internal/control/build.go

e3632a550366fe23b2619edba30fc58fab19b598
gitbay/internal/control/build.go history · blame · raw

771 lines · 28700 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), j.Image, 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		Image   string            `json:"image,omitempty"`
402		Secrets map[string]string `json:"secrets,omitempty"`
403	}{b.ID, repo.Path(), b.Number, b.Job, b.SHA, b.Ref, steps, b.Image, secrets}
404	return c.emit(d, func(w io.Writer) {
405		fmt.Fprintf(w, "build %d: %s %s @ %.10s\n", d.ID, d.Repo, d.Job, d.SHA)
406	})
407}
408
409func runRunnerLog(c *Ctx, args []string) int {
410	if code := requireRunner(c); code >= 0 {
411		return code
412	}
413	if len(args) != 1 {
414		return c.fail(protocol.ExitUsage, "usage: runner log <build-id> (chunk on stdin)")
415	}
416	id, err := strconv.ParseInt(args[0], 10, 64)
417	if err != nil {
418		return c.fail(protocol.ExitUsage, "bad build id %q", args[0])
419	}
420	// Stream stdin into the log in chunks so long builds appear live. An
421	// append that fails drops its chunk and the loop keeps draining: ending
422	// the session here breaks the runner's pipe, and a broken pipe is how a
423	// transient SQLITE_BUSY used to fail the build the log belonged to.
424	//
425	// The session is also how a running build is cancelled: while it is
426	// open the build's row is watched, and when the row stops saying
427	// running the session ends with ExitNotFound, which the runner reads as
428	// "stop this build". Any other end of the session is a lost stream.
429	type chunk struct {
430		data []byte
431		err  error
432	}
433	chunks := make(chan chunk, 4)
434	go func() {
435		buf := make([]byte, 64<<10)
436		for {
437			n, rerr := c.Stdin.Read(buf)
438			if n > 0 {
439				chunks <- chunk{data: append([]byte(nil), buf[:n]...)}
440			}
441			if rerr != nil {
442				chunks <- chunk{err: rerr}
443				return
444			}
445		}
446	}()
447	watch := time.NewTicker(2 * time.Second)
448	defer watch.Stop()
449	dropped := 0
450	for {
451		select {
452		case ch := <-chunks:
453			if len(ch.data) > 0 {
454				if err := c.Store.AppendBuildLog(id, ch.data); err != nil {
455					dropped++
456					slog.Warn("appending build log", "build", id, "err", err)
457				}
458			}
459			if ch.err != nil {
460				if dropped > 0 {
461					slog.Warn("build log incomplete", "build", id, "dropped_chunks", dropped)
462				}
463				return c.emit(map[string]string{"log": "ok"}, func(w io.Writer) {})
464			}
465		case <-watch.C:
466			if b, err := c.Store.BuildByID(id); err == nil && b.Status != "running" {
467				return c.fail(protocol.ExitNotFound, "build %d is %s; stop", id, b.Status)
468			}
469		}
470	}
471}
472
473func runRunnerDone(c *Ctx, args []string) int {
474	if code := requireRunner(c); code >= 0 {
475		return code
476	}
477	if len(args) != 2 || (args[1] != "success" && args[1] != "failure") {
478		return c.fail(protocol.ExitUsage, "usage: runner done <build-id> success|failure")
479	}
480	id, err := strconv.ParseInt(args[0], 10, 64)
481	if err != nil {
482		return c.fail(protocol.ExitUsage, "bad build id %q", args[0])
483	}
484	b, err := c.Store.BuildByID(id)
485	if err != nil {
486		return c.fail(protocol.ExitNotFound, "no build %d", id)
487	}
488	// Cancelled underneath the runner: its report is late, not wrong.
489	// The row, the status and the log were settled by the cancel.
490	if b.Status == "cancelled" {
491		c.Store.RunnerDone(c.User.ID)
492		return c.emit(map[string]any{"build": b.Number, "status": "cancelled"}, func(w io.Writer) {
493			fmt.Fprintf(w, "build %d was cancelled\n", b.Number)
494		})
495	}
496	if err := c.Store.FinishBuild(id, args[1]); err != nil {
497		return c.fail(protocol.ExitFailure, "finishing build %d: %v", id, err)
498	}
499	c.Store.RunnerDone(c.User.ID)
500	repo, err := c.Store.RepoByID(b.RepoID)
501	if err != nil {
502		return c.fail(protocol.ExitFailure, "%v", err)
503	}
504	url := fmt.Sprintf("%s/%s/builds/%d", c.Cfg.Server.SiteURL, repo.Path(), b.Number)
505	desc := "build " + args[1]
506	if err := c.Store.SetCommitStatus(repo.ID, b.SHA, "ci/"+b.Job, args[1], desc, url, c.User.ID); err != nil {
507		return c.fail(protocol.ExitFailure, "%v", err)
508	}
509	c.Store.RecordEvent(repo.ID, c.User.ID, "build."+args[1],
510		fmt.Sprintf(`{"number":%d,"job":%q}`, b.Number, b.Job))
511	// A red build mails the repo's notify targets with the log tail — a
512	// failed scheduled job must not wait to be noticed.
