commit dc7f387dda3e76c2de69d2c45b926cca1f289d65 · veryroundbird.house/core

+2 -1

spindle/config/config.go

···

       17
       17
        
       

     

       18
       18
        
       type Pipelines struct {

     

       19
       19
        
       	// TODO: change default to nixery.tangled.sh

     

       20
       20
       -
       	Nixery string `env:"NIXERY, default=nixery.dev"`

     

       20
       20
       +
       	Nixery      string `env:"NIXERY, default=nixery.dev"`

     

       21
       21
       +
       	StepTimeout string `env:"STEP_TIMEOUT, default=5m"`

     

       21
       22
        
       }

     

       22
       23
        
       

     

       23
       24
        
       type Config struct {

+2 -2

spindle/engine/ansi_stripper.go

···

       3
       3
        
       import (

     

       4
       4
        
       	"io"

     

       5
       5
        
       

     

       6
       6
       -
       	"github.com/go-enry/go-enry/v2/regex"

     

       6
       6
       +
       	"regexp"

     

       7
       7
        
       )

     

       8
       8
        
       

     

       9
       9
        
       // regex to match ANSI escape codes (e.g., color codes, cursor moves)

     

       10
       10
        
       const ansi = "[\u001B\u009B][[\\]()#;?]*(?:(?:(?:[a-zA-Z\\d]*(?:;[a-zA-Z\\d]*)*)?\u0007)|(?:(?:\\d{1,4}(?:;\\d{0,4})*)?[\\dA-PRZcf-ntqry=><~]))"

     

       11
       11
        
       

     

       12
       12
       -
       var re = regex.MustCompile(ansi)

     

       12
       12
       +
       var re = regexp.MustCompile(ansi)

     

       13
       13
        
       

     

       14
       14
        
       type ansiStrippingWriter struct {

     

       15
       15
        
       	underlying io.Writer

+52 -19

spindle/engine/engine.go

···

       3
       3
        
       import (

     

       4
       4
        
       	"bufio"

     

       5
       5
        
       	"context"

     

       6
       6
       +
       	"errors"

     

       6
       7
        
       	"fmt"

     

       7
       8
        
       	"io"

     

       8
       9
        
       	"log/slog"

     

       9
       10
        
       	"os"

     

       10
       11
        
       	"strings"

     

       11
       12
        
       	"sync"

     

       13
       13
       +
       	"time"

     

       12
       14
        
       

     

       13
       15
        
       	"github.com/docker/docker/api/types/container"

     

       14
       16
        
       	"github.com/docker/docker/api/types/image"

     
···

       176
       178
        
       // StartSteps starts all steps sequentially with the same base image.

     

       177
       179
        
       // ONLY marks pipeline as failed if container's exit code is non-zero.

     

       178
       180
        
       // All other errors are bubbled up.

     

       181
       181
       +
       // Fixed version of the step execution logic

     

       179
       182
        
       func (e *Engine) StartSteps(ctx context.Context, steps []models.Step, wid models.WorkflowId, image string) error {

     

       180
       180
       -
       	// set up logging channels

     

       183
       183
       +
       	stepTimeoutStr := e.cfg.Pipelines.StepTimeout

     

       184
       184
       +
       	stepTimeout, err := time.ParseDuration(stepTimeoutStr)

     

       185
       185
       +
       	if err != nil {

     

       186
       186
       +
       		e.l.Error("failed to parse step timeout", "error", err, "timeout", stepTimeoutStr)

     

       187
       187
       +
       		stepTimeout = 5 * time.Minute

     

       188
       188
       +
       	}

     

       189
       189
       +
       	e.l.Info("using step timeout", "timeout", stepTimeout)

     

       190
       190
       +
       

     

       181
       191
        
       	e.chanMu.Lock()

     

       182
       192
        
       	if _, exists := e.stdoutChans[wid.String()]; !exists {

     

       183
       193
        
       		e.stdoutChans[wid.String()] = make(chan string, 100)

     
···

       216
       226
        
       			return fmt.Errorf("connecting network: %w", err)

     

       217
       227
        
       		}

     

       218
       228
        
       

     

       219
       219
       -
       		err = e.docker.ContainerStart(ctx, resp.ID, container.StartOptions{})

     

       229
       229
       +
       		stepCtx, stepCancel := context.WithTimeout(ctx, stepTimeout)

     

       230
       230
       +
       

     

       231
       231
       +
       		err = e.docker.ContainerStart(stepCtx, resp.ID, container.StartOptions{})

     

       220
       232
        
       		if err != nil {

     

       233
       233
       +
       			stepCancel()

     

       221
       234
        
       			return err

     

       222
       235
        
       		}

     

       223
       236
        
       		e.l.Info("started container", "name", resp.ID, "step", step.Name)

     

       224
       237
        
       

     

       225
       225
       -
       		wg := sync.WaitGroup{}

     

