internal/control/build.go
756 lines · 27885 bytes
1package control
2
3import (
4 "encoding/json"
5 "errors"
6 "fmt"
7 "io"
8 "log/slog"
9 "regexp"
10 "strconv"
11 "strings"
12 "time"
13
14 "gitbay.org/gitbay/internal/ci"
15 "gitbay.org/gitbay/internal/gitutil"
16 "gitbay.org/gitbay/internal/policy"
17 "gitbay.org/gitbay/internal/protocol"
18 "gitbay.org/gitbay/internal/store"
19)
20
21func init() {
22 register(Command{Path: []string{"build", "list"},
23 Summary: "list recent builds",
24 Usage: "build list <owner/name>", ReadOnly: true, Run: runBuildList})
25 register(Command{Path: []string{"build", "show"},
26 Summary: "show one build",
27 Usage: "build show <owner/name> <n>", ReadOnly: true, Run: runBuildShow})
28 register(Command{Path: []string{"build", "log"},
29 Summary: "print a build's log",
30 Usage: "build log <owner/name> <n>", ReadOnly: true, Run: runBuildLog})
31
32 register(Command{Path: []string{"build", "jobs"},
33 Summary: "list the jobs a trigger can name",
34 Usage: "build jobs <owner/name>", ReadOnly: true, Run: runBuildJobs})
35
36 register(Command{Path: []string{"build", "cancel"},
37 Summary: "withdraw a queued build before a runner claims it",
38 Usage: "build cancel <owner/name> <n>", Run: runBuildCancel})
39 register(Command{Path: []string{"build", "trigger"},
40 Summary: "queue a job now (scheduled or not)",
41 Usage: "build trigger <owner/name> <job>", Run: runBuildTrigger})
42 // Secrets: set over stdin, listed by name only, injected into the
43 // repo's builds as environment variables. Same discipline as mirror
44 // tokens — the value never appears in argv, logs, or output.
45 register(Command{Path: []string{"repo", "secret", "set"},
46 Summary: "set a build secret",
47 Usage: "repo secret set <owner/name> <NAME> (value on stdin)",
48 ReadsStdin: true, SSHOnly: true, Run: runSecretSet})
49 register(Command{Path: []string{"repo", "secret", "remove"},
50 Summary: "remove a build secret",
51 Usage: "repo secret remove <owner/name> <NAME>", Run: runSecretRemove})
52 register(Command{Path: []string{"repo", "secret", "list"},
53 Summary: "list build secret names",
54 Usage: "repo secret list <owner/name>", ReadOnly: true, Run: runSecretList})
55
56 // Runner commands: the claim/report loop for gitbay-runner. A runner
57 // executes arbitrary repo code, so handing out jobs is the instance
58 // operator's call: a key added with --scope runner, which the
59 // dispatcher confines to these three commands and read-only git, or
60 // an admin key, which a runner host should not hold (#92).
61 register(Command{Path: []string{"runner", "next"},
62 Summary: "claim the oldest pending build (runner protocol)",
63 Usage: "runner next [<owner/name>...]", SSHOnly: true, Run: runRunnerNext})
64 register(Command{Path: []string{"runner", "log"},
65 Summary: "append a build's log from stdin",
66 Usage: "runner log <build-id>", SSHOnly: true, ReadsStdin: true, Run: runRunnerLog})
67 register(Command{Path: []string{"runner", "done"},
68 Summary: "finish a build",
69 Usage: "runner done <build-id> success|failure", SSHOnly: true, Run: runRunnerDone})
70}
71
72type BuildOut struct {
73 Number int64 `json:"number"`
74 Job string `json:"job"`
75 Status string `json:"status"`
76 SHA string `json:"sha"`
77 Ref string `json:"ref"`
78 CreatedAt string `json:"created_at"`
