | − | let repos = service(env)?; | |
| 2682 | − | match change { | |
| 2683 | − | forks::PullChange::Settled => repos.pull_settled(&pull_id).await?, | |
| 2684 | − | forks::PullChange::Reopened => repos.pull_reopened(&pull_id).await?, | |
| 2708 | + | match handle_event(message.body(), env, registry, identity).await { | |
| 2709 | + | Ok(()) => message.ack(), | |
| 2710 | + | Err(error) => { | |
| 2711 | + | worker::console_error!("repos: event {} failed: {error}", message.body().id); | |
| 2712 | + | message.retry(); | |
| 2685 | 2713 | } | |
| 2686 | − | continue; | |
| 2687 | 2714 | } | |
| 2688 | − | // A workspace deleted, restored or purged: its repositories go with | |
| 2689 | − | // it, come back with it, or are purged with it (lifecycle.rs). | |
| 2690 | − | if event.kind == "workspace.deleting" { | |
| 2691 | − | match serde_json::from_value::<WorkspaceDeleting>(event.data.clone()) { | |
| 2692 | − | Ok(deleting) => service(env)?.delete_with_workspace(&deleting, &protected_workspaces(env)).await?, | |
| 2693 | − | Err(_) => worker::console_error!("workspace.deleting {} could not be read", event.id), | |
| 2694 | − | } | |
| 2695 | − | continue; | |
| 2715 | + | } | |
| 2716 | + | Ok(()) | |
| 2717 | + | } | |
| 2718 | + | ||
| 2719 | + | async fn handle_event(event: &Event, env: &Env, registry: &Registry, identity: &Fetcher) -> Result<()> { | |
| 2720 | + | // A pull request merged, closed or reopened: its working copy is | |
| 2721 | + | // kept or let go (forks.rs). | |
| 2722 | + | if let Some(change) = forks::pull_change(&event.kind) { | |
| 2723 | + | let Some(pull_id) = forks::pull_id_of(&event.data) else { | |
| 2724 | + | worker::console_error!("{} {} names no pull request", event.kind, event.id); | |
| 2725 | + | return Ok(()); | |
| 2726 | + | }; | |
| 2727 | + | let repos = service(env)?; | |
| 2728 | + | match change { | |
| 2729 | + | forks::PullChange::Settled => repos.pull_settled(&pull_id).await?, | |
| 2730 | + | forks::PullChange::Reopened => repos.pull_reopened(&pull_id).await?, | |
| 2696 | 2731 | } | |
| 2697 | − | if event.kind == "workspace.restored" { | |
| 2698 | − | match serde_json::from_value::<WorkspaceRestored>(event.data.clone()) { | |
| 2699 | − | Ok(restored) => service(env)?.restore_with_workspace(&restored).await?, | |
| 2700 | − | Err(_) => worker::console_error!("workspace.restored {} could not be read", event.id), | |
| 2701 | − | } | |
| 2702 | − | continue; | |
| 2732 | + | return Ok(()); | |
| 2733 | + | } | |
| 2734 | + | // A workspace deleted, restored or purged: its repositories go with | |
| 2735 | + | // it, come back with it, or are purged with it (lifecycle.rs). | |
| 2736 | + | if event.kind == "workspace.deleting" { | |
| 2737 | + | match serde_json::from_value::<WorkspaceDeleting>(event.data.clone()) { | |
| 2738 | + | Ok(deleting) => service(env)?.delete_with_workspace(&deleting, &protected_workspaces(env)).await?, | |
| 2739 | + | Err(_) => worker::console_error!("workspace.deleting {} could not be read", event.id), | |
| 2703 | 2740 | } | |
| 2704 | − | if event.kind == "workspace.deleted" { | |
| 2705 | − | match serde_json::from_value::<WorkspaceDeleted>(event.data.clone()) { | |
| 2706 | − | Ok(deleted) => service(env)?.purge_workspace(&deleted, &protected_workspaces(env)).await?, | |
| 2707 | − | Err(_) => worker::console_error!("workspace.deleted {} could not be read", event.id), | |
| 2708 | − | } | |
