alice.pds.demo.boltless.dev
This repository has no description
1package microvm
2
3import (
4 "context"
5 "encoding/json"
6 "errors"
7 "fmt"
8 "io"
9 "log/slog"
10 "net/http"
11 "os"
12 "path/filepath"
13 "slices"
14 "sync"
15 "sync/atomic"
16 "time"
17
18 "gopkg.in/yaml.v3"
19
20 "tangled.org/core/api/tangled"
21 "tangled.org/core/log"
22 "tangled.org/core/spindle/agentproto"
23 agentv1 "tangled.org/core/spindle/agentproto/gen"
24 "tangled.org/core/spindle/config"
25 "tangled.org/core/spindle/db"
26 "tangled.org/core/spindle/engine"
27 "tangled.org/core/spindle/models"
28 "tangled.org/core/spindle/secrets"
29)
30
31const (
32 guestWorkDir = "/workspace/repo"
33 guestBasePATH = "/run/current-system/sw/bin:/nix/var/nix/profiles/default/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin"
34 guestDevShellEnvPath = "/run/spindle/devshell-env.sh"
35 activationStepAction = "activate-config"
36 agentAcceptTimeout = 2 * time.Minute
37 agentHandshakeTimeout = 30 * time.Second
38 cacheDrainTimeout = 5 * time.Minute
39 vmShutdownTimeout = 10 * time.Second
40 guestTimeoutGrace = 5 * time.Second
41)
42
43type cleanupFunc func(context.Context) error
44
45type Engine struct {
46 l *slog.Logger
47 cfg *config.Config
48 db *db.DB
49 agent *agentHub
50 scheduler *engine.ResourceScheduler[Resources]
51 cgroupParent *CgroupParent
52
53 cleanupMu sync.Mutex
54 cleanup map[string][]cleanupFunc
55}
56
57type Step struct {
58 name string
59 kind models.StepKind
60 command string
61 environment map[string]string
62 action string
63 config manifestConfig
64 configKey string
65}
66
67func (s Step) Name() string { return s.name }
68func (s Step) Command() string { return s.command }
69func (s Step) Kind() models.StepKind { return s.kind }
70
71func New(ctx context.Context, cfg *config.Config, d *db.DB) (*Engine, error) {
72 l := log.FromContext(ctx).With("component", "engine.microvm")
73 port := cfg.MicroVMPipelines.AgentPort
74 if port == 0 {
75 port = agentproto.DefaultPort
76 }
77 agent, err := newAgentHub(port, l)
78 if err != nil {
79 return nil, err
80 }
81 budget, max, agingThreshold := newVMBudgetConfig(cfg.MicroVMPipelines)
82 l.Info("initialized microVM workflow budget", "budget", budget.String(), "maxWorkflow", max.String(), "agingThreshold", agingThreshold)
83
84 var cgroupParent *CgroupParent
85 if cfg.MicroVMPipelines.EnableCgroups {
86 cgroupParent, err = initCgroupParent(cfg.MicroVMPipelines.CgroupParent, cfg.MicroVMPipelines.CgroupSupervisorMemoryMinMiB, l)
87 if err != nil {
88 return nil, err
89 }
90 }
91
92 return &Engine{
93 l: l,
94 cfg: cfg,
95 db: d,
96 agent: agent,
97 scheduler: engine.NewResourceScheduler(budget, max, agingThreshold),
98 cgroupParent: cgroupParent,
99 cleanup: make(map[string][]cleanupFunc),
100 }, nil
101}
102
103func (e *Engine) InitWorkflow(twf tangled.Pipeline_Workflow, tpl tangled.Pipeline) (*models.Workflow, error) {
104 swf := &models.Workflow{}
105 var dwf manifestWorkflow
106
107 if err := engine.DescribeManifestError(twf.Raw, manifestWorkflow{}); err != nil {
108 return nil, err
109 }
110 if err := yaml.Unmarshal([]byte(twf.Raw), &dwf); err != nil {
111 return nil, err
112 }
113
114 for _, dstep := range dwf.Steps {
115 swf.Steps = append(swf.Steps, Step{
116 name: dstep.Name,
117 kind: models.StepKindUser,
118 command: dstep.Command,
119 environment: dstep.Environment,
120 })
121 }
122 swf.Name = twf.Name
123 swf.Environment = dwf.Environment
124
125 if tpl.TriggerMetadata != nil {
126 if clone := models.BuildCloneStep(twf, *tpl.TriggerMetadata, e.cfg.Server.Dev); clone.Command() != "" {
127 swf.Steps = append([]models.Step{clone}, swf.Steps...)