79 FinishedAt string `json:"finished_at,omitempty"`
80}
81
82func buildToOut(b store.Build) BuildOut {
83 return BuildOut{b.Number, b.Job, b.Status, b.SHA, b.Ref, b.CreatedAt, b.FinishedAt}
84}
85
86func buildRef(c *Ctx, args []string) (store.Repo, store.Build, int) {
87 if len(args) != 2 {
88 return store.Repo{}, store.Build{}, c.fail(protocol.ExitUsage, "expected <owner/name> <number>")
89 }
90 repo, code := resolveRepo(c, args[0], policy.CanRead)
91 if code >= 0 {
92 return repo, store.Build{}, code
93 }
94 n, err := strconv.ParseInt(args[1], 10, 64)
95 if err != nil {
96 return repo, store.Build{}, c.fail(protocol.ExitUsage, "bad build number %q", args[1])
97 }
98 b, err := c.Store.BuildByNumber(repo.ID, n)
99 if err != nil {
100 return repo, b, c.fail(protocol.ExitNotFound, "no build %d on %s", n, repo.Path())
101 }
102 return repo, b, -1
103}
104
105func runBuildList(c *Ctx, args []string) int {
106 if len(args) != 1 {
107 return c.fail(protocol.ExitUsage, "usage: build list <owner/name>")
108 }
109 repo, code := resolveRepo(c, args[0], policy.CanRead)
110 if code >= 0 {
111 return code
112 }
113 builds, err := c.Store.ListBuilds(repo.ID, 50)
114 if err != nil {
115 return c.fail(protocol.ExitFailure, "%v", err)
116 }
117 var ds []BuildOut
118 for _, b := range builds {
119 ds = append(ds, buildToOut(b))
120 }
121 return c.emit(ds, func(w io.Writer) {
122 for _, d := range ds {
123 fmt.Fprintf(w, "%d\t%s\t%s\t%.10s\t%s\n", d.Number, d.Job, d.Status, d.SHA, d.Ref)
124 }
125 })
126}
127
128func runBuildShow(c *Ctx, args []string) int {
129 _, b, code := buildRef(c, args)
130 if code >= 0 {
131 return code
132 }
133 d := buildToOut(b)
134 return c.emit(d, func(w io.Writer) {
135 fmt.Fprintf(w, "build %d\t%s\t%s\n%.10s on %s\nqueued %s", d.Number, d.Job, d.Status, d.SHA, d.Ref, d.CreatedAt)
136 if d.FinishedAt != "" {
137 fmt.Fprintf(w, ", finished %s", d.FinishedAt)
138 }
139 fmt.Fprintln(w)
140 })
141}
142
143func runBuildLog(c *Ctx, args []string) int {
144 _, b, code := buildRef(c, args)
145 if code >= 0 {
146 return code
147 }
148 log, err := c.Store.BuildLog(b.ID)
149 if err != nil {
150 return c.fail(protocol.ExitFailure, "%v", err)
151 }
152 c.Stdout.Write(log)
153 return protocol.ExitOK
154}
155
156type JobOut struct {
157 Name string `json:"name"`
158 Schedule string `json:"schedule,omitempty"`
159 Tags string `json:"tags,omitempty"`
160}
161
162// repoJobs reads the CI config on the default branch — the same file the
163// scheduler reads — and returns its jobs with the sha they came from.
164func repoJobs(c *Ctx, repo store.Repo) ([]ci.Job, string, int) {
165 dir := RepoDir(c.Cfg.Server.Root, repo.OwnerName, repo.Name)
166 sha, err := gitutil.ResolveRef(dir, "refs/heads/"+repo.DefaultBranch)
167 if err != nil {
168 return nil, "", c.fail(protocol.ExitFailure, "resolving %s: %v", repo.DefaultBranch, err)
169 }
170 raw, err := gitutil.ReadBlob(dir, sha, ci.ConfigPath, 1<<16)
171 if err != nil {
172 return nil, "", c.fail(protocol.ExitNotFound, "%s has no %s on %s", repo.Path(), ci.ConfigPath, repo.DefaultBranch)
173 }
174 jobs, err := ci.Parse(raw)
175 if err != nil {
176 return nil, "", c.failErr(err)
177 }
178 return jobs, sha, -1
179}
180
181// runBuildJobs answers "what can I trigger?". Without it only a surface
182// that can read the repository's git could offer the choice.