| 2709 | − | continue; | |
| 2741 | + | return Ok(()); | |
| 2742 | + | } | |
| 2743 | + | if event.kind == "workspace.restored" { | |
| 2744 | + | match serde_json::from_value::<WorkspaceRestored>(event.data.clone()) { | |
| 2745 | + | Ok(restored) => service(env)?.restore_with_workspace(&restored).await?, | |
| 2746 | + | Err(_) => worker::console_error!("workspace.restored {} could not be read", event.id), | |
| 2710 | 2747 | } | |
| 2711 | − | // An account deleted or restored: kept contributors that name it | |
| 2712 | − | // (or ghost, for a restore) are worked out again, so it shows as | |
| 2713 | − | // ghost for its 30 days, and as itself again if restored. | |
| 2714 | − | if let Some(needle) = stats::shown_differently(&event.kind, &event.data) { | |
| 2715 | − | let db = env.d1("DB")?; | |
| 2716 | − | if let Err(error) = stats::rework_naming(&db, &needle).await { | |
| 2717 | − | worker::console_error!("{} {}: contributors not marked to be counted again: {error}", event.kind, event.id); | |
| 2718 | − | } | |
| 2719 | − | continue; | |
| 2748 | + | return Ok(()); | |
| 2749 | + | } | |
| 2750 | + | if event.kind == "workspace.deleted" { | |
| 2751 | + | match serde_json::from_value::<WorkspaceDeleted>(event.data.clone()) { | |
| 2752 | + | Ok(deleted) => service(env)?.purge_workspace(&deleted, &protected_workspaces(env)).await?, | |
| 2753 | + | Err(_) => worker::console_error!("workspace.deleted {} could not be read", event.id), | |
| 2720 | 2754 | } | |
| 2721 | − | if event.kind != "workspace.renamed" { | |
| 2722 | − | continue; | |
| 2755 | + | return Ok(()); | |
| 2756 | + | } | |
| 2757 | + | // An account deleted or restored: kept contributors that name it | |
| 2758 | + | // (or ghost, for a restore) are worked out again, so it shows as | |
| 2759 | + | // ghost for its 30 days, and as itself again if restored. | |
| 2760 | + | if let Some(needle) = stats::shown_differently(&event.kind, &event.data) { | |
| 2761 | + | let db = env.d1("DB")?; | |
| 2762 | + | if let Err(error) = stats::rework_naming(&db, &needle).await { | |
| 2763 | + | worker::console_error!("{} {}: contributors not marked to be counted again: {error}", event.kind, event.id); | |
| 2723 | 2764 | } | |
| 2724 | − | let Ok(renamed) = serde_json::from_value::<WorkspaceRenamed>(event.data.clone()) else { | |
| 2725 | − | worker::console_error!("workspace.renamed {} could not be read", event.id); | |
| 2726 | − | continue; | |
| 2727 | − | }; | |
| 2728 | − | let names: HashMap<String, String> = g1t_kit::call( | |
| 2729 | − | identity, | |
| 2730 | − | "usernames", | |
| 2731 | − | &g1t_contracts::identity::UsernamesArgs { | |
| 2732 | − | ids: vec![renamed.workspace_id.clone()], | |
| 2733 | − | }, | |
| 2734 | − | ) | |
| 2765 | + | return Ok(()); | |
| 2766 | + | } | |
| 2767 | + | if event.kind != "workspace.renamed" { | |
| 2768 | + | return Ok(()); | |
| 2769 | + | } | |
| 2770 | + | let Ok(renamed) = serde_json::from_value::<WorkspaceRenamed>(event.data.clone()) else { | |
| 2771 | + | worker::console_error!("workspace.renamed {} could not be read", event.id); | |
| 2772 | + | return Ok(()); | |
| 2773 | + | }; | |
| 2774 | + | let names: HashMap<String, String> = g1t_kit::call( | |
| 2775 | + | identity, | |
| 2776 | + | "usernames", | |
| 2777 | + | &g1t_contracts::identity::UsernamesArgs { | |
| 2778 | + | ids: vec![renamed.workspace_id.clone()], | |
| 2779 | + | }, | |
