internal/control/build.go
784 lines · 29513 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 tree, _ := gitutil.ResolveTree(RepoDir(c.Cfg.Server.Root, repo.OwnerName, repo.Name), sha)
231 n, err := c.Store.CreateBuild(repo.ID, j.Name, sha, repo.DefaultBranch, string(steps), j.Image, tree, true)
232 if err != nil {
233 return c.fail(protocol.ExitFailure, "%v", err)
234 }
235 url := fmt.Sprintf("%s/%s/builds/%d", c.Cfg.Server.SiteURL, repo.Path(), n)
236 c.Store.SetCommitStatus(repo.ID, sha, "ci/"+j.Name, "pending", "triggered", url, c.User.ID)
237 return c.emit(map[string]any{"build": n, "job": j.Name, "sha": sha}, func(w io.Writer) {
238 fmt.Fprintf(w, "queued build %d (%s @ %.10s)\n", n, j.Name, sha)
239 })
240 }
241 return c.fail(protocol.ExitNotFound, "no job %q in %s", args[1], ci.ConfigPath)
242}
243
244// secretName is env-var shaped: the value lands in the build environment.
245var secretName = regexp.MustCompile(`^[A-Z_][A-Z0-9_]{0,63}$`)
246
247func runSecretSet(c *Ctx, args []string) int {
248 if len(args) != 2 {
249 return c.fail(protocol.ExitUsage, "usage: repo secret set <owner/name> <NAME> (value on stdin)")
250 }
251 if !secretName.MatchString(args[1]) {
252 return c.fail(protocol.ExitUsage, "secret names are env-var shaped: uppercase letters, digits, _")
253 }
254 repo, code := resolveRepo(c, args[0], policy.CanAdmin)
255 if code >= 0 {
256 return code
257 }
258 raw, err := io.ReadAll(io.LimitReader(c.Stdin, 64<<10))
259 if err != nil {
260 return c.fail(protocol.ExitFailure, "reading secret: %v", err)
261 }
262 value := strings.TrimRight(string(raw), "\n")
263 if value == "" {
264 return c.fail(protocol.ExitUsage, "no value on stdin (pipe it: printf %%s TOKEN | ...)")
265 }
266 if err := c.Store.SetBuildSecret(repo.ID, args[1], value); err != nil {
267 return c.fail(protocol.ExitFailure, "%v", err)
268 }
269 return c.emit(map[string]string{"secret": args[1]}, func(w io.Writer) {
270 fmt.Fprintf(w, "secret %s set on %s\n", args[1], repo.Path())
271 })
272}
273
274func runSecretRemove(c *Ctx, args []string) int {
275 if len(args) != 2 {
276 return c.fail(protocol.ExitUsage, "usage: repo secret remove <owner/name> <NAME>")
277 }
278 repo, code := resolveRepo(c, args[0], policy.CanAdmin)
279 if code >= 0 {
280 return code
281 }
282 if err := c.Store.RemoveBuildSecret(repo.ID, args[1]); err != nil {
283 if errors.Is(err, store.ErrNotFound) {
284 return c.fail(protocol.ExitNotFound, "no secret %s on %s", args[1], repo.Path())
285 }
286 return c.fail(protocol.ExitFailure, "%v", err)
287 }
288 return c.emit(map[string]string{"removed": args[1]}, func(w io.Writer) {
289 fmt.Fprintf(w, "removed %s\n", args[1])
290 })
291}
292
293func runSecretList(c *Ctx, args []string) int {
294 if len(args) != 1 {
295 return c.fail(protocol.ExitUsage, "usage: repo secret list <owner/name>")
296 }
297 repo, code := resolveRepo(c, args[0], policy.CanAdmin)
298 if code >= 0 {
299 return code
300 }
301 names, err := c.Store.ListBuildSecretNames(repo.ID)
302 if err != nil {
303 return c.fail(protocol.ExitFailure, "%v", err)
304 }
305 return c.emit(names, func(w io.Writer) {
306 for _, n := range names {
307 fmt.Fprintln(w, n)
308 }
309 })
310}
311
312func requireRunner(c *Ctx) int {
313 if c.Scope != "runner" && !c.User.IsAdmin {
314 return c.fail(protocol.ExitDenied, "runner commands need a key added with --scope runner")
315 }
316 return -1
317}
318
319// maxOrphanSkip bounds how many claimed builds runRunnerNext will find
320// unreachable and cancel in one call before giving up. Only fast-forward
321// merges are allowed here, so any branch whose target advances gets
322// rebased and force-pushed, and a stack of branches can do that repeatedly
323// in one sitting — the issue this guards saw five in an afternoon. The cap
324// is well above that, so a real backlog is never cut short, while a
325// repository whose queue is orphaned end to end still returns rather than
326// walking it forever.