183func runBuildJobs(c *Ctx, args []string) int {
184 if len(args) != 1 {
185 return c.fail(protocol.ExitUsage, "usage: build jobs <owner/name>")
186 }
187 repo, code := resolveRepo(c, args[0], policy.CanRead)
188 if code >= 0 {
189 return code
190 }
191 jobs, _, code := repoJobs(c, repo)
192 if code >= 0 {
193 return code
194 }
195 out := make([]JobOut, 0, len(jobs))
196 for _, j := range jobs {
197 out = append(out, JobOut{Name: j.Name, Schedule: j.Schedule, Tags: j.Tags})
198 }
199 return c.emit(out, func(w io.Writer) {
200 for _, j := range out {
201 switch {
202 case j.Schedule != "":
203 fmt.Fprintf(w, "%s\tschedule %s\n", j.Name, j.Schedule)
204 case j.Tags != "":
205 fmt.Fprintf(w, "%s\ttags %s\n", j.Name, j.Tags)
206 default:
207 fmt.Fprintf(w, "%s\ton push\n", j.Name)
208 }
209 }
210 })
211}
212
213func runBuildTrigger(c *Ctx, args []string) int {
214 if len(args) != 2 {
215 return c.fail(protocol.ExitUsage, "usage: build trigger <owner/name> <job>")
216 }
217 repo, code := resolveRepo(c, args[0], policy.CanWrite)
218 if code >= 0 {
219 return code
220 }
221 jobs, sha, code := repoJobs(c, repo)
222 if code >= 0 {
223 return code
224 }
225 for _, j := range jobs {
226 if j.Name != args[1] {
227 continue
228 }
229 steps, _ := json.Marshal(j.Steps)
230 n, err := c.Store.CreateBuild(repo.ID, j.Name, sha, repo.DefaultBranch, string(steps), true)
231 if err != nil {
232 return c.fail(protocol.ExitFailure, "%v", err)
233 }
234 url := fmt.Sprintf("%s/%s/builds/%d", c.Cfg.Server.SiteURL, repo.Path(), n)
235 c.Store.SetCommitStatus(repo.ID, sha, "ci/"+j.Name, "pending", "triggered", url, c.User.ID)
236 return c.emit(map[string]any{"build": n, "job": j.Name, "sha": sha}, func(w io.Writer) {
237 fmt.Fprintf(w, "queued build %d (%s @ %.10s)\n", n, j.Name, sha)
238 })
239 }
240 return c.fail(protocol.ExitNotFound, "no job %q in %s", args[1], ci.ConfigPath)
241}
242
243// secretName is env-var shaped: the value lands in the build environment.
244var secretName = regexp.MustCompile(`^[A-Z_][A-Z0-9_]{0,63}$`)
245
246func runSecretSet(c *Ctx, args []string) int {
247 if len(args) != 2 {
248 return c.fail(protocol.ExitUsage, "usage: repo secret set <owner/name> <NAME> (value on stdin)")
249 }
250 if !secretName.MatchString(args[1]) {
251 return c.fail(protocol.ExitUsage, "secret names are env-var shaped: uppercase letters, digits, _")
252 }
253 repo, code := resolveRepo(c, args[0], policy.CanAdmin)
254 if code >= 0 {
255 return code
256 }
257 raw, err := io.ReadAll(io.LimitReader(c.Stdin, 64<<10))
258 if err != nil {
259 return c.fail(protocol.ExitFailure, "reading secret: %v", err)
260 }
261 value := strings.TrimRight(string(raw), "\n")
262 if value == "" {
263 return c.fail(protocol.ExitUsage, "no value on stdin (pipe it: printf %%s TOKEN | ...)")