513	if args[1] == "failure" {
514		if targets, err := c.Store.RepoNotifyTargets(repo); err == nil {
515			tail := ""
516			if log, err := c.Store.BuildLog(id); err == nil && len(log) > 0 {
517				if len(log) > 2000 {
518					log = log[len(log)-2000:]
519				}
520				tail = string(log)
521			}
522			notify(c, targets, notice{repo: repo, kind: "build",
523				subject: fmt.Sprintf("[%s] build %d failed: %s on %s", repo.Path(), b.Number, b.Job, b.Ref),
524				action:  fmt.Sprintf("build %d failed: %s on %s", b.Number, b.Job, b.Ref),
525				body:    fmt.Sprintf("job %s failed at %.10s.\n\n…%s\n\n%s\n", b.Job, b.SHA, tail, url),
526				path:    fmt.Sprintf("%s/builds/%d", repo.Path(), b.Number)})
527		}
528	}
529	return c.emit(map[string]any{"build": b.Number, "status": args[1]}, func(w io.Writer) {
530		fmt.Fprintf(w, "build %d %s\n", b.Number, args[1])
531	})
532}
533
534// QueueBranchBuilds reads .gitbay/ci.yml at sha and creates one pending
535// build per push job, with a pending commit status the runner resolves.
536// A broken config surfaces as a failed "ci/config" status, not silence.
537//
538// Both paths that move a branch call this: post-receive for a push, and
539// the merge path for a merge, which updates the ref directly and so never
540// reaches a hook. old is the branch's sha before this update, the diff
541// base a job's path filters run against; a new branch has no prior
542// commit and sends old as empty or all zeros. queueJobs falls back to
543// the merge base with the default branch in that case, so a filter
544// still applies to a branch's first push — the shape most changes have,
545// since branch-then-MR is the normal workflow here.
546func QueueBranchBuilds(
547	st *store.Store, root, siteURL string,
548	repo store.Repo, userID int64, branch, old, sha string, now time.Time,
549) {
550	queueJobs(st, root, siteURL, repo, userID, branch, old, sha, now, true, branch == repo.DefaultBranch, true)
551}
552
553// QueueMRBuilds queues the push jobs for a merge request head fetched
554// from another repository, which the target holds at
555// refs/merge-requests/<n>/head, so a fork's merge request has ci/<job>
556// statuses for require-checks to gate on (#98). The head is untrusted:
557// its build runs without the target's secrets. A same-repository head is
558// the branch push's job and is not queued here; a failed one is rebuilt
559// when it lands, not when it is proposed.
560func QueueMRBuilds(
561	st *store.Store, root, siteURL string,
562	repo store.Repo, userID, n int64, sha string,
563) {
564	// No old sha, and unlike QueueBranchBuilds, no merge-base fallback
565	// either: this deliberately keeps failing open and running every
566	// job. require_checks refuses a merge when an MR head has no
567	// statuses at all (mr.go), so filtering a head down to zero jobs
568	// would make it unmergeable rather than just unfiltered (#172).
569	queueJobs(st, root, siteURL, repo, userID, mrHeadRef(n), "", sha, time.Now(), false, false, false)
570}
571
572// skipReason names why a job's path filters excluded this push, mirroring
573// the order ci.Selected checks them in: an unmatched paths list rules a
574// job out before paths-ignore is even considered.