327const maxOrphanSkip = 50
328
329func runRunnerNext(c *Ctx, args []string) int {
330 if code := requireRunner(c); code >= 0 {
331 return code
332 }
333 // A runner may limit itself to named repositories. The operator chooses
334 // what a given runner executes by how they start it; this is scoping the
335 // runner asks for, not an ACL the server holds over it.
336 var repoIDs []int64
337 for _, arg := range args {
338 repo, code := resolveRepo(c, arg, policy.CanRead)
339 if code >= 0 {
340 return code
341 }
342 repoIDs = append(repoIDs, repo.ID)
343 }
344 var b store.Build
345 var repo store.Repo
346 var ok bool
347 var err error
348 for attempt := 0; attempt < maxOrphanSkip; attempt++ {
349 b, ok, err = c.Store.ClaimBuild(repoIDs)
350 if err != nil {
351 return c.fail(protocol.ExitFailure, "%v", err)
352 }
353 if !ok {
354 break
355 }
356 repo, err = c.Store.RepoByID(b.RepoID)
357 if err != nil {
358 return c.fail(protocol.ExitFailure, "%v", err)
359 }
360 // Only fast-forward merges are allowed here, so a target that
361 // advances gets rebased and force-pushed, orphaning whatever was
362 // queued for the old head: the runner would clone the repo and
363 // fail at checkout with a git internal error that reads exactly
364 // like a real failure. Catch it here instead. A check that itself
365 // fails is not evidence of anything — the build runs for real and
366 // is left to fail on its own terms, never cancelled on a guess.
367 reachable, err := gitutil.Reachable(RepoDir(c.Cfg.Server.Root, repo.OwnerName, repo.Name), b.SHA)
368 if err != nil || reachable {
369 break
370 }
371 if code := cancelOrphanedBuild(c, repo, b); code >= 0 {
372 return code
373 }
374 // Cancelled, not claimed: if the cap is hit right here, the runner
375 // heartbeat below must not record this build as the one handed out.
376 b, ok = store.Build{}, false
377 }
378 // The poll itself is the runner's heartbeat: admin runners reads it.
379 c.Store.TouchRunner(c.User.ID, strings.Join(args, ","), b.ID)
380 if !ok {
381 return c.emit(map[string]any{}, func(w io.Writer) { fmt.Fprintln(w, "no pending builds") })
382 }
383 var steps []string
384 json.Unmarshal([]byte(b.Steps), &steps)
385 // Secrets ride the claim: this channel is admin-only and the values
386 // land in the build's environment, nowhere else.
387 var secrets map[string]string
388 if b.Trusted {
389 secrets, err = c.Store.BuildSecrets(b.RepoID)
390 if err != nil {
391 return c.fail(protocol.ExitFailure, "%v", err)
392 }
393 }
394 d := struct {
395 ID int64 `json:"id"`
396 Repo string `json:"repo"`
397 Number int64 `json:"number"`
398 Job string `json:"job"`
399 SHA string `json:"sha"`
400 Ref string `json:"ref"`
401 Steps []string `json:"steps"`
402 Image string `json:"image,omitempty"`
403 Secrets map[string]string `json:"secrets,omitempty"`
404 }{b.ID, repo.Path(), b.Number, b.Job, b.SHA, b.Ref, steps, b.Image, secrets}
405 return c.emit(d, func(w io.Writer) {
406 fmt.Fprintf(w, "build %d: %s %s @ %.10s\n", d.ID, d.Repo, d.Job, d.SHA)
407 })
408}
409
410func runRunnerLog(c *Ctx, args []string) int {
411 if code := requireRunner(c); code >= 0 {
412 return code
413 }
414 if len(args) != 1 {
415 return c.fail(protocol.ExitUsage, "usage: runner log <build-id> (chunk on stdin)")
416 }
417 id, err := strconv.ParseInt(args[0], 10, 64)
418 if err != nil {
419 return c.fail(protocol.ExitUsage, "bad build id %q", args[0])
420 }
421 // Stream stdin into the log in chunks so long builds appear live. An
422 // append that fails drops its chunk and the loop keeps draining: ending
423 // the session here breaks the runner's pipe, and a broken pipe is how a
424 // transient SQLITE_BUSY used to fail the build the log belonged to.
