alice.pds.demo.boltless.dev
This repository has no description
1package nixery
2
3import (
4 "bufio"
5 "context"
6 "errors"
7 "fmt"
8 "io"
9 "log/slog"
10 "path"
11 "runtime"
12 "sync"
13 "time"
14
15 "github.com/docker/docker/api/types/container"
16 "github.com/docker/docker/api/types/image"
17 "github.com/docker/docker/api/types/mount"
18 "github.com/docker/docker/api/types/network"
19 "github.com/docker/docker/client"
20 "github.com/docker/docker/pkg/stdcopy"
21 "gopkg.in/yaml.v3"
22 "tangled.org/core/api/tangled"
23 "tangled.org/core/log"
24 "tangled.org/core/spindle/config"
25 "tangled.org/core/spindle/engine"
26 "tangled.org/core/spindle/models"
27 "tangled.org/core/spindle/secrets"
28)
29
30const (
31 workspaceDir = "/tangled/workspace"
32 homeDir = "/tangled/home"
33)
34
35type cleanupFunc func(context.Context) error
36
37type Engine struct {
38 dockerMu sync.Mutex
39 docker client.APIClient
40 l *slog.Logger
41 cfg *config.Config
42
43 slotter engine.WorkflowSlotter
44
45 cleanupMu sync.Mutex
46 cleanup map[string][]cleanupFunc
47}
48
49type Step struct {
50 name string
51 kind models.StepKind
52 command string
53 environment map[string]string
54}
55
56func (s Step) Name() string {
57 return s.name
58}
59
60func (s Step) Command() string {
61 return s.command
62}
63
64func (s Step) Kind() models.StepKind {
65 return s.kind
66}
67
68// setupSteps get added to start of Steps
69type setupSteps []models.Step
70
71// addStep adds a step to the beginning of the workflow's steps.
72func (ss *setupSteps) addStep(step models.Step) {
73 *ss = append(*ss, step)
74}
75
76type addlFields struct {
77 image string
78 container string
79 mounts []mount.Mount
80}
81
82func (e *Engine) InitWorkflow(twf tangled.Pipeline_Workflow, tpl tangled.Pipeline) (*models.Workflow, error) {
83 swf := &models.Workflow{}
84 addl := addlFields{}
85
86 dwf := &struct {
87 Steps []struct {
88 Command string `yaml:"command"`
89 Name string `yaml:"name"`
90 Environment map[string]string `yaml:"environment"`
91 } `yaml:"steps"`
92 Dependencies map[string][]string `yaml:"dependencies"`
93 Environment map[string]string `yaml:"environment"`
94 }{}
95 if err := engine.DescribeManifestError(twf.Raw, dwf); err != nil {
96 return nil, err
97 }
98 if err := yaml.Unmarshal([]byte(twf.Raw), &dwf); err != nil {
99 return nil, err
100 }
101
102 for _, dstep := range dwf.Steps {
103 sstep := Step{}
104 sstep.environment = dstep.Environment
105 sstep.command = dstep.Command
106 sstep.name = dstep.Name
107 sstep.kind = models.StepKindUser
108 swf.Steps = append(swf.Steps, sstep)
109 }
110 swf.Name = twf.Name
111 swf.Environment = dwf.Environment
112 addl.image = workflowImage(dwf.Dependencies, e.cfg.NixeryPipelines.Nixery)
113
114 if sock := e.cfg.Server.DockerSocket; sock != "" {
115 addl.mounts = append(addl.mounts, mount.Mount{
116 Type: mount.TypeBind,
117 Source: sock,
118 Target: sock,
119 ReadOnly: false,
120 })
121 }
122 setup := &setupSteps{}
123
124 setup.addStep(nixConfStep())
125 setup.addStep(models.BuildCloneStep(twf, *tpl.TriggerMetadata, e.cfg.Server.Dev))
126 // this step could be empty
127 if s := dependencyStep(dwf.Dependencies); s != nil {
128 setup.addStep(*s)
129 }
130
131 // append setup steps in order to the start of workflow steps
132 swf.Steps = append(*setup, swf.Steps...)
133 swf.Data = addl
134
135 return swf, nil
136}
137
138func (e *Engine) WorkflowTimeout() time.Duration {
139 workflowTimeoutStr := e.cfg.NixeryPipelines.WorkflowTimeout
140 workflowTimeout, err := time.ParseDuration(workflowTimeoutStr)
141 if err != nil {
142 e.l.Error("failed to parse workflow timeout", "error", err, "timeout", workflowTimeoutStr)
143 workflowTimeout = 5 * time.Minute
144 }
145
146 return workflowTimeout
147}
148
149func workflowImage(deps map[string][]string, nixery string) string {
150 var dependencies string
151 for reg, ds := range deps {
152 if reg == "nixpkgs" {
153 dependencies = path.Join(ds...)