575func skipReason(j ci.Job, changed []string) string {
576	if len(j.Paths) > 0 {
577		hit := false
578		for _, f := range changed {
579			for _, p := range j.Paths {
580				if ci.Match(p, f) {
581					hit = true
582				}
583			}
584		}
585		if !hit {
586			return "no changed file matches paths"
587		}
588	}
589	return "every changed file matched paths-ignore"
590}
591
592func queueJobs(
593	st *store.Store, root, siteURL string,
594	repo store.Repo, userID int64, ref, old, sha string, now time.Time,
595	trusted, syncSchedules, deriveMergeBase bool,
596) {
597	dir := RepoDir(root, repo.OwnerName, repo.Name)
598	raw, err := gitutil.ReadBlob(dir, sha, ci.ConfigPath, 1<<16)
599	if err != nil {
600		return // no CI config at this commit
601	}
602	jobs, err := ci.Parse(raw)
603	if err != nil {
604		st.SetCommitStatus(repo.ID, sha, "ci/config", "failure", err.Error(), "", userID)
605		return
606	}
607	// A build is a fact about a commit, not a ref: a job has no branch
608	// filter, so a commit that already passed a job on another branch has
609	// nothing left to prove when a fast-forward lands it here, and one
610	// still queued or running there will say soon enough. A failed,
611	// abandoned or cancelled build does not count; that commit runs again.
612	built, err := st.BuildsForCommit(repo.ID, sha)
613	if err != nil {
614		built = nil
615	}
616	// The changed-file list a job's path filters run against, computed
617	// once and only if some job actually declares one. When the diff
618	// base does not exist or the diff itself fails, filtered stays
619	// false and every job runs: a filter that cannot be evaluated must
620	// not silently skip CI.
621	//
622	// A branch's first push has no old sha, but a diff base still
623	// exists: the merge base with the default branch. Without deriving
624	// one, every job runs on every new branch, and since branch-then-MR
625	// is the normal workflow, that is the push path filters matter most
626	// for. The merge base of the default branch's tip with itself is
627	// the tip, carrying no diff — that covers the default branch's own
628	// first push on a fresh repository, and must fail open rather than
629	// read as "nothing changed".
630	filtered := false
631	var changed []string
632	for _, j := range jobs {
633		if len(j.Paths) == 0 && len(j.PathsIgnore) == 0 {
634			continue
635		}
636		diffOld := old
637		// A force-push rewrote the branch, so the old tip is not an
638		// ancestor of the new one and old..new is not "what this push
639		// changed" — it is the difference between two histories. After a
640		// rebase that is whatever the new base added, typically nothing
641		// the branch itself touched, so every path filter concludes its
642		// job is unnecessary and the branch reads as green without its
643		// suite having run (#176). The merge base is the honest base:
644		// the filter is deciding about the branch's relationship to its
645		// target, which is what the merge base expresses.
646		if ci.HasDiffBase(diffOld) && deriveMergeBase {
647			if ok, err := gitutil.IsAncestor(dir, diffOld, sha); err != nil || !ok {
648				diffOld = ""
649			}
650		}
651		if !ci.HasDiffBase(diffOld) && deriveMergeBase {
652			if base, err := gitutil.MergeBase(dir, "refs/heads/"+repo.DefaultBranch, sha); err == nil && base != sha {
653				diffOld = base
654			}
655		}
656		if ci.HasDiffBase(diffOld) {
657			if files, err := gitutil.DiffFiles(dir, diffOld, sha); err == nil {
658				changed, filtered = files, true
659			}
660		}
661		break
662	}
663	var schedules []store.Schedule
664	for _, j := range jobs {
665		// Tag jobs run on matching tag pushes only.
666		if j.Tags != "" {
667			continue
668		}
669		if b, ok := built[j.Name]; ok && (b.Status == "success" || b.Status == "pending" || b.Status == "running") {
670			continue
671		}
672		// Scheduled jobs run on their cron, not on push; a default-branch
673		// push (re)registers them.
674		if j.Schedule != "" {
675			if syncSchedules {
676				schedules = append(schedules, store.Schedule{
677					RepoID: repo.ID, Job: j.Name, Cron: j.Schedule,
678					NextRun: ci.NextRun(j.Schedule, now),
679				})
680			}
681			continue
682		}
683		// A filter that excludes this push is not silence: it satisfies
684		// require_checks with a skipped status instead of leaving the
685		// commit with none at all, which the gate refuses outright (#172).
686		if filtered && !ci.Selected(j, changed) {
687			st.SetCommitStatus(repo.ID, sha, "ci/"+j.Name, "skipped", skipReason(j, changed), "", userID)
688			continue
689		}
690		steps, _ := json.Marshal(j.Steps)
691		n, err := st.CreateBuild(repo.ID, j.Name, sha, ref, string(steps), j.Image, trusted)
692		if err != nil {
693			slog.Error("queueing build", "repo", repo.Path(), "job", j.Name, "err", err)
694			continue
695		}
696		url := fmt.Sprintf("%s/%s/builds/%d", siteURL, repo.Path(), n)
697		st.SetCommitStatus(repo.ID, sha, "ci/"+j.Name, "pending", "queued", url, userID)
698	}
699	if syncSchedules {
700		if err := st.SyncSchedules(repo.ID, schedules); err != nil {
701			slog.Error("syncing schedules", "repo", repo.Path(), "err", err)
702		}
703	}
704}
705
706// resolveCancelledCommitStatus sets the commit status for a build that was
707// just cancelled: if the commit already passed this job on another ref,
708// that result stands again; otherwise the context reports the
709// cancellation as an error, so the queued status left behind is never
710// pending forever.