264 }
265 if err := c.Store.SetBuildSecret(repo.ID, args[1], value); err != nil {
266 return c.fail(protocol.ExitFailure, "%v", err)
267 }
268 return c.emit(map[string]string{"secret": args[1]}, func(w io.Writer) {
269 fmt.Fprintf(w, "secret %s set on %s\n", args[1], repo.Path())
270 })
271}
272
273func runSecretRemove(c *Ctx, args []string) int {
274 if len(args) != 2 {
275 return c.fail(protocol.ExitUsage, "usage: repo secret remove <owner/name> <NAME>")
276 }
277 repo, code := resolveRepo(c, args[0], policy.CanAdmin)
278 if code >= 0 {
279 return code
280 }
281 if err := c.Store.RemoveBuildSecret(repo.ID, args[1]); err != nil {
282 if errors.Is(err, store.ErrNotFound) {
283 return c.fail(protocol.ExitNotFound, "no secret %s on %s", args[1], repo.Path())
284 }
285 return c.fail(protocol.ExitFailure, "%v", err)
286 }
287 return c.emit(map[string]string{"removed": args[1]}, func(w io.Writer) {
288 fmt.Fprintf(w, "removed %s\n", args[1])
289 })
290}
291
292func runSecretList(c *Ctx, args []string) int {
293 if len(args) != 1 {
294 return c.fail(protocol.ExitUsage, "usage: repo secret list <owner/name>")
295 }
296 repo, code := resolveRepo(c, args[0], policy.CanAdmin)
297 if code >= 0 {
298 return code
299 }
300 names, err := c.Store.ListBuildSecretNames(repo.ID)
301 if err != nil {
302 return c.fail(protocol.ExitFailure, "%v", err)
303 }
304 return c.emit(names, func(w io.Writer) {
305 for _, n := range names {
306 fmt.Fprintln(w, n)
307 }
308 })
309}
310
311func requireRunner(c *Ctx) int {
312 if c.Scope != "runner" && !c.User.IsAdmin {
313 return c.fail(protocol.ExitDenied, "runner commands need a key added with --scope runner")
314 }
315 return -1
316}
317
318// maxOrphanSkip bounds how many claimed builds runRunnerNext will find
319// unreachable and cancel in one call before giving up. Only fast-forward
320// merges are allowed here, so any branch whose target advances gets
321// rebased and force-pushed, and a stack of branches can do that repeatedly
322// in one sitting — the issue this guards saw five in an afternoon. The cap
323// is well above that, so a real backlog is never cut short, while a
324// repository whose queue is orphaned end to end still returns rather than
325// walking it forever.
326const maxOrphanSkip = 50
327
328func runRunnerNext(c *Ctx, args []string) int {
329 if code := requireRunner(c); code >= 0 {
330 return code
331 }
332 // A runner may limit itself to named repositories. The operator chooses
333 // what a given runner executes by how they start it; this is scoping the
334 // runner asks for, not an ACL the server holds over it.
335 var repoIDs []int64
336 for _, arg := range args {
337 repo, code := resolveRepo(c, arg, policy.CanRead)
338 if code >= 0 {
339 return code
340 }
341 repoIDs = append(repoIDs, repo.ID)
342 }
343 var b store.Build
344 var repo store.Repo
345 var ok bool
346 var err error
347 for attempt := 0; attempt < maxOrphanSkip; attempt++ {
348 b, ok, err = c.Store.ClaimBuild(repoIDs)
349 if err != nil {
350 return c.fail(protocol.ExitFailure, "%v", err)
351 }
352 if !ok {
353 break
354 }
355 repo, err = c.Store.RepoByID(b.RepoID)
356 if err != nil {
357 return c.fail(protocol.ExitFailure, "%v", err)
358 }
359 // Only fast-forward merges are allowed here, so a target that
360 // advances gets rebased and force-pushed, orphaning whatever was
361 // queued for the old head: the runner would clone the repo and
362 // fail at checkout with a git internal error that reads exactly
363 // like a real failure. Catch it here instead. A check that itself
364 // fails is not evidence of anything — the build runs for real and
365 // is left to fail on its own terms, never cancelled on a guess.
366 reachable, err := gitutil.Reachable(RepoDir(c.Cfg.Server.Root, repo.OwnerName, repo.Name), b.SHA)
367 if err != nil || reachable {
368 break
369 }
370 if code := cancelOrphanedBuild(c, repo, b); code >= 0 {
371 return code
372 }
373 // Cancelled, not claimed: if the cap is hit right here, the runner
374 // heartbeat below must not record this build as the one handed out.
375 b, ok = store.Build{}, false
376 }
377 // The poll itself is the runner's heartbeat: admin runners reads it.
378 c.Store.TouchRunner(c.User.ID, strings.Join(args, ","), b.ID)
379 if !ok {
380 return c.emit(map[string]any{}, func(w io.Writer) { fmt.Fprintln(w, "no pending builds") })
381 }
382 var steps []string
383 json.Unmarshal([]byte(b.Steps), &steps)
384 // Secrets ride the claim: this channel is admin-only and the values
385 // land in the build's environment, nowhere else.