154 }
155 }
156
157 // load defaults from somewhere else
158 dependencies = path.Join(dependencies, "bash", "git", "coreutils", "nix")
159
160 if runtime.GOARCH == "arm64" {
161 dependencies = path.Join("arm64", dependencies)
162 }
163
164 return path.Join(nixery, dependencies)
165}
166
167func New(ctx context.Context, cfg *config.Config) (*Engine, error) {
168 l := log.FromContext(ctx).With("component", "spindle")
169
170 e := &Engine{
171 l: l,
172 cfg: cfg,
173 slotter: engine.NewSemaphoreSlotter(cfg.NixeryPipelines.MaxConcurrentWorkflows),
174 }
175
176 e.cleanup = make(map[string][]cleanupFunc)
177
178 return e, nil
179}
180
181func (e *Engine) ensureDocker() (client.APIClient, error) {
182 e.dockerMu.Lock()
183 defer e.dockerMu.Unlock()
184
185 if e.docker != nil {
186 return e.docker, nil
187 }
188
189 dcli, err := client.NewClientWithOpts(client.FromEnv, client.WithAPIVersionNegotiation())
190 if err != nil {
191 return nil, err
192 }
193 e.docker = dcli
194 return dcli, nil
195}
196
197func (e *Engine) AcquireWorkflowSlot(
198 ctx context.Context,
199 wid models.WorkflowId,
200 wf *models.Workflow,
201) (engine.WorkflowSlot, error) {
202 if e.slotter == nil {
203 return engine.NoopSlot{}, nil
204 }
205
206 return e.slotter.AcquireWorkflowSlot(ctx, wid, wf)
207}
208
209func (e *Engine) SetupWorkflow(ctx context.Context, wid models.WorkflowId, wf *models.Workflow, wfLogger models.WorkflowLogger) (err error) {
210 /// -------------------------INITIAL SETUP------------------------------------------
211 l := e.l.With("workflow", wid)
212 l.Info("setting up workflow")
213
214 setupStep := Step{
215 name: "Pull image from Nixery",
216 kind: models.StepKindSystem,
217 }
218 setupStepIdx := -1
219
220 wfLogger.ControlWriter(setupStepIdx, setupStep, models.StepStatusStart).Write([]byte{0})
221 defer wfLogger.ControlWriter(setupStepIdx, setupStep, models.StepStatusEnd).Write([]byte{0})
222
223 defer func() {
224 if err != nil {
225 err = fmt.Errorf("Failed to setup container:\n%w", err)
226 }
227 }()
228
229 if _, err := e.ensureDocker(); err != nil {
230 return err
231 }
232
233 /// -------------------------NETWORK CREATION---------------------------------------
234 _, err = e.docker.NetworkCreate(ctx, networkName(wid), network.CreateOptions{
235 Driver: "bridge",
236 })
237 if err != nil {
238 return err
239 }
240
241 e.registerCleanup(wid, func(ctx context.Context) error {
242 if err := e.docker.NetworkRemove(ctx, networkName(wid)); err != nil {
243 return fmt.Errorf("removing network: %w", err)
244 }
245 return nil
246 })
247
248 /// -------------------------IMAGE PULL---------------------------------------------
249 addl := wf.Data.(addlFields)
250 l.Info("pulling image", "image", addl.image)
251 fmt.Fprintf(
252 wfLogger.DataWriter(setupStepIdx, "stdout"),
253 "Pulling image: %s",
254 addl.image,
255 )
256
257 reader, err := e.docker.ImagePull(ctx, addl.image, image.PullOptions{})
258 if err != nil {
259 l.Error("pipeline image pull failed!", "error", err.Error())
260 fmt.Fprintf(wfLogger.DataWriter(setupStepIdx, "stderr"), "image pull failed: %s", err)
261 return fmt.Errorf("pulling image: %w", err)
262 }
263 defer reader.Close()
264
265 scanner := bufio.NewScanner(reader)
266 for scanner.Scan() {
267 line := scanner.Text()
268 wfLogger.DataWriter(setupStepIdx, "stdout").Write([]byte(line))
269 l.Info("image pull progress", "stdout", line)
270 }
271
272 /// -------------------------CONTAINER CREATION-------------------------------------
273 l.Info("creating container")
274 wfLogger.DataWriter(setupStepIdx, "stdout").Write([]byte("Creating container..."))