| 2780 | + | ) | |
| 2781 | + | .await?; | |
| 2782 | + | let current = names | |
| 2783 | + | .get(&renamed.workspace_id) | |
| 2784 | + | .cloned() | |
| 2785 | + | .unwrap_or_else(|| renamed.to.clone()); | |
| 2786 | + | let left = registry | |
| 2787 | + | .rename_namespace(&renamed.stale_slugs(¤t), ¤t) | |
| 2735 | 2788 | .await?; | |
| 2736 | − | let current = names | |
| 2737 | − | .get(&renamed.workspace_id) | |
| 2738 | − | .cloned() | |
| 2739 | − | .unwrap_or_else(|| renamed.to.clone()); | |
| 2740 | − | let left = registry | |
| 2741 | − | .rename_namespace(&renamed.stale_slugs(¤t), ¤t) | |
| 2742 | − | .await?; | |
| 2743 | − | if left > 0 { | |
| 2744 | − | worker::console_error!( | |
| 2745 | − | "{left} repositories stayed under {} or {}: {current} already has repositories of the same names", | |
| 2746 | − | renamed.from, | |
| 2747 | − | renamed.to | |
| 2748 | − | ); | |
| 2749 | − | } | |
| 2789 | + | if left > 0 { | |
| 2790 | + | worker::console_error!( | |
| 2791 | + | "{left} repositories stayed under {} or {}: {current} already has repositories of the same names", | |
| 2792 | + | renamed.from, | |
| 2793 | + | renamed.to | |
| 2794 | + | ); | |
| 2750 | 2795 | } | |
| 2751 | 2796 | Ok(()) | |
| 2752 | 2797 | } | |
| ⋯ | |||
| 2828 | 2873 | } | |
| 2829 | 2874 | ||
| 2830 | 2875 | #[cfg(test)] | |
| 2876 | + | mod announced_refs_tests { | |
| 2877 | + | use super::*; | |
| 2878 | + | ||
| 2879 | + | fn pushed(git_ref: String) -> git_http::Pushed { | |
| 2880 | + | git_http::Pushed { git_ref, before: None, after: "abc".into() } | |
| 2881 | + | } | |
| 2882 | + | ||
| 2883 | + | fn refs(announced: Vec<&git_http::Pushed>) -> Vec<&str> { | |
| 2884 | + | announced.into_iter().map(|pushed| pushed.git_ref.as_str()).collect() | |
| 2885 | + | } | |
| 2886 | + | ||
| 2887 | + | #[test] | |
| 2888 | + | fn a_few_tags_are_announced_and_many_are_not() { | |
| 2889 | + | let few: Vec<_> = (1..=3).map(|n| pushed(format!("refs/tags/v{n}"))).chain([pushed("refs/heads/main".into())]).collect(); | |
| 2890 | + | assert_eq!(refs(announced_refs(&few, "main")), ["refs/heads/main", "refs/tags/v1", "refs/tags/v2", "refs/tags/v3"]); | |
| 2891 | + | let many: Vec<_> = (1..=10_000).map(|n| pushed(format!("refs/tags/v{n}"))).chain([pushed("refs/heads/main".into())]).collect(); | |
| 2892 | + | assert_eq!(refs(announced_refs(&many, "main")), ["refs/heads/main"]); | |
| 2893 | + | } | |
| 2894 | + | ||
| 2895 | + | #[test] | |
| 2896 | + | fn past_the_branch_cap_only_the_default_branch_is_announced() { | |
| 2897 | + | let at_cap: Vec<_> = (0..MAX_PUSH_BRANCH_EVENTS).map(|n| pushed(format!("refs/heads/b{n}"))).collect(); | |
| 2898 | + | assert_eq!(announced_refs(&at_cap, "main").len(), MAX_PUSH_BRANCH_EVENTS); | |
| 2899 | + | let over: Vec<_> = (0..=MAX_PUSH_BRANCH_EVENTS) | |
| 2900 | + | .map(|n| pushed(format!("refs/heads/b{n}"))) | |
| 2901 | + | .chain([pushed("refs/heads/main".into())]) | |
| 2902 | + | .collect(); | |
| 2903 | + | assert_eq!(refs(announced_refs(&over, "main")), ["refs/heads/main"]); | |
| 2904 | + | assert!(announced_refs(&over, "trunk").is_empty()); | |
| 2905 | + | } | |
| 2906 | + | } | |
| 2907 | + | ||
| 2908 | + | #[cfg(test)] | |
| 2831 | 2909 | mod push_to_create_tests { | |
| 2832 | 2910 | use super::*; | |
| 2833 | 2911 | ||
| 122 | 122 | // A workspace renamed moves its repositories to the new slug; one | |
