Bound concurrent image processing and upstream fetches (closes #64)

max_concurrent_processing (default: the number of CPUs Go uses) bounds
the images processed at once, and upstream_connections (default 64) the
fetches from all upstream hosts together, beside the per-host limit. A
request that finds either full waits up to 10 seconds, then gets 503
"server busy, try again later". The processor holds its slot from before
it reads the input until it returns, and takes a free slot even after the
request context has ended; a fetch holds its connection until the
response body is closed, after its image is processed. libvips now starts
with one worker thread per image and no operation cache. Both settings
have PIXA_ variables and are in README.md and config.example.yml.

Model: opus-5-5

internal/imageprocessor/imageprocessor.go

@@ -7,7 +7,9 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"runtime"
 	"sync"
+	"time"
 
 	"github.com/davidbyttow/govips/v2/vips"
 )
@@ -17,11 +19,21 @@ import (
 //nolint:gochecknoglobals // package-level sync.Once for one-time vips init
 var vipsOnce sync.Once
 
-// initVips initializes libvips with quiet logging.
+// initVips initializes libvips with quiet logging, one worker thread per
+// image and no operation cache. Process already works on one image per CPU
+// by default, so more threads per image would only compete for the CPUs.
+// Each request decodes different source bytes, so the operation cache
+// would rarely be hit and would hold memory outside MaxConcurrentProcessing;
+// repeated requests are served from pixa's disk cache instead.
 func initVips() {
 	vipsOnce.Do(func() {
 		vips.LoggingSettings(nil, vips.LogLevelError)
-		vips.Startup(nil)
+		vips.Startup(&vips.Config{
+			ConcurrencyLevel: 1,
+			MaxCacheSize:     0,
+			MaxCacheMem:      0,
+			MaxCacheFiles:    0,
+		})
 	})
 }
 
@@ -106,9 +118,23 @@ var ErrInputDataTooLarge = errors.New("input data exceeds maximum allowed size")
 // not supported.
 var ErrUnsupportedOutputFormat = errors.New("unsupported output format")
 
+// ErrTooManyImages is returned when MaxConcurrentProcessing images are being
+// processed and none finishes within ProcessingWaitTimeout.
+var ErrTooManyImages = errors.New("too many images being processed at once")
+
+// ProcessingWaitTimeout is how long Process waits for a free slot when
+// MaxConcurrentProcessing images are already being processed.
+const ProcessingWaitTimeout = 10 * time.Second
+
 // ImageProcessor implements image transformation using libvips via govips.
 type ImageProcessor struct {
 	maxInputBytes int64
+	// processingSemaphore has one slot per image that may be processed at
+	// once. Process holds a slot from before it reads its input until it
+	// returns, so the input, the decoded image and the output all count.
+	processingSemaphore chan struct{}
+	// processingWaitTimeout is ProcessingWaitTimeout; tests shorten it.
+	processingWaitTimeout time.Duration
 }
 
 // Params holds configuration for creating an ImageProcessor.
@@ -117,6 +143,9 @@ type Params struct {
 	// MaxInputBytes is the maximum allowed input size in bytes.
 	// If <= 0, DefaultMaxInputBytes is used.
 	MaxInputBytes int64
+	// MaxConcurrentProcessing is the most images processed at once.
+	// If <= 0, the number of CPUs Go uses (runtime.GOMAXPROCS(0)) is used.
+	MaxConcurrentProcessing int
 }
 
 // New creates a new image processor with the given parameters.
@@ -129,17 +158,34 @@ func New(params Params) *ImageProcessor {
 		maxInputBytes = DefaultMaxInputBytes
 	}
 
+	maxConcurrentProcessing := params.MaxConcurrentProcessing
+	if maxConcurrentProcessing <= 0 {
+		maxConcurrentProcessing = runtime.GOMAXPROCS(0)
+	}
+
 	return &ImageProcessor{
-		maxInputBytes: maxInputBytes,
+		maxInputBytes:         maxInputBytes,
+		processingSemaphore:   make(chan struct{}, maxConcurrentProcessing),
+		processingWaitTimeout: ProcessingWaitTimeout,
 	}
 }
 
-// Process transforms an image according to the request.
+// Process transforms an image according to the request. When
+// MaxConcurrentProcessing images are already being processed, it waits up
+// to ProcessingWaitTimeout for one to finish, then fails with
+// ErrTooManyImages.
 func (p *ImageProcessor) Process(
-	_ context.Context,
+	ctx context.Context,
 	input io.Reader,
 	req *Request,
 ) (*Result, error) {
+	release, err := p.acquireSlot(ctx)
+	if err != nil {
+		return nil, err
+	}
+
+	defer release()
+
 	// Read input with a size limit to prevent unbounded memory consumption.
 	// We read at most maxInputBytes+1 so we can detect if the input exceeds
 	// the limit without consuming additional memory.
@@ -285,6 +331,29 @@ func FormatToMIME(format Format) string {
 	}
 }
 
+// acquireSlot takes a slot in processingSemaphore, waiting at most
+// processingWaitTimeout for one to free up, and returns the func that gives
+// it back. A free slot is taken even when ctx has ended; only the wait for
+// one stops when ctx ends, as the rest of Process does not check ctx.
+func (p *ImageProcessor) acquireSlot(ctx context.Context) (func(), error) {
+	release := func() { <-p.processingSemaphore }
+
+	select {
+	case p.processingSemaphore <- struct{}{}:
+		return release, nil
+	default:
+	}
+
+	select {
+	case p.processingSemaphore <- struct{}{}:
+		return release, nil
+	case <-time.After(p.processingWaitTimeout):
+		return nil, ErrTooManyImages
+	case <-ctx.Done():
+		return nil, ctx.Err()
+	}
+}
+
 // detectFormat returns the format string from a vips image.
 func (p *ImageProcessor) detectFormat(img *vips.ImageRef) string {
 	format := img.Format()