275
276 extraHosts := []string{"host.docker.internal:host-gateway"}
277 for _, h := range e.cfg.Server.DevExtraHosts {
278 extraHosts = append(extraHosts, h+":host-gateway")
279 }
280
281 resp, err := e.docker.ContainerCreate(ctx, &container.Config{
282 Image: addl.image,
283 Cmd: []string{"cat"},
284 OpenStdin: true, // so cat stays alive :3
285 Tty: false,
286 Hostname: "spindle",
287 WorkingDir: workspaceDir,
288 Labels: map[string]string{
289 "sh.tangled.pipeline/workflow_id": wid.String(),
290 },
291 // TODO(winter): investigate whether environment variables passed here
292 // get propagated to ContainerExec processes
293 }, &container.HostConfig{
294 Mounts: append([]mount.Mount{
295 {
296 Type: mount.TypeTmpfs,
297 Target: "/tmp",
298 ReadOnly: false,
299 TmpfsOptions: &mount.TmpfsOptions{
300 Mode: 0o1777, // world-writable sticky bit
301 Options: [][]string{
302 {"exec"},
303 },
304 },
305 },
306 }, addl.mounts...),
307 ReadonlyRootfs: false,
308 CapDrop: []string{"ALL"},
309 CapAdd: []string{"CAP_DAC_OVERRIDE", "CAP_CHOWN", "CAP_FOWNER", "CAP_SETUID", "CAP_SETGID"},
310 SecurityOpt: []string{"no-new-privileges"},
311 ExtraHosts: extraHosts,
312 Resources: container.Resources{
313 Memory: e.cfg.NixeryPipelines.MaxJobMemoryMB * 1024 * 1024,
314 },
315 }, nil, nil, "")
316 if err != nil {
317 fmt.Fprintf(
318 wfLogger.DataWriter(setupStepIdx, "stderr"),
319 "container creation failed: %s",
320 err,
321 )
322 return fmt.Errorf("creating container: %w", err)
323 }
324
325 e.registerCleanup(wid, func(ctx context.Context) error {
326 if err := e.docker.ContainerStop(ctx, resp.ID, container.StopOptions{}); err != nil {
327 return fmt.Errorf("stopping container: %w", err)
328 }
329
330 err := e.docker.ContainerRemove(ctx, resp.ID, container.RemoveOptions{
331 RemoveVolumes: true,
332 RemoveLinks: false,
333 Force: false,
334 })
335 if err != nil {
336 return fmt.Errorf("removing container: %w", err)
337 }
338
339 return nil
340 })
341
342 /// -------------------------CONTAINER START----------------------------------------
343 wfLogger.DataWriter(setupStepIdx, "stdout").Write([]byte("Starting container..."))
344 if err := e.docker.ContainerStart(ctx, resp.ID, container.StartOptions{}); err != nil {
345 return fmt.Errorf("starting container: %w", err)
346 }
347
348 mkExecResp, err := e.docker.ContainerExecCreate(ctx, resp.ID, container.ExecOptions{
349 Cmd: []string{"mkdir", "-p", workspaceDir, homeDir},
350 AttachStdout: true, // NOTE(winter): pretty sure this will make it so that when stdout read is done below, mkdir is done. maybe??
351 AttachStderr: true, // for good measure, backed up by docker/cli ("If -d is not set, attach to everything by default")
352 })
353 if err != nil {
354 return err
355 }
356
357 // This actually *starts* the command. Thanks, Docker!
358 execResp, err := e.docker.ContainerExecAttach(ctx, mkExecResp.ID, container.ExecAttachOptions{})
359 if err != nil {
360 return err
361 }
362 defer execResp.Close()
363
364 // This is apparently best way to wait for the command to complete.
365 _, err = io.ReadAll(execResp.Reader)
366 if err != nil {
367 return err
368 }
369
370 /// -----------------------------------FINISH---------------------------------------
371 execInspectResp, err := e.docker.ContainerExecInspect(ctx, mkExecResp.ID)
372 if err != nil {
373 return err
374 }
375
376 if execInspectResp.ExitCode != 0 {
377 return fmt.Errorf("mkdir exited with exit code %d", execInspectResp.ExitCode)
378 } else if execInspectResp.Running {
379 return errors.New("mkdir is somehow still running??")
380 }
381
382 addl.container = resp.ID
383 wf.Data = addl
384
385 return nil
386}
387
388func (e *Engine) RunStep(ctx context.Context, wid models.WorkflowId, w *models.Workflow, idx int, secrets []secrets.UnlockedSecret, wfLogger models.WorkflowLogger) error {
389 addl := w.Data.(addlFields)
390 workflowEnvs := ConstructEnvs(w.Environment)
391 // TODO(winter): should SetupWorkflow also have secret access?
392 // IMO yes, but probably worth thinking on.
393 for _, s := range secrets {
394 workflowEnvs.AddEnv(s.Key, s.Value)
395 }
396
397 step := w.Steps[idx]
398
399 select {
400 case <-ctx.Done():
401 return ctx.Err()
402 default:
403 }
404
405 envs := append(EnvVars(nil), workflowEnvs...)