| 123 | 123 | // deleted purges the repositories it left in Recently deleted. | |
| 124 | 124 | "queues": { | |
| 125 | − | "consumers": [{ "queue": "g1t-events-repos", "max_batch_size": 20, "max_batch_timeout": 1 }] | |
| 125 | + | "consumers": [{ "queue": "g1t-events-repos", "max_batch_size": 20, "max_batch_timeout": 1, "max_retries": 3, "dead_letter_queue": "g1t-events-dlq" }] | |
| 126 | 126 | }, | |
| 127 | 127 | "observability": { "enabled": true } | |
| 128 | 128 | } |
| 91 | 91 | import { BACKUP_MINUTES, backupEnv, backupPace, backupSandboxName } from "./backup"; | |
| 92 | 92 | import { type ProjectSurroundings, readableSurroundings } from "./surroundings"; | |
| 93 | 93 | import { holdCredentials, pushGrant, remotePath, revokeCredentials, runCredential } from "./credentials"; | |
| 94 | − | import { buildMentionPrompt, describeThread, handleMention, planMention } from "./mentions"; | |
| 94 | + | import { buildMentionPrompt, describeThread, handleMention, jobTokenRefusal, planMention } from "./mentions"; | |
| 95 | 95 | import { instructionsFor, repoInstructions, withBlock } from "./repo-instructions"; | |
| 96 | 96 | import { cancelTask, enqueueTask, handedOverStep, selfHostedRoute, taskEnv, taskRepo } from "./self-hosted"; | |
| 97 | 97 | import { | |
| ⋯ | |||
| 2438 | 2438 | * the actor is a member or a collaborator. | |
| 2439 | 2439 | */ | |
| 2440 | 2440 | private async refusal(actor: User, repo: RepoPath): Promise<Result<never> | null> { | |
| 2441 | + | // Agent compute is never started by a workflow job's token. | |
| 2442 | + | const byJob = jobTokenRefusal(actor); | |
| 2443 | + | if (byJob) return fail("forbidden", byJob); | |
| 2441 | 2444 | const closed = await this.closedRepo(actor, repo); | |
| 2442 | 2445 | if (closed) return closed; | |
| 2443 | 2446 | if (!(await this.workspaceAllowed(repo.namespace))) { | |
| ⋯ | |||
| 2840 | 2843 | ||
| 2841 | 2844 | async delegate(actor: User, repo: RepoPath, input: DelegateInput): Promise<Result<Delegated>> { | |
| 2842 | 2845 | // Who may put agents to work here is settled before anything is opened. | |
| 2846 | + | const byJob = jobTokenRefusal(actor); | |
| 2847 | + | if (byJob) return fail("forbidden", byJob); | |
| 2843 | 2848 | const closed = await this.closedRepo(actor, repo); | |
| 2844 | 2849 | if (closed) return closed; | |
| 2845 | 2850 | if (!actor || !(await this.repoAllows(actor, repo, "run"))) return fail("forbidden", needs("run")); | |
| 3 | 3 | ||
| 4 | 4 | import type { LifecycleJob, MentionJob, Result } from "@g1t/contracts"; | |
| 5 | 5 | ||
| 6 | − | import { type MentionPorts, buildMentionPrompt, handleMention, planMention } from "./mentions.ts"; | |
| 6 | + | import { type MentionPorts, buildMentionPrompt, handleMention, jobTokenRefusal, planMention } from "./mentions.ts"; | |
| 7 | 7 | ||
| 8 | 8 | const ana = { id: "usr_ana", username: "ana", verified: true, workspaces: [{ slug: "acme", role: "member" as const }] }; | |
| 9 | 9 | ||
| ⋯ | |||
| 162 | 162 | }; | |
| 163 | 163 | return { ports, replies, started, recorded }; | |
| 164 | 164 | } | |
| 165 | + | ||
| 166 | + | test("a workflow job's token never puts g1t to work; a person's own token can", () => { | |
| 167 | + | const job = { ...ana, token: { scopes: ["repo"], job: { run_id: "run_9", job_id: "job_1" } } }; | |
| 168 | + | assert.match(jobTokenRefusal(job) ?? "", /job's token/); | |
| 169 | + | assert.equal(jobTokenRefusal({ ...ana, token: { scopes: ["repo"] } }), null); | |