128 }
129 }
130
131 imageSpec, imageSpecPath, imageName, err := e.resolveImage(dwf.Image)
132 if err != nil {
133 return nil, err
134 }
135 configKey := ""
136 config := manifestConfig{
137 Services: dwf.Services,
138 Virtualisation: dwf.Virtualisation,
139 Dependencies: dwf.Dependencies,
140 Registry: dwf.Registry,
141 }
142 if config.Enabled() {
143 if !imageSpec.SupportsConfigActivation() {
144 return nil, fmt.Errorf(
145 "microVM image %q is not a NixOS image: services, virtualisation, dependencies and registry workflow options require a NixOS image",
146 imageName,
147 )
148 }
149 var err error
150 configKey, err = buildConfigKey(imageSpec, config)
151 if err != nil {
152 return nil, fmt.Errorf("build config key: %w", err)
153 }
154 activationStep := Step{
155 name: "NixOS config activation",
156 kind: models.StepKindSystem,
157 command: "activate nixos config",
158 action: activationStepAction,
159 config: config,
160 configKey: configKey,
161 }
162
163 insertAt := 0
164 if len(swf.Steps) > 0 && swf.Steps[0].Kind() == models.StepKindSystem {
165 insertAt = 1
166 }
167 swf.Steps = append(swf.Steps, nil)
168 copy(swf.Steps[insertAt+1:], swf.Steps[insertAt:])
169 swf.Steps[insertAt] = activationStep
170 }
171
172 cacheURLs, cacheKeys, err := workflowCaches(dwf.Caches)
173 if err != nil {
174 return nil, err
175 }
176
177 swf.Data = &workflowState{
178 ImageSpec: imageSpec,
179 ImageSpecPath: imageSpecPath,
180 Config: config,
181 ConfigKey: configKey,
182 Image: imageName,
183 CacheReadURLs: cacheURLs,
184 CacheTrustedPublicKeys: cacheKeys,
185 NixOSToplevelCache: newNixOSToplevelCacheStore(e.db),
186 }
187 return swf, nil
188}
189
190func (e *Engine) SetupWorkflow(ctx context.Context, wid models.WorkflowId, wf *models.Workflow, wfLogger models.WorkflowLogger) error {
191 l := e.l.With("workflow", wid)
192 setupStep := Step{name: "microVM setup", kind: models.StepKindSystem}
193
194 wfLogger.ControlWriter(-1, setupStep, models.StepStatusStart).Write([]byte{0})
195 defer wfLogger.ControlWriter(-1, setupStep, models.StepStatusEnd).Write([]byte{0})
196
197 state, ok := wf.Data.(*workflowState)
198 if !ok || state == nil {
199 return fmt.Errorf("workflow state is not initialized")
200 }
201
202 cid, err := AllocateCID()
203 if err != nil {
204 return err
205 }
206 connCh, unregister, err := e.agent.expect(cid)
207 if err != nil {
208 return err
209 }
210 defer unregister()
211
212 workDirBase := e.cfg.MicroVMPipelines.OverlayDir
213 if workDirBase == "" {
214 workDirBase = os.TempDir()
215 }
216 workDir, err := os.MkdirTemp(workDirBase, "spindle-microvm-"+wid.String()+"-*")
217 if err != nil {
218 return fmt.Errorf("create workflow microVM directory: %w", err)
219 }
220 state.WorkDir = workDir
221
222 setupDone := false
223 defer func() {
224 if setupDone {
225 return
226 }
227 if detail := vmCrashLog(state.VM); detail != "" {
228 l.Error("microVM setup failed", "detail", detail)
229 }
230 if err := e.cleanupState(context.Background(), wid, state); err != nil {
231 l.Error("failed to cleanup failed setup", "error", err)
232 }
233 }()
234
235 upstreams, err := BuildCacheUpstreams(e.cfg.NixCache.ReadURLs, state.CacheReadURLs)
236 if err != nil {
237 return err
238 }
239 readCache, err := StartReadCacheProxy(ctx, cid, upstreams, l)
240 if err != nil {
241 return err
242 }
243 state.ReadCache = readCache
244 stagingDir := filepath.Join(workDir, "upload-cache")
245 uploadCache, err := StartUploadCacheProxy(ctx, cid, e.cfg.NixCache.UploadURL, upstreams, stagingDir, l)
246 if err != nil {
247 return err
248 }
249 state.UploadCache = uploadCache
250 dnsProxy, err := StartDNSProxy(ctx, cid, l)
251 if err != nil {
252 return err
253 }
254 state.DNSProxy = dnsProxy
255
256 port := e.cfg.MicroVMPipelines.AgentPort
257 if port == 0 {
258 port = agentproto.DefaultPort
259 }
260 state.ImageSpec.BootArgs = fmt.Sprintf("%s shuttle.vsock_port=%d", state.ImageSpec.BootArgs, port)