406 if nixStep, ok := step.(Step); ok {
407 for k, v := range nixStep.environment {
408 envs.AddEnv(k, v)
409 }
410 }
411
412 envs.AddEnv("HOME", homeDir)
413 existingPath := "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin"
414 envs.AddEnv("PATH", fmt.Sprintf("%s/.nix-profile/bin:/nix/var/nix/profiles/default/bin:%s", homeDir, existingPath))
415 if sock := e.cfg.Server.DockerSocket; sock != "" {
416 envs.AddEnv("DOCKER_HOST", fmt.Sprintf("unix://%s", sock))
417 }
418
419 mkExecResp, err := e.docker.ContainerExecCreate(ctx, addl.container, container.ExecOptions{
420 Cmd: []string{"bash", "-c", step.Command()},
421 AttachStdout: true,
422 AttachStderr: true,
423 Env: envs,
424 })
425 if err != nil {
426 return fmt.Errorf("User step error:\ncreating exec: %w", err)
427 }
428
429 // start tailing logs in background
430 tailDone := make(chan error, 1)
431 go func() {
432 tailDone <- e.tailStep(ctx, wfLogger, mkExecResp.ID, idx)
433 }()
434
435 select {
436 case <-tailDone:
437
438 case <-ctx.Done():
439 // cleanup will be handled by DestroyWorkflow, since
440 // Docker doesn't provide an API to kill an exec run
441 // (sure, we could grab the PID and kill it ourselves,
442 // but that's wasted effort)
443 e.l.Warn("step timed out", "step", step.Name())
444
445 <-tailDone
446
447 return engine.ErrTimedOut
448 }
449
450 select {
451 case <-ctx.Done():
452 return ctx.Err()
453 default:
454 }
455
456 execInspectResp, err := e.docker.ContainerExecInspect(ctx, mkExecResp.ID)
457 if err != nil {
458 return fmt.Errorf("User step error:\n%w", err)
459 }
460
461 if execInspectResp.ExitCode != 0 {
462 inspectResp, err := e.docker.ContainerInspect(ctx, addl.container)
463 if err != nil {
464 return fmt.Errorf("User step error:\n%w", err)
465 }
466
467 e.l.Error("workflow failed!", "workflow_id", wid.String(), "exit_code", execInspectResp.ExitCode, "oom_killed", inspectResp.State.OOMKilled)
468
469 if inspectResp.State.OOMKilled {
470 return fmt.Errorf("User step error:\n%w", ErrOOMKilled)
471 }
472 return fmt.Errorf("User step error: exited with code %d", execInspectResp.ExitCode)
473 }
474
475 return nil
476}
477
478func (e *Engine) tailStep(ctx context.Context, wfLogger models.WorkflowLogger, execID string, stepIdx int) error {
479 if wfLogger == nil {
480 return nil
481 }
482
483 // This actually *starts* the command. Thanks, Docker!
484 logs, err := e.docker.ContainerExecAttach(ctx, execID, container.ExecAttachOptions{})
485 if err != nil {
486 return err
487 }
488 defer logs.Close()
489
490 _, err = stdcopy.StdCopy(
491 wfLogger.DataWriter(stepIdx, "stdout"),
492 wfLogger.DataWriter(stepIdx, "stderr"),
493 logs.Reader,
494 )
495 if err != nil && err != io.EOF && !errors.Is(err, context.DeadlineExceeded) {
496 return fmt.Errorf("failed to copy logs: %w", err)
497 }
498
499 return nil
500}
501
502func (e *Engine) DestroyWorkflow(ctx context.Context, wid models.WorkflowId) error {
503 fns := e.drainCleanups(wid)
504
505 for _, fn := range fns {
506 if err := fn(ctx); err != nil {
507 e.l.Error("failed to cleanup workflow resource", "workflowId", wid, "error", err)
508 }
509 }
510 return nil
511}
512
513func (e *Engine) registerCleanup(wid models.WorkflowId, fn cleanupFunc) {
514 e.cleanupMu.Lock()
515 defer e.cleanupMu.Unlock()
516
517 key := wid.String()
518 e.cleanup[key] = append(e.cleanup[key], fn)
519}
520
521func (e *Engine) drainCleanups(wid models.WorkflowId) []cleanupFunc {
522 e.cleanupMu.Lock()
523 key := wid.String()
524
525 fns := e.cleanup[key]
526 delete(e.cleanup, key)
527 e.cleanupMu.Unlock()
528
529 return fns
530}
531
532func networkName(wid models.WorkflowId) string {
533 return fmt.Sprintf("workflow-network-%s", wid)
534}