| 170 | + | assert.equal(jobTokenRefusal(ana), null); | |
| 171 | + | assert.equal(jobTokenRefusal(null), null); | |
| 172 | + | }); | |
| 11 | 11 | */ | |
| 12 | 12 | import type { Comment, LifecycleJob, MentionJob, MentionsApi, Pull, RepoPath, Result, User } from "@g1t/contracts"; | |
| 13 | 13 | ||
| 14 | + | /** | |
| 15 | + | * Why a workflow job's token (`G1T_TOKEN`) may not put g1t to work, or | |
| 16 | + | * null for anyone else. A workflow that could summon an agent on a failing | |
| 17 | + | * check would start one whose push runs the workflow again, without end. | |
| 18 | + | */ | |
| 19 | + | export function jobTokenRefusal(actor: User | null | undefined): string | null { | |
| 20 | + | return actor?.token?.job | |
| 21 | + | ? "A workflow job's token (G1T_TOKEN) cannot put g1t to work. A person, or a token of their own, can." | |
| 22 | + | : null; | |
| 23 | + | } | |
| 24 | + | ||
| 14 | 25 | /** What a mention leads to. */ | |
| 15 | 26 | export type MentionPlan = | |
| 16 | 27 | | { kind: "not_member" } |
| 67 | 67 | "triggers": { "crons": ["*/5 * * * *"] }, | |
| 68 | 68 | // Events it reacts to: a pull request ready for review, or its head moving. | |
| 69 | 69 | "queues": { | |
| 70 | − | "consumers": [{ "queue": "g1t-events-runner", "max_batch_size": 20, "max_batch_timeout": 1 }] | |
| 70 | + | "consumers": [{ "queue": "g1t-events-runner", "max_batch_size": 20, "max_batch_timeout": 1, "max_retries": 3, "dead_letter_queue": "g1t-events-dlq" }] | |
| 71 | 71 | }, | |
| 72 | 72 | "vars": { | |
| 73 | 73 | // Each workspace chooses where its agents' model spend goes: its own |
| 29 | 29 | // of the backfill. One at a time, so each stays within its caps. | |
| 30 | 30 | "producers": [{ "binding": "JOBS", "queue": "g1t-search-jobs" }], | |
| 31 | 31 | "consumers": [ | |
| 32 | − | // Every event, to keep the index current. | |
| 33 | − | { "queue": "g1t-events-search", "max_batch_size": 20, "max_batch_timeout": 2 }, | |
| 34 | − | { "queue": "g1t-search-jobs", "max_batch_size": 1, "max_batch_timeout": 1, "max_retries": 3 } | |
| 32 | + | // The events that change what is indexed, to keep it current. | |
| 33 | + | { "queue": "g1t-events-search", "max_batch_size": 20, "max_batch_timeout": 2, "max_retries": 3, "dead_letter_queue": "g1t-events-dlq" }, | |
| 34 | + | { "queue": "g1t-search-jobs", "max_batch_size": 1, "max_batch_timeout": 1, "max_retries": 3, "dead_letter_queue": "g1t-events-dlq" } | |
| 35 | 35 | ] | |
| 36 | 36 | }, | |
| 37 | 37 | "observability": { "enabled": true } |
| 36 | 36 | // failing checks (security updates move on), new repositories (their | |
| 37 | 37 | // history is scanned once) and renamed workspaces. | |
| 38 | 38 | "queues": { | |
| 39 | − | "consumers": [{ "queue": "g1t-events-security", "max_batch_size": 10, "max_batch_timeout": 5 }] | |
| 39 | + | "consumers": [{ "queue": "g1t-events-security", "max_batch_size": 10, "max_batch_timeout": 5, "max_retries": 3, "dead_letter_queue": "g1t-events-dlq" }] | |
| 40 | 40 | }, | |
| 41 | 41 | // Continues history scans a page at a time, and reads every | |
| 42 | 42 | // repository's dependencies again once a day (every 30 minutes); runs |
| 23 | 23 | ], | |
| 24 | 24 | // Every event on the bus, to deliver to the webhooks that want it. | |
| 25 | 25 | "queues": { | |
| 26 | − | "consumers": [{ "queue": "g1t-events-webhooks", "max_batch_size": 20, "max_batch_timeout": 1 }] | |
