kot.pink
1package nixery
2
3import (
4 "context"
5 "errors"
6 "fmt"
7 "io"
8 "log/slog"
9 "os"
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.sh/tangled.sh/core/api/tangled"
23 "tangled.sh/tangled.sh/core/log"
24 "tangled.sh/tangled.sh/core/spindle/config"
25 "tangled.sh/tangled.sh/core/spindle/engine"
26 "tangled.sh/tangled.sh/core/spindle/models"
27 "tangled.sh/tangled.sh/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 docker client.APIClient
39 l *slog.Logger
40 cfg *config.Config
41
42 cleanupMu sync.Mutex
43 cleanup map[string][]cleanupFunc
44}
45
46type Step struct {
47 name string
48 kind models.StepKind
49 command string
50 environment map[string]string
51}
52
53func (s Step) Name() string {
54 return s.name
55}
56
57func (s Step) Command() string {
58 return s.command
59}
60
61func (s Step) Kind() models.StepKind {
62 return s.kind
63}
64
65// setupSteps get added to start of Steps
66type setupSteps []models.Step
67
68// addStep adds a step to the beginning of the workflow's steps.
69func (ss *setupSteps) addStep(step models.Step) {
70 *ss = append(*ss, step)
71}
72
73type addlFields struct {
74 image string
75 container string
76 env map[string]string
77}
78
79func (e *Engine) InitWorkflow(twf tangled.Pipeline_Workflow, tpl tangled.Pipeline) (*models.Workflow, error) {
80 swf := &models.Workflow{}
81 addl := addlFields{}
82
83 dwf := &struct {
84 Steps []struct {
85 Command string `yaml:"command"`
86 Name string `yaml:"name"`
87 Environment map[string]string `yaml:"environment"`
88 } `yaml:"steps"`
89 Dependencies map[string][]string `yaml:"dependencies"`
90 Environment map[string]string `yaml:"environment"`
91 }{}
92 err := yaml.Unmarshal([]byte(twf.Raw), &dwf)
93 if err != nil {
94 return nil, err
95 }
96
97 for _, dstep := range dwf.Steps {
98 sstep := Step{}
99 sstep.environment = dstep.Environment
100 sstep.command = dstep.Command
101 sstep.name = dstep.Name
102 sstep.kind = models.StepKindUser
103 swf.Steps = append(swf.Steps, sstep)
104 }
105 swf.Name = twf.Name
106 addl.env = dwf.Environment
107 addl.image = workflowImage(dwf.Dependencies, e.cfg.NixeryPipelines.Nixery)
108
109 setup := &setupSteps{}
110
111 setup.addStep(nixConfStep())
112 setup.addStep(cloneStep(twf, *tpl.TriggerMetadata, e.cfg.Server.Dev))
113 // this step could be empty
114 if s := dependencyStep(dwf.Dependencies); s != nil {
115 setup.addStep(*s)
116 }
117
118 // append setup steps in order to the start of workflow steps
119 swf.Steps = append(*setup, swf.Steps...)
120 swf.Data = addl
121
122 return swf, nil
123}
124
125func (e *Engine) WorkflowTimeout() time.Duration {
126 workflowTimeoutStr := e.cfg.NixeryPipelines.WorkflowTimeout
127 workflowTimeout, err := time.ParseDuration(workflowTimeoutStr)
128 if err != nil {
129 e.l.Error("failed to parse workflow timeout", "error", err, "timeout", workflowTimeoutStr)
130 workflowTimeout = 5 * time.Minute
131 }
132
133 return workflowTimeout
134}
135
136func workflowImage(deps map[string][]string, nixery string) string {
137 var dependencies string
138 for reg, ds := range deps {
139 if reg == "nixpkgs" {
140 dependencies = path.Join(ds...)