425 //
426 // The session is also how a running build is cancelled: while it is
427 // open the build's row is watched, and when the row stops saying
428 // running the session ends with ExitNotFound, which the runner reads as
429 // "stop this build". Any other end of the session is a lost stream.
430 type chunk struct {
431 data []byte
432 err error
433 }
434 chunks := make(chan chunk, 4)
435 go func() {
436 buf := make([]byte, 64<<10)
437 for {
438 n, rerr := c.Stdin.Read(buf)
439 if n > 0 {
440 chunks <- chunk{data: append([]byte(nil), buf[:n]...)}
441 }
442 if rerr != nil {
443 chunks <- chunk{err: rerr}
444 return
445 }
446 }
447 }()
448 watch := time.NewTicker(2 * time.Second)
449 defer watch.Stop()
450 dropped := 0
451 for {
452 select {
453 case ch := <-chunks:
454 if len(ch.data) > 0 {
455 if err := c.Store.AppendBuildLog(id, ch.data); err != nil {
456 dropped++
457 slog.Warn("appending build log", "build", id, "err", err)
458 }
459 }
460 if ch.err != nil {
461 if dropped > 0 {
462 slog.Warn("build log incomplete", "build", id, "dropped_chunks", dropped)
463 }
464 return c.emit(map[string]string{"log": "ok"}, func(w io.Writer) {})
465 }
466 case <-watch.C:
467 if b, err := c.Store.BuildByID(id); err == nil && b.Status != "running" {
468 return c.fail(protocol.ExitNotFound, "build %d is %s; stop", id, b.Status)
469 }
470 }
471 }
472}
473
474func runRunnerDone(c *Ctx, args []string) int {
475 if code := requireRunner(c); code >= 0 {
476 return code
477 }
478 if len(args) != 2 || (args[1] != "success" && args[1] != "failure") {
479 return c.fail(protocol.ExitUsage, "usage: runner done <build-id> success|failure")
480 }
481 id, err := strconv.ParseInt(args[0], 10, 64)
482 if err != nil {
483 return c.fail(protocol.ExitUsage, "bad build id %q", args[0])
484 }
485 b, err := c.Store.BuildByID(id)
486 if err != nil {
487 return c.fail(protocol.ExitNotFound, "no build %d", id)
488 }
489 // Cancelled underneath the runner: its report is late, not wrong.
490 // The row, the status and the log were settled by the cancel.
491 if b.Status == "cancelled" {
492 c.Store.RunnerDone(c.User.ID)
493 return c.emit(map[string]any{"build": b.Number, "status": "cancelled"}, func(w io.Writer) {
494 fmt.Fprintf(w, "build %d was cancelled\n", b.Number)
495 })
496 }
497 if err := c.Store.FinishBuild(id, args[1]); err != nil {
498 return c.fail(protocol.ExitFailure, "finishing build %d: %v", id, err)
499 }
500 c.Store.RunnerDone(c.User.ID)
501 repo, err := c.Store.RepoByID(b.RepoID)
502 if err != nil {
503 return c.fail(protocol.ExitFailure, "%v", err)
504 }
505 url := fmt.Sprintf("%s/%s/builds/%d", c.Cfg.Server.SiteURL, repo.Path(), b.Number)
506 desc := "build " + args[1]
507 if err := c.Store.SetCommitStatus(repo.ID, b.SHA, "ci/"+b.Job, args[1], desc, url, c.User.ID); err != nil {
508 return c.fail(protocol.ExitFailure, "%v", err)
509 }
510 c.Store.RecordEvent(repo.ID, c.User.ID, "build."+args[1],
511 fmt.Sprintf(`{"number":%d,"job":%q}`, b.Number, b.Job))
512 // A red build mails the repo's notify targets with the log tail — a
513 // failed scheduled job must not wait to be noticed.