261
262 fmt.Fprintf(wfLogger.DataWriter(-1, "stdout"), "starting microVM image %s\n", state.Image)
263 l.Info("starting microVM workflow", "image", state.Image, "imageSpec", state.ImageSpecPath, "cid", cid, "workDir", workDir)
264
265 var vm VMHandle
266 vm, err = StartVM(ctx, VMConfig{
267 Image: state.ImageSpec,
268 CID: cid,
269 EnableKVM: e.cfg.MicroVMPipelines.EnableKVM,
270 WorkDir: workDir,
271 Cgroup: e.cgroupLimits(wid, state.ImageSpec),
272 Dev: e.cfg.Server.Dev,
273 }, l)
274 if err != nil {
275 return err
276 }
277 state.VM = vm
278
279 acceptCtx, cancelAccept := context.WithTimeout(ctx, agentAcceptTimeout)
280 defer cancelAccept()
281 conn, err := waitAgentConn(acceptCtx, connCh)
282 if err != nil {
283 return err
284 }
285
286 agentSession := NewAgentSession(conn, l)
287 initCtx, cancelInit := context.WithTimeout(ctx, agentHandshakeTimeout)
288 defer cancelInit()
289 if err := agentSession.Init(initCtx, &agentv1.Init{
290 JobId: wid.String(),
291 CacheTrustedPublicKeys: append(slices.Clone(e.cfg.NixCache.TrustedPublicKeys), state.CacheTrustedPublicKeys...),
292 CacheReadProxyPort: readCache.Port(),
293 CacheUploadProxyPort: uploadCache.Port(),
294 DnsProxyPort: dnsProxy.Port(),
295 }); err != nil {
296 _ = agentSession.Close()
297 return err
298 }
299 state.Agent = agentSession
300 wf.Data = state
301
302 e.registerCleanup(wid, func(ctx context.Context) error {
303 return e.cleanupState(ctx, wid, state)
304 })
305 setupDone = true
306
307 fmt.Fprintf(wfLogger.DataWriter(-1, "stdout"),
308 "agent connected; serial log: %s\n", vm.Logs().Serial,
309 )
310 return nil
311}
312
313func applyDepsSource(command string) string {
314 return fmt.Sprintf(
315 // check if it exists because not all images have this
316 `if [ -f %s ]; then . %s; export PATH="$PATH:%s"; fi; %s`,
317 guestDevShellEnvPath, guestDevShellEnvPath, guestBasePATH, command,
318 )
319}
320
321func (e *Engine) RunStep(ctx context.Context, wid models.WorkflowId, w *models.Workflow, idx int, secrets []secrets.UnlockedSecret, wfLogger models.WorkflowLogger) error {
322 state, ok := w.Data.(*workflowState)
323 if !ok || state == nil || state.Agent == nil {
324 return fmt.Errorf("microVM workflow is not connected to agent")
325 }
326
327 stderr := wfLogger.DataWriter(idx, "stderr")
328
329 execCtx, vmExited, cancelWatch := watchVMExit(ctx, state.VM)
330 defer cancelWatch()
331
332 step := w.Steps[idx]
333 if s, ok := step.(Step); ok && s.action == activationStepAction {
334 err := e.activateConfig(execCtx, wid, state, s, wfLogger.DataWriter(idx, "stdout"))
335 return e.classifyStepError(ctx, wid, step, state, stderr, vmExited, err)
336 }
337 env := []string{
338 "HOME=/workspace",
339 "LOGNAME=" + guestWorkflowUser,
340 "PATH=" + guestBasePATH,
341 "USER=" + guestWorkflowUser,
342 }
343 for k, v := range w.Environment {
344 env = append(env, k+"="+v)
345 }
346 for _, s := range secrets {
347 env = append(env, s.Key+"="+s.Value)
348 }
349 if s, ok := step.(Step); ok {
350 for k, v := range s.environment {
351 env = append(env, k+"="+v)
352 }
353 }
354
355 stdout := wfLogger.DataWriter(idx, "stdout")
356 exitCode, err := state.Agent.Exec(execCtx, AgentExec{
357 ID: fmt.Sprintf("%s-%d", wid.String(), idx),
358 ExecStart: &agentv1.ExecStart{
359 Argv: []string{state.ImageSpec.Shell, "-lc", applyDepsSource(step.Command())},
360 Env: env,
361 Cwd: guestWorkDir,
362 User: guestWorkflowUser,
363 // timeout not set here, Exec will fill it
364 },
365 Stdout: stdout,
366 Stderr: stderr,
367 })
368 if err != nil {
369 return e.classifyStepError(ctx, wid, step, state, stderr, vmExited, err)
370 }
371
372 if exitCode != 0 {
373 e.l.Debug("step exited non-zero", "workflow", wid, "step", step.Name(), "exitCode", exitCode)
374 return engine.ErrWorkflowFailed
375 }
376 return nil
377}
378