711func resolveCancelledCommitStatus(c *Ctx, repo store.Repo, b store.Build) {
712	if prev, ok, err := c.Store.SuccessBuildFor(repo.ID, b.SHA, b.Job); err == nil && ok {
713		url := fmt.Sprintf("%s/%s/builds/%d", c.Cfg.Server.SiteURL, repo.Path(), prev.Number)
714		c.Store.SetCommitStatus(repo.ID, b.SHA, "ci/"+b.Job, "success",
715			fmt.Sprintf("passed in build %d on %s", prev.Number, prev.Ref), url, c.User.ID)
716		return
717	}
718	url := fmt.Sprintf("%s/%s/builds/%d", c.Cfg.Server.SiteURL, repo.Path(), b.Number)
719	c.Store.SetCommitStatus(repo.ID, b.SHA, "ci/"+b.Job, "error", "cancelled", url, c.User.ID)
720}
721
722// cancelOrphanedBuild withdraws a build runRunnerNext claimed and then
723// found unreachable. It leaves the same shape behind as a build cancel a
724// person runs by hand: CancelBuild's status, a log line saying why, and
725// the commit status resolved rather than left pending. Returns -1 to mean
726// "handled, keep going"; anything else is the exit code to return.
727func cancelOrphanedBuild(c *Ctx, repo store.Repo, b store.Build) int {
728	if err := c.Store.CancelBuild(b.ID); err != nil {
729		return c.fail(protocol.ExitFailure, "%v", err)
730	}
731	c.Store.AppendBuildLog(b.ID, []byte(fmt.Sprintf(
732		"cancelled: %.10s is not reachable from any ref; the sha was likely orphaned by a force-push\n", b.SHA)))
733	resolveCancelledCommitStatus(c, repo, b)
734	c.Store.RecordEvent(repo.ID, c.User.ID, "build.cancelled", fmt.Sprintf(`{"number":%d,"job":%q}`, b.Number, b.Job))
735	return -1
736}
737
738func runBuildCancel(c *Ctx, args []string) int {
739	repo, b, code := buildRef(c, args)
740	if code >= 0 {
741		return code
742	}
743	grant, err := c.Store.AccessRole(repo.ID, c.User.ID)
744	if err != nil {
745		return c.fail(protocol.ExitFailure, "%v", err)
746	}
747	if !policy.CanWrite(c.User, repo, grant) {
748		return c.fail(protocol.ExitDenied, "cancelling a build needs write access to %s", repo.Path())
749	}
750	if b.Status != "pending" && b.Status != "running" {
751		return c.fail(protocol.ExitUsage, "build %d is %s; only a queued or running build can be cancelled", b.Number, b.Status)
752	}
753	if err := c.Store.CancelBuild(b.ID); err != nil {
754		return c.fail(protocol.ExitFailure, "%v", err)
755	}
756	if b.Status == "running" {
757		c.Store.AppendBuildLog(b.ID, []byte(fmt.Sprintf("\ncancelled by %s while running; the runner stops at its next check\n", c.User.Username)))
758	} else {
759		c.Store.AppendBuildLog(b.ID, []byte(fmt.Sprintf("cancelled by %s before a runner claimed it\n", c.User.Username)))
760	}
761	// The queued status replaced whatever the commit had for this job.
762	resolveCancelledCommitStatus(c, repo, b)
763	c.Store.RecordEvent(repo.ID, c.User.ID, "build.cancelled", fmt.Sprintf(`{"number":%d,"job":%q}`, b.Number, b.Job))
764	return c.emit(map[string]any{"number": b.Number, "job": b.Job, "status": "cancelled", "was": b.Status}, func(w io.Writer) {
765		if b.Status == "running" {
766			fmt.Fprintf(w, "cancelled %s build %d (%s); the runner stops at its next check\n", repo.Path(), b.Number, b.Job)
767			return
768		}
769		fmt.Fprintf(w, "cancelled %s build %d (%s)\n", repo.Path(), b.Number, b.Job)
770	})
771}