| 26 | + | "consumers": [{ "queue": "g1t-events-webhooks", "max_batch_size": 20, "max_batch_timeout": 1, "max_retries": 3, "dead_letter_queue": "g1t-events-dlq" }] | |
| 27 | 27 | }, | |
| 28 | 28 | // Retries that are due, and deliveries old enough to forget. | |
| 29 | 29 | "triggers": { "crons": ["* * * * *"] }, |
| 1 | + | -- When a mention sent g1t back to its pull request, so that no more than | |
| 2 | + | -- a day's ceiling of them do (settings.rs MAX_MENTION_REVISIONS_PER_DAY). | |
| 3 | + | ALTER TABLE agent_mentions ADD COLUMN revised_at TEXT; | |
| 4 | + | CREATE INDEX agent_mentions_revised ON agent_mentions (pull_id, revised_at) | |
| 5 | + | WHERE revised_at IS NOT NULL; |
| 490 | 490 | let repo = check!(self.repo(&a.repo, &Some(a.actor.clone())).await?); | |
| 491 | 491 | check!(writable(&repo)); | |
| 492 | 492 | check!(allowed(Some(&a.actor), &repo, Capability::Run)); | |
| 493 | + | if let Some(refused) = mentions::refuse_job_token(&a.actor) { | |
| 494 | + | return Ok(refused); | |
| 495 | + | } | |
| 493 | 496 | self.open_issue(OpenIssueArgs { | |
| 494 | 497 | actor: a.actor, | |
| 495 | 498 | repo: a.repo, |
| 23 | 23 | use worker::wasm_bindgen::JsValue; | |
| 24 | 24 | ||
| 25 | 25 | use crate::Work; | |
| 26 | + | use crate::settings::MAX_MENTION_REVISIONS_PER_DAY; | |
| 26 | 27 | use crate::lifecycle::{POLICY_ACTOR_ID, POLICY_ACTOR_NAME, made_by_g1t}; | |
| 27 | 28 | use crate::reviews::{AGENT_ID, AGENT_NAME}; | |
| 28 | 29 | use crate::rows::{NumberRow, ValueRow}; | |
| ⋯ | |||
| 388 | 389 | Ok(Some(label.to_lowercase())) | |
| 389 | 390 | } | |
| 390 | 391 | ||
| 392 | + | /// Whether `actor` mentioning `@g1t` may set it to work. g1t and other | |
| 393 | + | /// agents may not, so no agent can set another to work; neither may a | |
| 394 | + | /// workflow job's token (`G1T_TOKEN`), or a workflow that comments on a | |
| 395 | + | /// failing check would start an agent whose push runs it again, and so on | |
| 396 | + | /// without end. | |
| 397 | + | pub(crate) fn may_summon(actor: &User) -> bool { | |
| 398 | + | actor.kind != PrincipalKind::Agent | |
| 399 | + | && !actor.is_system() | |
| 400 | + | && actor.id != AGENT_ID | |
| 401 | + | && g1t_contracts::events::job_run_of(actor).is_none() | |
| 402 | + | } | |
| 403 | + | ||
| 404 | + | /// Why a workflow job's token may not put g1t to work (`may_summon`). | |
| 405 | + | pub(crate) const JOB_TOKEN_REFUSED: &str = | |
| 406 | + | "A workflow job's token (G1T_TOKEN) cannot put g1t to work. A person, or a token of their own, can."; | |
| 407 | + | ||
| 408 | + | /// A refusal for an actor acting with a workflow job's token, which may | |
| 409 | + | /// not start agents by any path: queueing an issue, assigning a plan's or | |
| 410 | + | /// handing one over. | |
| 411 | + | pub(crate) fn refuse_job_token<T>(actor: &User) -> Option<Outcome<T>> { | |
| 412 | + | g1t_contracts::events::job_run_of(actor).map(|_| Outcome::fail(FailureCode::Forbidden, JOB_TOKEN_REFUSED)) | |
| 413 | + | } | |
| 414 | + | ||
| 415 | + | /// The start of the day `mention_revision` counts back over, from `now`. | |
| 416 | + | fn day_before(now: u64) -> String { | |
| 417 | + | rfc3339(now.saturating_sub(24 * 60 * 60 * 1000)) | |
| 418 | + | } | |
| 419 | + | ||
| 391 | 420 | impl Work { | |
| 392 | 421 | /// Records a comment's mention of `@g1t`, if it makes one, for | |