386 var secrets map[string]string
387 if b.Trusted {
388 secrets, err = c.Store.BuildSecrets(b.RepoID)
389 if err != nil {
390 return c.fail(protocol.ExitFailure, "%v", err)
391 }
392 }
393 d := struct {
394 ID int64 `json:"id"`
395 Repo string `json:"repo"`
396 Number int64 `json:"number"`
397 Job string `json:"job"`
398 SHA string `json:"sha"`
399 Ref string `json:"ref"`
400 Steps []string `json:"steps"`
401 Secrets map[string]string `json:"secrets,omitempty"`
402 }{b.ID, repo.Path(), b.Number, b.Job, b.SHA, b.Ref, steps, secrets}
403 return c.emit(d, func(w io.Writer) {
404 fmt.Fprintf(w, "build %d: %s %s @ %.10s\n", d.ID, d.Repo, d.Job, d.SHA)
405 })
406}
407
408func runRunnerLog(c *Ctx, args []string) int {
409 if code := requireRunner(c); code >= 0 {
410 return code
411 }
412 if len(args) != 1 {
413 return c.fail(protocol.ExitUsage, "usage: runner log <build-id> (chunk on stdin)")
414 }
415 id, err := strconv.ParseInt(args[0], 10, 64)
416 if err != nil {
417 return c.fail(protocol.ExitUsage, "bad build id %q", args[0])
418 }
419 // Stream stdin into the log in chunks so long builds appear live. An
420 // append that fails drops its chunk and the loop keeps draining: ending
421 // the session here breaks the runner's pipe, and a broken pipe is how a
422 // transient SQLITE_BUSY used to fail the build the log belonged to.
423 //
424 // The session is also how a running build is cancelled: while it is
425 // open the build's row is watched, and when the row stops saying
426 // running the session ends with ExitNotFound, which the runner reads as
427 // "stop this build". Any other end of the session is a lost stream.
428 type chunk struct {
429 data []byte
430 err error
431 }
432 chunks := make(chan chunk, 4)
433 go func() {
434 buf := make([]byte, 64<<10)
435 for {
436 n, rerr := c.Stdin.Read(buf)
437 if n > 0 {
438 chunks <- chunk{data: append([]byte(nil), buf[:n]...)}
439 }
440 if rerr != nil {
441 chunks <- chunk{err: rerr}
442 return
443 }
444 }
445 }()
446 watch := time.NewTicker(2 * time.Second)
447 defer watch.Stop()
448 dropped := 0
449 for {
450 select {
451 case ch := <-chunks:
452 if len(ch.data) > 0 {
453 if err := c.Store.AppendBuildLog(id, ch.data); err != nil {
454 dropped++
455 slog.Warn("appending build log", "build", id, "err", err)
456 }
457 }
458 if ch.err != nil {
459 if dropped > 0 {
460 slog.Warn("build log incomplete", "build", id, "dropped_chunks", dropped)
461 }
462 return c.emit(map[string]string{"log": "ok"}, func(w io.Writer) {})
463 }
464 case <-watch.C:
465 if b, err := c.Store.BuildByID(id); err == nil && b.Status != "running" {
466 return c.fail(protocol.ExitNotFound, "build %d is %s; stop", id, b.Status)
467 }
468 }
469 }
470}
471
472func runRunnerDone(c *Ctx, args []string) int {
473 if code := requireRunner(c); code >= 0 {
474 return code
475 }
476 if len(args) != 2 || (args[1] != "success" && args[1] != "failure") {
477 return c.fail(protocol.ExitUsage, "usage: runner done <build-id> success|failure")
478 }
479 id, err := strconv.ParseInt(args[0], 10, 64)
480 if err != nil {
481 return c.fail(protocol.ExitUsage, "bad build id %q", args[0])
482 }
483 b, err := c.Store.BuildByID(id)
484 if err != nil {
485 return c.fail(protocol.ExitNotFound, "no build %d", id)
486 }
487 // Cancelled underneath the runner: its report is late, not wrong.
488 // The row, the status and the log were settled by the cancel.