514 if args[1] == "failure" {
515 if targets, err := c.Store.RepoNotifyTargets(repo); err == nil {
516 tail := ""
517 if log, err := c.Store.BuildLog(id); err == nil && len(log) > 0 {
518 if len(log) > 2000 {
519 log = log[len(log)-2000:]
520 }
521 tail = string(log)
522 }
523 notify(c, targets, notice{repo: repo, kind: "build",
524 subject: fmt.Sprintf("[%s] build %d failed: %s on %s", repo.Path(), b.Number, b.Job, b.Ref),
525 action: fmt.Sprintf("build %d failed: %s on %s", b.Number, b.Job, b.Ref),
526 body: fmt.Sprintf("job %s failed at %.10s.\n\n…%s\n\n%s\n", b.Job, b.SHA, tail, url),
527 path: fmt.Sprintf("%s/builds/%d", repo.Path(), b.Number)})
528 }
529 }
530 return c.emit(map[string]any{"build": b.Number, "status": args[1]}, func(w io.Writer) {
531 fmt.Fprintf(w, "build %d %s\n", b.Number, args[1])
532 })
533}
534
535// QueueBranchBuilds reads .gitbay/ci.yml at sha and creates one pending
536// build per push job, with a pending commit status the runner resolves.
537// A broken config surfaces as a failed "ci/config" status, not silence.
538//
539// Both paths that move a branch call this: post-receive for a push, and
540// the merge path for a merge, which updates the ref directly and so never
541// reaches a hook. old is the branch's sha before this update, the diff
542// base a job's path filters run against; a new branch has no prior
543// commit and sends old as empty or all zeros. queueJobs falls back to
544// the merge base with the default branch in that case, so a filter
545// still applies to a branch's first push — the shape most changes have,
546// since branch-then-MR is the normal workflow here.
547func QueueBranchBuilds(
548 st *store.Store, root, siteURL string,
549 repo store.Repo, userID int64, branch, old, sha string, now time.Time,
550) {
551 queueJobs(st, root, siteURL, repo, userID, branch, old, sha, now, true, branch == repo.DefaultBranch, true)
552}
553
554// QueueMRBuilds queues the push jobs for a merge request head fetched
555// from another repository, which the target holds at
556// refs/merge-requests/<n>/head, so a fork's merge request has ci/<job>
557// statuses for require-checks to gate on (#98). The head is untrusted:
558// its build runs without the target's secrets. A same-repository head is
559// the branch push's job and is not queued here; a failed one is rebuilt
560// when it lands, not when it is proposed.
561func QueueMRBuilds(
562 st *store.Store, root, siteURL string,
563 repo store.Repo, userID, n int64, sha string,
564) {
565 // No old sha, and unlike QueueBranchBuilds, no merge-base fallback
566 // either: this deliberately keeps failing open and running every
567 // job. require_checks refuses a merge when an MR head has no
568 // statuses at all (mr.go), so filtering a head down to zero jobs
569 // would make it unmergeable rather than just unfiltered (#172).
570 queueJobs(st, root, siteURL, repo, userID, mrHeadRef(n), "", sha, time.Now(), false, false, false)
571}
572
573// skipReason names why a job's path filters excluded this push, mirroring
574// the order ci.Selected checks them in: an unmatched paths list rules a
575// job out before paths-ignore is even considered.
576func skipReason(j ci.Job, changed []string) string {
577 if len(j.Paths) > 0 {
578 hit := false
579 for _, f := range changed {
580 for _, p := range j.Paths {
581 if ci.Match(p, f) {
582 hit = true
583 }
584 }
585 }
586 if !hit {
587 return "no changed file matches paths"
588 }
589 }
590 return "every changed file matched paths-ignore"
591}
592
593func queueJobs(
594 st *store.Store, root, siteURL string,
595 repo store.Repo, userID int64, ref, old, sha string, now time.Time,
596 trusted, syncSchedules, deriveMergeBase bool,
597) {
598 dir := RepoDir(root, repo.OwnerName, repo.Name)
599 raw, err := gitutil.ReadBlob(dir, sha, ci.ConfigPath, 1<<16)
600 if err != nil {
601 return // no CI config at this commit
602 }
603 jobs, err := ci.Parse(raw)
604 if err != nil {
605 st.SetCommitStatus(repo.ID, sha, "ci/config", "failure", err.Error(), "", userID)
606 return
607 }
608 // A build is a fact about a commit, not a ref: a job has no branch
609 // filter, so a commit that already passed a job on another branch has
610 // nothing left to prove when a fast-forward lands it here, and one
611 // still queued or running there will say soon enough. A failed,
612 // abandoned or cancelled build does not count; that commit runs again.
613 built, err := st.BuildsForCommit(repo.ID, sha)
614 if err != nil {
615 built = nil
616 }
617 // A job's result is a property of the tree, not the commit: a rebase
618 // onto a base that touched nothing the branch did gives every commit
619 // a new sha and the same tree, and re-running the suite over it
620 // proves nothing it did not already prove (#177). A success recorded
621 // against the tree stands for the new commit.