| 393 | − | /// the runner to take when it hears of the comment. g1t mentioning | |
| 394 | − | /// itself is not recorded, so no agent can set another to work. | |
| 422 | + | /// the runner to take when it hears of the comment. What agents and | |
| 423 | + | /// workflow jobs say is not recorded (`may_summon`). | |
| 395 | 424 | pub(crate) async fn note_mention( | |
| 396 | 425 | &self, | |
| 397 | 426 | actor: &User, | |
| ⋯ | |||
| 400 | 429 | comment: &Comment, | |
| 401 | 430 | pull_id: Option<&str>, | |
| 402 | 431 | ) -> Result<()> { | |
| 403 | − | if actor.kind == PrincipalKind::Agent | |
| 404 | − | || actor.is_system() | |
| 405 | − | || actor.id == AGENT_ID | |
| 406 | − | || !mentions_agent(&comment.body) | |
| 407 | − | { | |
| 432 | + | if !may_summon(actor) || !mentions_agent(&comment.body) { | |
| 408 | 433 | return Ok(()); | |
| 409 | 434 | } | |
| 410 | 435 | // Whether it may set the agent to work: mentioning spends compute. | |
| ⋯ | |||
| 571 | 596 | return Ok(Outcome::fail(FailureCode::NotFound, "Pull request not found.")); | |
| 572 | 597 | }; | |
| 573 | 598 | let base = pull.base_branch(&repo.default_branch).to_owned(); | |
| 574 | − | // A person asking outranks a stop and the limit on revisions. | |
| 599 | + | // A person asking outranks a stop and the limit on revisions, but | |
| 600 | + | // not this ceiling: each revision spends compute, and nothing | |
| 601 | + | // should send g1t back to one pull request without end. | |
| 602 | + | let now = now_ms(); | |
| 603 | + | let today = self | |
| 604 | + | .db | |
| 605 | + | .prepare("SELECT count(*) AS n FROM agent_mentions WHERE pull_id = ? AND revised_at >= ?") | |
| 606 | + | .bind(&[pull.id.as_str().into(), day_before(now).into()])? | |
| 607 | + | .first::<NumberRow>(None) | |
| 608 | + | .await? | |
| 609 | + | .map_or(0, |row| row.n); | |
| 610 | + | if today >= MAX_MENTION_REVISIONS_PER_DAY { | |
| 611 | + | return Ok(Outcome::fail( | |
| 612 | + | FailureCode::Limit, | |
| 613 | + | format!( | |
| 614 | + | "g1t has been sent back to this pull request {MAX_MENTION_REVISIONS_PER_DAY} times in the last day, the most it takes. Mention it again tomorrow, or push the change yourself." | |
| 615 | + | ), | |
| 616 | + | )); | |
| 617 | + | } | |
| 575 | 618 | let was_stalled = self.is_stalled(&pull.id).await?; | |
| 576 | 619 | self.db | |
| 577 | 620 | .prepare("UPDATE pulls SET stalled = NULL WHERE id = ?") | |
| ⋯ | |||
| 587 | 630 | "g1t is already taking a step on this pull request.", | |
| 588 | 631 | )); | |
| 589 | 632 | } | |
| 633 | + | self.db | |
| 634 | + | .prepare("UPDATE agent_mentions SET revised_at = ? WHERE comment_id = ?") | |
| 635 | + | .bind(&[rfc3339(now).into(), row.comment_id.as_str().into()])? | |
| 636 | + | .run() | |
| 637 | + | .await?; | |
| 590 | 638 | let round = self | |
| 591 | 639 | .db | |
| 592 | 640 | .prepare("SELECT revisions FROM pulls WHERE id = ?") | |
| ⋯ | |||
| 787 | 835 | // the repository is public does not matter here. | |
| 788 | 836 | let target = access::RepoRef { id: &issue.repo_id, namespace: &path.namespace, private: true }; | |
| 789 | 837 | if !actor.verified | |
| 790 | − | || actor.kind == PrincipalKind::Agent | |
| 838 | + | || !may_summon(actor) | |
| 791 | 839 | || !access::can(Some(actor), target, Capability::Run) | |
| 792 | 840 | { | |
| 793 | 841 | return Ok(()); | |
| ⋯ | |||
| 819 | 867 | use super::*; | |
| 820 | 868 | ||
| 821 | 869 | #[test] | |