379// reads the vm serial logs so we report the tail of that as an error instead of
380// just "guest agent connection lost: EOF"
381func (e *Engine) classifyStepError(ctx context.Context, wid models.WorkflowId, step models.Step, state *workflowState, stderr io.Writer, vmExited *atomic.Bool, err error) error {
382 if err == nil {
383 return nil
384 }
385 l := e.l.With("workflow", wid, "step", step.Name())
386
387 if vmExited != nil && vmExited.Load() {
388 reason := "microVM exited unexpectedly"
389 oom := state.VM != nil && state.VM.OOMKilled()
390 if oom {
391 reason = "microVM killed by OOM (cgroup memory limit exceeded)"
392 }
393 if detail := vmCrashLog(state.VM); detail != "" {
394 fmt.Fprintf(stderr, "%s:\n%s\n", reason, detail)
395 l.Error(reason, "oom", oom, "detail", detail)
396 } else {
397 fmt.Fprintln(stderr, reason)
398 l.Error(reason, "oom", oom)
399 }
400 return errors.New(reason + "; see workflow logs for serial output")
401 }
402
403 if errors.Is(err, errGuestTimedOut) || ctx.Err() != nil {
404 l.Debug("step timed out", "guestReported", errors.Is(err, errGuestTimedOut))
405 return engine.ErrTimedOut
406 }
407
408 // the agent connection dropped while qemu stayed up (eg. the guest kernel
409 // OOM-killed the agent or a guest panic), so surface serial logs, those
410 // will be more helpful.
411 if detail := vmCrashLog(state.VM); detail != "" {
412 fmt.Fprintf(stderr, "step failed (%v):\n%s\n", err, detail)
413 l.Error("step failed", "error", err, "detail", detail)
414 } else {
415 l.Error("step failed", "error", err)
416 }
417 return err
418}
419
420func (e *Engine) activateConfig(ctx context.Context, wid models.WorkflowId, state *workflowState, step Step, out io.Writer) error {
421 cfg := step.config
422 if !cfg.Enabled() {
423 return nil
424 }
425
426 configKey := step.configKey
427 if configKey == "" {
428 configKey = state.ConfigKey
429 }
430
431 userConfigJSON, err := json.Marshal(cfg)
432 if err != nil {
433 return fmt.Errorf("encode user config: %w", err)
434 }
435
436 var cachedToplevel string
437 if configKey != "" {
438 if record, ok, err := state.NixOSToplevelCache.Lookup(configKey); err != nil {
439 return err
440 } else if ok {
441 // todo(dawn): we should probably use gc roots to eliminate TOCTOU
442 // the spindle will have to manage the gc roots, and for remote we have to
443 // ssh in to the host and add / remove gc root.
444 // we need to have this check anyway since the only check http caches can
445 // use is this one, since we cant manage gc roots there...
446 if e.anyCacheHasPath(ctx, state, record.Toplevel) {
447 cachedToplevel = record.Toplevel
448 fmt.Fprintf(out, "realizing cached NixOS config %s\n", cachedToplevel)
449 }
450 }
451 }
452 if cachedToplevel == "" {
453 fmt.Fprintf(out, "building NixOS config from user config\n")
454 }
455
456 baseHash, err := BaseConfigHash(state.ImageSpec)
457 if err != nil {
458 return fmt.Errorf("calculate base config hash: %w", err)
459 }
460
461 result, err := state.Agent.ActivateConfig(ctx, fmt.Sprintf("%s-config", wid.String()), &agentv1.ActivateConfig{
462 ConfigKey: configKey,
463 BaseConfigHash: baseHash,
464 UserConfig: string(userConfigJSON),
465 Toplevel: cachedToplevel,
466 }, out)
467 if err != nil {
468 return err
469 }
470 fmt.Fprintf(out, "activated NixOS config toplevel %s\n", result.Toplevel)
471
472 if cachedToplevel != "" || configKey == "" {
473 return nil
474 }
475 if e.cfg.NixCache.UploadURL == "" {
476 e.l.Warn("not committing config cache metadata: no upload URL configured", "workflow", wid, "configKey", configKey, "toplevel", result.Toplevel)
477 return nil
478 }
479
480 if err := e.drainNixCache(ctx, state); err != nil {
481 // a partial upload would leave the cache unable to realize this toplevel,
482 // so skip the metadata commit rather than poison it with an un-realizable
483 // key. the config still activated fine, so don't fail the workflow.