141 }
142 }
143
144 // load defaults from somewhere else
145 dependencies = path.Join(dependencies, "bash", "git", "coreutils", "nix")
146
147 if runtime.GOARCH == "arm64" {
148 dependencies = path.Join("arm64", dependencies)
149 }
150
151 return path.Join(nixery, dependencies)
152}
153
154func New(ctx context.Context, cfg *config.Config) (*Engine, error) {
155 dcli, err := client.NewClientWithOpts(client.FromEnv, client.WithAPIVersionNegotiation())
156 if err != nil {
157 return nil, err
158 }
159
160 l := log.FromContext(ctx).With("component", "spindle")
161
162 e := &Engine{
163 docker: dcli,
164 l: l,
165 cfg: cfg,
166 }
167
168 e.cleanup = make(map[string][]cleanupFunc)
169
170 return e, nil
171}
172
173func (e *Engine) SetupWorkflow(ctx context.Context, wid models.WorkflowId, wf *models.Workflow) error {
174 e.l.Info("setting up workflow", "workflow", wid)
175
176 _, err := e.docker.NetworkCreate(ctx, networkName(wid), network.CreateOptions{
177 Driver: "bridge",
178 })
179 if err != nil {
180 return err
181 }
182 e.registerCleanup(wid, func(ctx context.Context) error {
183 return e.docker.NetworkRemove(ctx, networkName(wid))
184 })
185
186 addl := wf.Data.(addlFields)
187
188 reader, err := e.docker.ImagePull(ctx, addl.image, image.PullOptions{})
189 if err != nil {
190 e.l.Error("pipeline image pull failed!", "image", addl.image, "workflowId", wid, "error", err.Error())
191
192 return fmt.Errorf("pulling image: %w", err)
193 }
194 defer reader.Close()
195 io.Copy(os.Stdout, reader)
196
197 resp, err := e.docker.ContainerCreate(ctx, &container.Config{
198 Image: addl.image,
199 Cmd: []string{"cat"},
200 OpenStdin: true, // so cat stays alive :3
201 Tty: false,
202 Hostname: "spindle",
203 // TODO(winter): investigate whether environment variables passed here
204 // get propagated to ContainerExec processes
205 }, &container.HostConfig{
206 Mounts: []mount.Mount{
207 {
208 Type: mount.TypeTmpfs,
209 Target: "/tmp",
210 ReadOnly: false,
211 TmpfsOptions: &mount.TmpfsOptions{
212 Mode: 0o1777, // world-writeable sticky bit
213 Options: [][]string{
214 {"exec"},
215 },
216 },
217 },
218 },
219 ReadonlyRootfs: false,
220 CapDrop: []string{"ALL"},
221 CapAdd: []string{"CAP_DAC_OVERRIDE"},
222 SecurityOpt: []string{"no-new-privileges"},
223 ExtraHosts: []string{"host.docker.internal:host-gateway"},
224 }, nil, nil, "")
225 if err != nil {
226 return fmt.Errorf("creating container: %w", err)
227 }
228 e.registerCleanup(wid, func(ctx context.Context) error {
229 err = e.docker.ContainerStop(ctx, resp.ID, container.StopOptions{})
230 if err != nil {
231 return err
232 }
233
234 return e.docker.ContainerRemove(ctx, resp.ID, container.RemoveOptions{
235 RemoveVolumes: true,
236 RemoveLinks: false,
237 Force: false,
238 })
239 })
240
241 err = e.docker.ContainerStart(ctx, resp.ID, container.StartOptions{})
242 if err != nil {
243 return fmt.Errorf("starting container: %w", err)
244 }
245
246 mkExecResp, err := e.docker.ContainerExecCreate(ctx, resp.ID, container.ExecOptions{
247 Cmd: []string{"mkdir", "-p", workspaceDir, homeDir},
248 AttachStdout: true, // NOTE(winter): pretty sure this will make it so that when stdout read is done below, mkdir is done. maybe??
249 AttachStderr: true, // for good measure, backed up by docker/cli ("If -d is not set, attach to everything by default")
250 })
251 if err != nil {
252 return err
253 }
254
255 // This actually *starts* the command. Thanks, Docker!
256 execResp, err := e.docker.ContainerExecAttach(ctx, mkExecResp.ID, container.ExecAttachOptions{})
257 if err != nil {
258 return err
259 }
260 defer execResp.Close()
261
262 // This is apparently best way to wait for the command to complete.
263 _, err = io.ReadAll(execResp.Reader)
264 if err != nil {
265 return err
266 }
267
268 execInspectResp, err := e.docker.ContainerExecInspect(ctx, mkExecResp.ID)
269 if err != nil {
270 return err
271 }
272
273 if execInspectResp.ExitCode != 0 {
274 return fmt.Errorf("mkdir exited with exit code %d", execInspectResp.ExitCode)
275 } else if execInspectResp.Running {
276 return errors.New("mkdir is somehow still running??")