489 if b.Status == "cancelled" {
490 c.Store.RunnerDone(c.User.ID)
491 return c.emit(map[string]any{"build": b.Number, "status": "cancelled"}, func(w io.Writer) {
492 fmt.Fprintf(w, "build %d was cancelled\n", b.Number)
493 })
494 }
495 if err := c.Store.FinishBuild(id, args[1]); err != nil {
496 return c.fail(protocol.ExitFailure, "finishing build %d: %v", id, err)
497 }
498 c.Store.RunnerDone(c.User.ID)
499 repo, err := c.Store.RepoByID(b.RepoID)
500 if err != nil {
501 return c.fail(protocol.ExitFailure, "%v", err)
502 }
503 url := fmt.Sprintf("%s/%s/builds/%d", c.Cfg.Server.SiteURL, repo.Path(), b.Number)
504 desc := "build " + args[1]
505 if err := c.Store.SetCommitStatus(repo.ID, b.SHA, "ci/"+b.Job, args[1], desc, url, c.User.ID); err != nil {
506 return c.fail(protocol.ExitFailure, "%v", err)
507 }
508 c.Store.RecordEvent(repo.ID, c.User.ID, "build."+args[1],
509 fmt.Sprintf(`{"number":%d,"job":%q}`, b.Number, b.Job))
510 // A red build mails the repo's notify targets with the log tail — a
511 // failed scheduled job must not wait to be noticed.
512 if args[1] == "failure" {
513 if targets, err := c.Store.RepoNotifyTargets(repo); err == nil {
514 tail := ""
515 if log, err := c.Store.BuildLog(id); err == nil && len(log) > 0 {
516 if len(log) > 2000 {
517 log = log[len(log)-2000:]
518 }
519 tail = string(log)
520 }
521 notify(c, targets, notice{repo: repo, kind: "build",
522 subject: fmt.Sprintf("[%s] build %d failed: %s on %s", repo.Path(), b.Number, b.Job, b.Ref),
523 action: fmt.Sprintf("build %d failed: %s on %s", b.Number, b.Job, b.Ref),
524 body: fmt.Sprintf("job %s failed at %.10s.\n\n…%s\n\n%s\n", b.Job, b.SHA, tail, url),
525 path: fmt.Sprintf("%s/builds/%d", repo.Path(), b.Number)})
526 }
527 }
528 return c.emit(map[string]any{"build": b.Number, "status": args[1]}, func(w io.Writer) {
529 fmt.Fprintf(w, "build %d %s\n", b.Number, args[1])
530 })
531}
532
533// QueueBranchBuilds reads .gitbay/ci.yml at sha and creates one pending
534// build per push job, with a pending commit status the runner resolves.
535// A broken config surfaces as a failed "ci/config" status, not silence.
536//
537// Both paths that move a branch call this: post-receive for a push, and
538// the merge path for a merge, which updates the ref directly and so never
539// reaches a hook. old is the branch's sha before this update, the diff
540// base a job's path filters run against; a new branch has no prior
541// commit and sends old as empty or all zeros. queueJobs falls back to
542// the merge base with the default branch in that case, so a filter
543// still applies to a branch's first push — the shape most changes have,
544// since branch-then-MR is the normal workflow here.
545func QueueBranchBuilds(
546 st *store.Store, root, siteURL string,
547 repo store.Repo, userID int64, branch, old, sha string, now time.Time,
548) {
549 queueJobs(st, root, siteURL, repo, userID, branch, old, sha, now, true, branch == repo.DefaultBranch, true)
550}
551
552// QueueMRBuilds queues the push jobs for a merge request head fetched
553// from another repository, which the target holds at
554// refs/merge-requests/<n>/head, so a fork's merge request has ci/<job>
555// statuses for require-checks to gate on (#98). The head is untrusted:
556// its build runs without the target's secrets. A same-repository head is
557// the branch push's job and is not queued here; a failed one is rebuilt
558// when it lands, not when it is proposed.
559func QueueMRBuilds(
560 st *store.Store, root, siteURL string,
561 repo store.Repo, userID, n int64, sha string,
562) {
563 // No old sha, and unlike QueueBranchBuilds, no merge-base fallback
564 // either: this deliberately keeps failing open and running every
565 // job. require_checks refuses a merge when an MR head has no
566 // statuses at all (mr.go), so filtering a head down to zero jobs
567 // would make it unmergeable rather than just unfiltered (#172).
568 queueJobs(st, root, siteURL, repo, userID, mrHeadRef(n), "", sha, time.Now(), false, false, false)
569}
570
571// skipReason names why a job's path filters excluded this push, mirroring
572// the order ci.Selected checks them in: an unmatched paths list rules a
573// job out before paths-ignore is even considered.