       238
       238
       +
       		// start tailing logs in background

     

       239
       239
       +
       		tailDone := make(chan error, 1)

     

       240
       240
       +
       		go func() {

     

       241
       241
       +
       			tailDone <- e.TailStep(stepCtx, resp.ID, wid)

     

       242
       242
       +
       		}()

     

       226
       243
        
       

     

       227
       227
       -
       		wg.Add(1)

     

       244
       244
       +
       		// wait for container completion or timeout

     

       245
       245
       +
       		waitDone := make(chan struct{})

     

       246
       246
       +
       		var state *container.State

     

       247
       247
       +
       		var waitErr error

     

       248
       248
       +
       

     

       228
       249
        
       		go func() {

     

       229
       229
       -
       			defer wg.Done()

     

       230
       230
       -
       			err := e.TailStep(ctx, resp.ID, wid)

     

       231
       231
       -
       			if err != nil {

     

       232
       232
       -
       				e.l.Error("failed to tail container", "container", resp.ID)

     

       233
       233
       -
       				return

     

       234
       234
       -
       			}

     

       250
       250
       +
       			defer close(waitDone)

     

       251
       251
       +
       			state, waitErr = e.WaitStep(stepCtx, resp.ID)

     

       235
       252
        
       		}()

     

       236
       253
        
       

     

       237
       237
       -
       		// wait until all logs are piped

     

       238
       238
       -
       		wg.Wait()

     

       254
       254
       +
       		select {

     

       255
       255
       +
       		case <-waitDone:

     

       256
       256
       +
       			// container finished normally

     

       257
       257
       +
       			stepCancel()

     

       258
       258
       +
       

     

       259
       259
       +
       			// wait for tailing to complete

     

       260
       260
       +
       			<-tailDone

     

       261
       261
       +
       

     

       262
       262
       +
       		case <-stepCtx.Done():

     

       263
       263
       +
       			e.l.Warn("step timed out; killing container", "container", resp.ID, "timeout", stepTimeout)

     

       264
       264
       +
       

     

       265
       265
       +
       			_ = e.DestroyStep(ctx, resp.ID)

     

       266
       266
       +
       

     

       267
       267
       +
       			// wait for both goroutines to finish

     

       268
       268
       +
       			<-waitDone

     

       269
       269
       +
       			<-tailDone

     

       270
       270
       +
       

     

       271
       271
       +
       			stepCancel()

     

       272
       272
       +
       			return fmt.Errorf("step timed out after %v", stepTimeout)

     

       273
       273
       +
       		}

     

       239
       274
        
       

     

       240
       240
       -
       		state, err := e.WaitStep(ctx, resp.ID)

     

       241
       241
       -
       		if err != nil {

     

       242
       242
       -
       			return err

     

       275
       275
       +
       		if waitErr != nil {

     

       276
       276
       +
       			return waitErr

     

       243
       277
        
       		}

     

       244
       278
        
       

     

       245
       279
        
       		err = e.DestroyStep(ctx, resp.ID)

     
···

       253
       287
        
       			if err != nil {

     

       254
       288
        
       				return err

     

       255
       289
        
       			}

     

       256
       256
       -
       			return fmt.Errorf("error: %s, exit code: %d, oom: %s", state.Error, state.ExitCode, state.OOMKilled)

     

       290
       290
       +
       			return fmt.Errorf("error: %s, exit code: %d, oom: %t", state.Error, state.ExitCode, state.OOMKilled)

     

       257
       291
        
       		}

     

       258
       292
        
       	}

     

       259
       293
        
       

     

       260
       294
        
       	return nil

     

       261
       261
       -
       

     

       262
       295
        
       }

     

       263
       296
        
       

     

       264
       297
        
       func (e *Engine) WaitStep(ctx context.Context, containerID string) (*container.State, error) {

     
···

       318
       351
        
       		defer wpipeOut.Close()

     

       319
       352
        
       		defer wpipeErr.Close()

     

       320
       353
        
       		_, err := stdcopy.StdCopy(wpipeOut, wpipeErr, tee)

     

       321
       321
       -
       		if err != nil && err != io.EOF {

     

       354
       354
       +
       		if err != nil && err != io.EOF && !errors.Is(context.DeadlineExceeded, err) {

     

       322
       355
        
       			e.l.Error("failed to copy logs", "error", err)

     

       323
       356
        
       		}

     

       324
       357
        
       	}()

     
···

       393
       426
        
       

     

       394
       427
        
       	for _, fn := range fns {

     

       395
       428
        
       		if err := fn(ctx); err != nil {

     

       396
       396
       -
       			e.l.Error("failed to cleanup workflow resource", "workflowId", wid)

     

       429
       429
       +
       			e.l.Error("failed to cleanup workflow resource", "workflowId", wid, "error", err)

     

       397
       430
        
       		}

     

       398
       431
        
       	}

     

       399
       432
        
       	return nil