277 }
278
279 addl.container = resp.ID
280 wf.Data = addl
281
282 return nil
283}
284
285func (e *Engine) RunStep(ctx context.Context, wid models.WorkflowId, w *models.Workflow, idx int, secrets []secrets.UnlockedSecret, wfLogger *models.WorkflowLogger) error {
286 addl := w.Data.(addlFields)
287 workflowEnvs := ConstructEnvs(addl.env)
288 // TODO(winter): should SetupWorkflow also have secret access?
289 // IMO yes, but probably worth thinking on.
290 for _, s := range secrets {
291 workflowEnvs.AddEnv(s.Key, s.Value)
292 }
293
294 step := w.Steps[idx].(Step)
295
296 select {
297 case <-ctx.Done():
298 return ctx.Err()
299 default:
300 }
301
302 envs := append(EnvVars(nil), workflowEnvs...)
303 for k, v := range step.environment {
304 envs.AddEnv(k, v)
305 }
306 envs.AddEnv("HOME", homeDir)
307 e.l.Debug("envs for step", "step", step.Name, "envs", envs.Slice())
308
309 mkExecResp, err := e.docker.ContainerExecCreate(ctx, addl.container, container.ExecOptions{
310 Cmd: []string{"bash", "-c", step.command},
311 AttachStdout: true,
312 AttachStderr: true,
313 })
314 if err != nil {
315 return fmt.Errorf("creating exec: %w", err)
316 }
317
318 // start tailing logs in background
319 tailDone := make(chan error, 1)
320 go func() {
321 tailDone <- e.tailStep(ctx, wfLogger, mkExecResp.ID, wid, idx, step)
322 }()
323
324 select {
325 case <-tailDone:
326
327 case <-ctx.Done():
328 // cleanup will be handled by DestroyWorkflow, since
329 // Docker doesn't provide an API to kill an exec run
330 // (sure, we could grab the PID and kill it ourselves,
331 // but that's wasted effort)
332 e.l.Warn("step timed out", "step", step.Name)
333
334 <-tailDone
335
336 return engine.ErrTimedOut
337 }
338
339 select {
340 case <-ctx.Done():
341 return ctx.Err()
342 default:
343 }
344
345 execInspectResp, err := e.docker.ContainerExecInspect(ctx, mkExecResp.ID)
346 if err != nil {
347 return err
348 }
349
350 if execInspectResp.ExitCode != 0 {
351 inspectResp, err := e.docker.ContainerInspect(ctx, addl.container)
352 if err != nil {
353 return err
354 }
355
356 e.l.Error("workflow failed!", "workflow_id", wid.String(), "exit_code", execInspectResp.ExitCode, "oom_killed", inspectResp.State.OOMKilled)
357
358 if inspectResp.State.OOMKilled {
359 return ErrOOMKilled
360 }
361 return engine.ErrWorkflowFailed
362 }
363
364 return nil
365}
366
367func (e *Engine) tailStep(ctx context.Context, wfLogger *models.WorkflowLogger, execID string, wid models.WorkflowId, stepIdx int, step models.Step) error {
368 if wfLogger == nil {
369 return nil
370 }
371
372 // This actually *starts* the command. Thanks, Docker!
373 logs, err := e.docker.ContainerExecAttach(ctx, execID, container.ExecAttachOptions{})
374 if err != nil {
375 return err
376 }
377 defer logs.Close()
378
379 _, err = stdcopy.StdCopy(
380 wfLogger.DataWriter("stdout"),
381 wfLogger.DataWriter("stderr"),
382 logs.Reader,
383 )
384 if err != nil && err != io.EOF && !errors.Is(err, context.DeadlineExceeded) {
385 return fmt.Errorf("failed to copy logs: %w", err)
386 }
387
388 return nil
389}
390
391func (e *Engine) DestroyWorkflow(ctx context.Context, wid models.WorkflowId) error {
392 e.cleanupMu.Lock()
393 key := wid.String()
394
395 fns := e.cleanup[key]
396 delete(e.cleanup, key)
397 e.cleanupMu.Unlock()
398
399 for _, fn := range fns {
400 if err := fn(ctx); err != nil {
401 e.l.Error("failed to cleanup workflow resource", "workflowId", wid, "error", err)
402 }
403 }
404 return nil
405}
406
407func (e *Engine) registerCleanup(wid models.WorkflowId, fn cleanupFunc) {
408 e.cleanupMu.Lock()
409 defer e.cleanupMu.Unlock()
410
411 key := wid.String()
412 e.cleanup[key] = append(e.cleanup[key], fn)
413}
414
415func networkName(wid models.WorkflowId) string {
416 return fmt.Sprintf("workflow-network-%s", wid)
417}