574func skipReason(j ci.Job, changed []string) string {
575 if len(j.Paths) > 0 {
576 hit := false
577 for _, f := range changed {
578 for _, p := range j.Paths {
579 if ci.Match(p, f) {
580 hit = true
581 }
582 }
583 }
584 if !hit {
585 return "no changed file matches paths"
586 }
587 }
588 return "every changed file matched paths-ignore"
589}
590
591func queueJobs(
592 st *store.Store, root, siteURL string,
593 repo store.Repo, userID int64, ref, old, sha string, now time.Time,
594 trusted, syncSchedules, deriveMergeBase bool,
595) {
596 dir := RepoDir(root, repo.OwnerName, repo.Name)
597 raw, err := gitutil.ReadBlob(dir, sha, ci.ConfigPath, 1<<16)
598 if err != nil {
599 return // no CI config at this commit
600 }
601 jobs, err := ci.Parse(raw)
602 if err != nil {
603 st.SetCommitStatus(repo.ID, sha, "ci/config", "failure", err.Error(), "", userID)
604 return
605 }
606 // A build is a fact about a commit, not a ref: a job has no branch
607 // filter, so a commit that already passed a job on another branch has
608 // nothing left to prove when a fast-forward lands it here, and one
609 // still queued or running there will say soon enough. A failed,
610 // abandoned or cancelled build does not count; that commit runs again.
611 built, err := st.BuildsForCommit(repo.ID, sha)
612 if err != nil {
613 built = nil
614 }
615 // The changed-file list a job's path filters run against, computed
616 // once and only if some job actually declares one. When the diff
617 // base does not exist or the diff itself fails, filtered stays
618 // false and every job runs: a filter that cannot be evaluated must
619 // not silently skip CI.
620 //
621 // A branch's first push has no old sha, but a diff base still
622 // exists: the merge base with the default branch. Without deriving
623 // one, every job runs on every new branch, and since branch-then-MR
624 // is the normal workflow, that is the push path filters matter most
625 // for. The merge base of the default branch's tip with itself is
626 // the tip, carrying no diff — that covers the default branch's own
627 // first push on a fresh repository, and must fail open rather than
628 // read as "nothing changed".
629 filtered := false
630 var changed []string
631 for _, j := range jobs {
632 if len(j.Paths) == 0 && len(j.PathsIgnore) == 0 {
633 continue
634 }
635 diffOld := old
636 if !ci.HasDiffBase(diffOld) && deriveMergeBase {
637 if base, err := gitutil.MergeBase(dir, "refs/heads/"+repo.DefaultBranch, sha); err == nil && base != sha {
638 diffOld = base
639 }
640 }
641 if ci.HasDiffBase(diffOld) {
642 if files, err := gitutil.DiffFiles(dir, diffOld, sha); err == nil {
643 changed, filtered = files, true
644 }
645 }
646 break
647 }
648 var schedules []store.Schedule
649 for _, j := range jobs {
650 // Tag jobs run on matching tag pushes only.
651 if j.Tags != "" {
652 continue
653 }
654 if b, ok := built[j.Name]; ok && (b.Status == "success" || b.Status == "pending" || b.Status == "running") {
655 continue
656 }
657 // Scheduled jobs run on their cron, not on push; a default-branch
658 // push (re)registers them.
659 if j.Schedule != "" {
660 if syncSchedules {
661 schedules = append(schedules, store.Schedule{
662 RepoID: repo.ID, Job: j.Name, Cron: j.Schedule,
663 NextRun: ci.NextRun(j.Schedule, now),
664 })
665 }
666 continue
667 }
668 // A filter that excludes this push is not silence: it satisfies
669 // require_checks with a skipped status instead of leaving the
670 // commit with none at all, which the gate refuses outright (#172).
671 if filtered && !ci.Selected(j, changed) {
672 st.SetCommitStatus(repo.ID, sha, "ci/"+j.Name, "skipped", skipReason(j, changed), "", userID)
673 continue
674 }
675 steps, _ := json.Marshal(j.Steps)
676 n, err := st.CreateBuild(repo.ID, j.Name, sha, ref, string(steps), trusted)
677 if err != nil {
678 slog.Error("queueing build", "repo", repo.Path(), "job", j.Name, "err", err)
679 continue
680 }
681 url := fmt.Sprintf("%s/%s/builds/%d", siteURL, repo.Path(), n)
682 st.SetCommitStatus(repo.ID, sha, "ci/"+j.Name, "pending", "queued", url, userID)
683 }
684 if syncSchedules {
685 if err := st.SyncSchedules(repo.ID, schedules); err != nil {
686 slog.Error("syncing schedules", "repo", repo.Path(), "err", err)
687 }
688 }
689}
690
691// resolveCancelledCommitStatus sets the commit status for a build that was
692// just cancelled: if the commit already passed this job on another ref,
693// that result stands again; otherwise the context reports the
694// cancellation as an error, so the queued status left behind is never
695// pending forever.