| 870 | + | fn only_people_and_their_own_tokens_may_summon_g1t() { | |
| 871 | + | let mut ana = User { id: "usr_ana".into(), username: "ana".into(), verified: true, ..User::default() }; | |
| 872 | + | assert!(may_summon(&ana)); | |
| 873 | + | assert!(refuse_job_token::<()>(&ana).is_none()); | |
| 874 | + | ana.token = Some(Box::new(g1t_contracts::scopes::TokenAccess::default())); | |
| 875 | + | assert!(may_summon(&ana), "a person's own token speaks for them"); | |
| 876 | + | ana.token = Some(Box::new(g1t_contracts::scopes::TokenAccess { | |
| 877 | + | job: Some(g1t_contracts::scopes::JobToken { run_id: "run_9".into(), job_id: "job_1".into(), pull_requests: true }), | |
| 878 | + | ..Default::default() | |
| 879 | + | })); | |
| 880 | + | assert!(!may_summon(&ana), "a workflow's comment would set off the workflow again"); | |
| 881 | + | assert!(matches!(refuse_job_token::<()>(&ana), Some(Outcome::Fail(_)))); | |
| 882 | + | let g1t = User { id: AGENT_ID.into(), username: AGENT_NAME.into(), ..User::default() }; | |
| 883 | + | assert!(!may_summon(&g1t)); | |
| 884 | + | } | |
| 885 | + | ||
| 886 | + | #[test] | |
| 887 | + | fn a_day_of_mention_revisions_counts_back_from_now() { | |
| 888 | + | assert_eq!(day_before(2 * 24 * 60 * 60 * 1000), rfc3339(24 * 60 * 60 * 1000)); | |
| 889 | + | assert_eq!(day_before(5), rfc3339(0)); | |
| 890 | + | } | |
| 891 | + | ||
| 892 | + | #[test] | |
| 822 | 893 | fn a_mention_is_found_whatever_its_case() { | |
| 823 | 894 | assert!(mentions_agent("@g1t take this")); | |
| 824 | 895 | assert!(mentions_agent("@G1T take this")); | |
| 379 | 379 | if let Some(refused) = may_plan(&a.actor, &repo) { | |
| 380 | 380 | return Ok(refused); | |
| 381 | 381 | } | |
| 382 | + | if a.assign | |
| 383 | + | && let Some(refused) = crate::mentions::refuse_job_token(&a.actor) | |
| 384 | + | { | |
| 385 | + | return Ok(refused); | |
| 386 | + | } | |
| 382 | 387 | let Some(row) = self | |
| 383 | 388 | .plan_row(&a.id) | |
| 384 | 389 | .await? | |
| ⋯ | |||
| 520 | 525 | Outcome::Fail(failure) => return Ok(Outcome::Fail(failure)), | |
| 521 | 526 | }; | |
| 522 | 527 | if a.queued { | |
| 528 | + | if let Some(refused) = crate::mentions::refuse_job_token(&a.actor) { | |
| 529 | + | return Ok(refused); | |
| 530 | + | } | |
| 523 | 531 | let repo = match self.repo(&a.repo, &Some(a.actor.clone())).await? { | |
| 524 | 532 | Outcome::Ok(repo) => repo, | |
| 525 | 533 | Outcome::Fail(failure) => return Ok(Outcome::Fail(failure)), | |
| 17 | 17 | ||
| 18 | 18 | const MAX_REQUIRED_APPROVALS: u32 = 6; | |
| 19 | 19 | const MAX_REVISIONS: u32 = 5; | |
| 20 | + | /// The most times in a day people's mentions may send g1t back to one pull | |
| 21 | + | /// request. They outrank `MAX_REVISIONS` and a stall, but not this. | |
| 22 | + | pub(crate) const MAX_MENTION_REVISIONS_PER_DAY: u32 = 10; | |
| 20 | 23 | ||
| 21 | 24 | #[derive(Deserialize)] | |
| 22 | 25 | struct SettingsRow { |
| 24 | 24 | { "binding": "ACTIONS", "service": "g1t-actions" } | |
| 25 | 25 | ], | |
| 26 | 26 | "queues": { | |
| 27 | − | "consumers": [{ "queue": "g1t-events-work", "max_batch_size": 100, "max_batch_timeout": 1 }] | |
| 27 | + | "consumers": [{ "queue": "g1t-events-work", "max_batch_size": 100, "max_batch_timeout": 1, "max_retries": 3, "dead_letter_queue": "g1t-events-dlq" }] | |
| 28 | 28 | }, | |
| 29 | 29 | "observability": { "enabled": true } | |
| 30 | 30 | } |