622 tree, _ := gitutil.ResolveTree(dir, sha)
623 // The changed-file list a job's path filters run against, computed
624 // once and only if some job actually declares one. When the diff
625 // base does not exist or the diff itself fails, filtered stays
626 // false and every job runs: a filter that cannot be evaluated must
627 // not silently skip CI.
628 //
629 // A branch's first push has no old sha, but a diff base still
630 // exists: the merge base with the default branch. Without deriving
631 // one, every job runs on every new branch, and since branch-then-MR
632 // is the normal workflow, that is the push path filters matter most
633 // for. The merge base of the default branch's tip with itself is
634 // the tip, carrying no diff — that covers the default branch's own
635 // first push on a fresh repository, and must fail open rather than
636 // read as "nothing changed".
637 filtered := false
638 var changed []string
639 for _, j := range jobs {
640 if len(j.Paths) == 0 && len(j.PathsIgnore) == 0 {
641 continue
642 }
643 diffOld := old
644 // A force-push rewrote the branch, so the old tip is not an
645 // ancestor of the new one and old..new is not "what this push
646 // changed" — it is the difference between two histories. After a
647 // rebase that is whatever the new base added, typically nothing
648 // the branch itself touched, so every path filter concludes its
649 // job is unnecessary and the branch reads as green without its
650 // suite having run (#176). The merge base is the honest base:
651 // the filter is deciding about the branch's relationship to its
652 // target, which is what the merge base expresses.
653 if ci.HasDiffBase(diffOld) && deriveMergeBase {
654 if ok, err := gitutil.IsAncestor(dir, diffOld, sha); err != nil || !ok {
655 diffOld = ""
656 }
657 }
658 if !ci.HasDiffBase(diffOld) && deriveMergeBase {
659 if base, err := gitutil.MergeBase(dir, "refs/heads/"+repo.DefaultBranch, sha); err == nil && base != sha {
660 diffOld = base
661 }
662 }
663 if ci.HasDiffBase(diffOld) {
664 if files, err := gitutil.DiffFiles(dir, diffOld, sha); err == nil {
665 changed, filtered = files, true
666 }
667 }
668 break
669 }
670 var schedules []store.Schedule
671 for _, j := range jobs {
672 // Tag jobs run on matching tag pushes only.
673 if j.Tags != "" {
674 continue
675 }
676 if b, ok := built[j.Name]; ok && (b.Status == "success" || b.Status == "pending" || b.Status == "running") {
677 continue
678 }
679 if prev, ok, _ := st.SuccessBuildForTree(repo.ID, tree, j.Name); ok && prev.SHA != sha {
680 url := fmt.Sprintf("%s/%s/builds/%d", siteURL, repo.Path(), prev.Number)
681 st.SetCommitStatus(repo.ID, sha, "ci/"+j.Name, "success",
682 fmt.Sprintf("passed in build %d as %.10s, same tree", prev.Number, prev.SHA), url, userID)
683 continue
684 }
685 // Scheduled jobs run on their cron, not on push; a default-branch
686 // push (re)registers them.
687 if j.Schedule != "" {
688 if syncSchedules {
689 schedules = append(schedules, store.Schedule{
690 RepoID: repo.ID, Job: j.Name, Cron: j.Schedule,
691 NextRun: ci.NextRun(j.Schedule, now),
692 })
693 }
694 continue
695 }
696 // A filter that excludes this push is not silence: it satisfies
697 // require_checks with a skipped status instead of leaving the
698 // commit with none at all, which the gate refuses outright (#172).
699 if filtered && !ci.Selected(j, changed) {
700 st.SetCommitStatus(repo.ID, sha, "ci/"+j.Name, "skipped", skipReason(j, changed), "", userID)
701 continue
702 }
703 steps, _ := json.Marshal(j.Steps)
704 n, err := st.CreateBuild(repo.ID, j.Name, sha, ref, string(steps), j.Image, tree, trusted)
705 if err != nil {
706 slog.Error("queueing build", "repo", repo.Path(), "job", j.Name, "err", err)
707 continue
708 }
709 url := fmt.Sprintf("%s/%s/builds/%d", siteURL, repo.Path(), n)
710 st.SetCommitStatus(repo.ID, sha, "ci/"+j.Name, "pending", "queued", url, userID)
711 }
712 if syncSchedules {
713 if err := st.SyncSchedules(repo.ID, schedules); err != nil {
714 slog.Error("syncing schedules", "repo", repo.Path(), "err", err)
715 }
716 }
717}
718
719// resolveCancelledCommitStatus sets the commit status for a build that was
720// just cancelled: if the commit already passed this job on another ref,
721// that result stands again; otherwise the context reports the
722// cancellation as an error, so the queued status left behind is never
723// pending forever.