484 e.l.Warn("cache drain failed; skipping config cache metadata commit", "workflow", wid, "configKey", configKey, "toplevel", result.Toplevel, "error", err)
485 return nil
486 }
487 if err := state.NixOSToplevelCache.Commit(configKey, result.Toplevel); err != nil {
488 return err
489 }
490 fmt.Fprintf(out, "committed config cache metadata %s -> %s\n", configKey, result.Toplevel)
491 return nil
492}
493
494func (e *Engine) anyCacheHasPath(ctx context.Context, state *workflowState, storePath string) bool {
495 upstreams, err := BuildCacheUpstreams(e.cfg.NixCache.ReadURLs, state.CacheReadURLs)
496 if err != nil {
497 e.l.Warn("config cache check: build upstreams failed; treating as absent", "path", storePath, "error", err)
498 return false
499 }
500 if len(upstreams) == 0 {
501 return false
502 }
503 hash, _, err := parseStorePath(storePath)
504 if err != nil {
505 e.l.Warn("config cache check: invalid toplevel path; treating as absent", "path", storePath, "error", err)
506 return false
507 }
508 req, err := http.NewRequestWithContext(ctx, http.MethodHead, "http://upstream/"+hash+".narinfo", nil)
509 if err != nil {
510 e.l.Warn("config cache check: build request failed; treating as absent", "path", storePath, "error", err)
511 return false
512 }
513 resp, err := newNarinfoExistenceTransport(upstreams, e.l).RoundTrip(req)
514 if err != nil {
515 e.l.Warn("config cache check: narinfo probe failed; treating as absent", "path", storePath, "error", err)
516 return false
517 }
518 defer resp.Body.Close()
519 _, _ = io.Copy(io.Discard, resp.Body)
520 return resp.StatusCode == http.StatusOK
521}
522
523func (e *Engine) DestroyWorkflow(ctx context.Context, wid models.WorkflowId) error {
524 fns := e.drainCleanups(wid)
525
526 var cleanupErr error
527 for i := len(fns) - 1; i >= 0; i-- {
528 if err := fns[i](ctx); err != nil {
529 e.l.Error("failed to cleanup workflow resource", "workflowId", wid, "error", err)
530 cleanupErr = errors.Join(cleanupErr, err)
531 }
532 }
533 return cleanupErr
534}
535
536func (e *Engine) FinalizeWorkflow(ctx context.Context, wid models.WorkflowId, w *models.Workflow, wfLogger models.WorkflowLogger) error {
537 return nil
538}
539
540func (e *Engine) WorkflowTimeout() time.Duration {
541 d, err := time.ParseDuration(e.cfg.MicroVMPipelines.WorkflowTimeout)
542 if err != nil {
543 d = 5 * time.Minute
544 }
545 return d + guestTimeoutGrace
546}
547
548func (e *Engine) registerCleanup(wid models.WorkflowId, fn cleanupFunc) {
549 e.cleanupMu.Lock()
550 defer e.cleanupMu.Unlock()
551 key := wid.String()
552 e.cleanup[key] = append(e.cleanup[key], fn)
553}
554
555func (e *Engine) drainCleanups(wid models.WorkflowId) []cleanupFunc {
556 e.cleanupMu.Lock()
557 defer e.cleanupMu.Unlock()
558 key := wid.String()
559 fns := e.cleanup[key]
560 delete(e.cleanup, key)
561 return fns
562}
563
564func (e *Engine) cgroupLimits(wid models.WorkflowId, spec ImageSpec) CgroupLimits {
565 cfg := e.cfg.MicroVMPipelines
566 return CgroupLimits{
567 Enabled: cfg.EnableCgroups,
568 Parent: e.cgroupParent,
569 Name: "workflow-" + wid.String(),
570 MemoryMaxMiB: resourcesForImage(spec).MemoryMiB,
571 SwapMaxMiB: cfg.CgroupSwapMaxMiB,
572 PidsMax: cfg.CgroupPidsMax,
573 }
574}