696func resolveCancelledCommitStatus(c *Ctx, repo store.Repo, b store.Build) {
697 if prev, ok, err := c.Store.SuccessBuildFor(repo.ID, b.SHA, b.Job); err == nil && ok {
698 url := fmt.Sprintf("%s/%s/builds/%d", c.Cfg.Server.SiteURL, repo.Path(), prev.Number)
699 c.Store.SetCommitStatus(repo.ID, b.SHA, "ci/"+b.Job, "success",
700 fmt.Sprintf("passed in build %d on %s", prev.Number, prev.Ref), url, c.User.ID)
701 return
702 }
703 url := fmt.Sprintf("%s/%s/builds/%d", c.Cfg.Server.SiteURL, repo.Path(), b.Number)
704 c.Store.SetCommitStatus(repo.ID, b.SHA, "ci/"+b.Job, "error", "cancelled", url, c.User.ID)
705}
706
707// cancelOrphanedBuild withdraws a build runRunnerNext claimed and then
708// found unreachable. It leaves the same shape behind as a build cancel a
709// person runs by hand: CancelBuild's status, a log line saying why, and
710// the commit status resolved rather than left pending. Returns -1 to mean
711// "handled, keep going"; anything else is the exit code to return.
712func cancelOrphanedBuild(c *Ctx, repo store.Repo, b store.Build) int {
713 if err := c.Store.CancelBuild(b.ID); err != nil {
714 return c.fail(protocol.ExitFailure, "%v", err)
715 }
716 c.Store.AppendBuildLog(b.ID, []byte(fmt.Sprintf(
717 "cancelled: %.10s is not reachable from any ref; the sha was likely orphaned by a force-push\n", b.SHA)))
718 resolveCancelledCommitStatus(c, repo, b)
719 c.Store.RecordEvent(repo.ID, c.User.ID, "build.cancelled", fmt.Sprintf(`{"number":%d,"job":%q}`, b.Number, b.Job))
720 return -1
721}
722
723func runBuildCancel(c *Ctx, args []string) int {
724 repo, b, code := buildRef(c, args)
725 if code >= 0 {
726 return code
727 }
728 grant, err := c.Store.AccessRole(repo.ID, c.User.ID)
729 if err != nil {
730 return c.fail(protocol.ExitFailure, "%v", err)
731 }
732 if !policy.CanWrite(c.User, repo, grant) {
733 return c.fail(protocol.ExitDenied, "cancelling a build needs write access to %s", repo.Path())
734 }
735 if b.Status != "pending" && b.Status != "running" {
736 return c.fail(protocol.ExitUsage, "build %d is %s; only a queued or running build can be cancelled", b.Number, b.Status)
737 }
738 if err := c.Store.CancelBuild(b.ID); err != nil {
739 return c.fail(protocol.ExitFailure, "%v", err)
740 }
741 if b.Status == "running" {
742 c.Store.AppendBuildLog(b.ID, []byte(fmt.Sprintf("\ncancelled by %s while running; the runner stops at its next check\n", c.User.Username)))
743 } else {
744 c.Store.AppendBuildLog(b.ID, []byte(fmt.Sprintf("cancelled by %s before a runner claimed it\n", c.User.Username)))
745 }
746 // The queued status replaced whatever the commit had for this job.
747 resolveCancelledCommitStatus(c, repo, b)
748 c.Store.RecordEvent(repo.ID, c.User.ID, "build.cancelled", fmt.Sprintf(`{"number":%d,"job":%q}`, b.Number, b.Job))
749 return c.emit(map[string]any{"number": b.Number, "job": b.Job, "status": "cancelled", "was": b.Status}, func(w io.Writer) {
750 if b.Status == "running" {
751 fmt.Fprintf(w, "cancelled %s build %d (%s); the runner stops at its next check\n", repo.Path(), b.Number, b.Job)
752 return
753 }
754 fmt.Fprintf(w, "cancelled %s build %d (%s)\n", repo.Path(), b.Number, b.Job)
755 })
756}