724func resolveCancelledCommitStatus(c *Ctx, repo store.Repo, b store.Build) {
725 if prev, ok, err := c.Store.SuccessBuildFor(repo.ID, b.SHA, b.Job); err == nil && ok {
726 url := fmt.Sprintf("%s/%s/builds/%d", c.Cfg.Server.SiteURL, repo.Path(), prev.Number)
727 c.Store.SetCommitStatus(repo.ID, b.SHA, "ci/"+b.Job, "success",
728 fmt.Sprintf("passed in build %d on %s", prev.Number, prev.Ref), url, c.User.ID)
729 return
730 }
731 url := fmt.Sprintf("%s/%s/builds/%d", c.Cfg.Server.SiteURL, repo.Path(), b.Number)
732 c.Store.SetCommitStatus(repo.ID, b.SHA, "ci/"+b.Job, "error", "cancelled", url, c.User.ID)
733}
734
735// cancelOrphanedBuild withdraws a build runRunnerNext claimed and then
736// found unreachable. It leaves the same shape behind as a build cancel a
737// person runs by hand: CancelBuild's status, a log line saying why, and
738// the commit status resolved rather than left pending. Returns -1 to mean
739// "handled, keep going"; anything else is the exit code to return.
740func cancelOrphanedBuild(c *Ctx, repo store.Repo, b store.Build) int {
741 if err := c.Store.CancelBuild(b.ID); err != nil {
742 return c.fail(protocol.ExitFailure, "%v", err)
743 }
744 c.Store.AppendBuildLog(b.ID, []byte(fmt.Sprintf(
745 "cancelled: %.10s is not reachable from any ref; the sha was likely orphaned by a force-push\n", b.SHA)))
746 resolveCancelledCommitStatus(c, repo, b)
747 c.Store.RecordEvent(repo.ID, c.User.ID, "build.cancelled", fmt.Sprintf(`{"number":%d,"job":%q}`, b.Number, b.Job))
748 return -1
749}
750
751func runBuildCancel(c *Ctx, args []string) int {
752 repo, b, code := buildRef(c, args)
753 if code >= 0 {
754 return code
755 }
756 grant, err := c.Store.AccessRole(repo.ID, c.User.ID)
757 if err != nil {
758 return c.fail(protocol.ExitFailure, "%v", err)
759 }
760 if !policy.CanWrite(c.User, repo, grant) {
761 return c.fail(protocol.ExitDenied, "cancelling a build needs write access to %s", repo.Path())
762 }
763 if b.Status != "pending" && b.Status != "running" {
764 return c.fail(protocol.ExitUsage, "build %d is %s; only a queued or running build can be cancelled", b.Number, b.Status)
765 }
766 if err := c.Store.CancelBuild(b.ID); err != nil {
767 return c.fail(protocol.ExitFailure, "%v", err)
768 }
769 if b.Status == "running" {
770 c.Store.AppendBuildLog(b.ID, []byte(fmt.Sprintf("\ncancelled by %s while running; the runner stops at its next check\n", c.User.Username)))
771 } else {
772 c.Store.AppendBuildLog(b.ID, []byte(fmt.Sprintf("cancelled by %s before a runner claimed it\n", c.User.Username)))
773 }
774 // The queued status replaced whatever the commit had for this job.
775 resolveCancelledCommitStatus(c, repo, b)
776 c.Store.RecordEvent(repo.ID, c.User.ID, "build.cancelled", fmt.Sprintf(`{"number":%d,"job":%q}`, b.Number, b.Job))
777 return c.emit(map[string]any{"number": b.Number, "job": b.Job, "status": "cancelled", "was": b.Status}, func(w io.Writer) {
778 if b.Status == "running" {
779 fmt.Fprintf(w, "cancelled %s build %d (%s); the runner stops at its next check\n", repo.Path(), b.Number, b.Job)
780 return
781 }
782 fmt.Fprintf(w, "cancelled %s build %d (%s)\n", repo.Path(), b.Number, b.Job)
783 })
784}