coder · aslilac · Jul 1, 2025 · Jun 23, 2025 · Jun 24, 2025 · Jun 24, 2025
diff --git a/coderd/files/cache.go b/coderd/files/cache.go
@@ -25,60 +25,61 @@ type FileAcquirer interface {
 
 // New returns a file cache that will fetch files from a database
 func New(registerer prometheus.Registerer, authz rbac.Authorizer) *Cache {
-	return (&Cache{
-		lock:  sync.Mutex{},
-		data:  make(map[uuid.UUID]*cacheEntry),
-		authz: authz,
-	}).registerMetrics(registerer)
+	return &Cache{
+		lock:         sync.Mutex{},
+		data:         make(map[uuid.UUID]*cacheEntry),
+		authz:        authz,
+		cacheMetrics: newCacheMetrics(registerer),
+	}
 }
 
-func (c *Cache) registerMetrics(registerer prometheus.Registerer) *Cache {
+func newCacheMetrics(registerer prometheus.Registerer) cacheMetrics {
 	subsystem := "file_cache"
 	f := promauto.With(registerer)
 
-	c.currentCacheSize = f.NewGauge(prometheus.GaugeOpts{
-		Namespace: "coderd",
-		Subsystem: subsystem,
-		Name:      "open_files_size_bytes_current",
-		Help:      "The current amount of memory of all files currently open in the file cache.",
-	})
-
-	c.totalCacheSize = f.NewCounter(prometheus.CounterOpts{
-		Namespace: "coderd",
-		Subsystem: subsystem,
-		Name:      "open_files_size_bytes_total",
-		Help:      "The total amount of memory ever opened in the file cache. This number never decrements.",
-	})
-
-	c.currentOpenFiles = f.NewGauge(prometheus.GaugeOpts{
-		Namespace: "coderd",
-		Subsystem: subsystem,
-		Name:      "open_files_current",
-		Help:      "The count of unique files currently open in the file cache.",
-	})
-
-	c.totalOpenedFiles = f.NewCounter(prometheus.CounterOpts{
-		Namespace: "coderd",
-		Subsystem: subsystem,
-		Name:      "open_files_total",
-		Help:      "The total count of unique files ever opened in the file cache.",
-	})
-
-	c.currentOpenFileReferences = f.NewGauge(prometheus.GaugeOpts{
-		Namespace: "coderd",
-		Subsystem: subsystem,
-		Name:      "open_file_refs_current",
-		Help:      "The count of file references currently open in the file cache. Multiple references can be held for the same file.",
-	})
-
-	c.totalOpenFileReferences = f.NewCounterVec(prometheus.CounterOpts{
-		Namespace: "coderd",
-		Subsystem: subsystem,
-		Name:      "open_file_refs_total",
-		Help:      "The total number of file references ever opened in the file cache. The 'hit' label indicates if the file was loaded from the cache.",
-	}, []string{"hit"})
-
-	return c
+	return cacheMetrics{
+		currentCacheSize: f.NewGauge(prometheus.GaugeOpts{
+			Namespace: "coderd",
+			Subsystem: subsystem,
+			Name:      "open_files_size_bytes_current",
+			Help:      "The current amount of memory of all files currently open in the file cache.",
+		}),
+
+		totalCacheSize: f.NewCounter(prometheus.CounterOpts{
+			Namespace: "coderd",
+			Subsystem: subsystem,
+			Name:      "open_files_size_bytes_total",
+			Help:      "The total amount of memory ever opened in the file cache. This number never decrements.",
+		}),
+
+		currentOpenFiles: f.NewGauge(prometheus.GaugeOpts{
+			Namespace: "coderd",
+			Subsystem: subsystem,
+			Name:      "open_files_current",
+			Help:      "The count of unique files currently open in the file cache.",
+		}),
+
+		totalOpenedFiles: f.NewCounter(prometheus.CounterOpts{
+			Namespace: "coderd",
+			Subsystem: subsystem,
+			Name:      "open_files_total",
+			Help:      "The total count of unique files ever opened in the file cache.",
+		}),
+
+		currentOpenFileReferences: f.NewGauge(prometheus.GaugeOpts{
+			Namespace: "coderd",
+			Subsystem: subsystem,
+			Name:      "open_file_refs_current",
+			Help:      "The count of file references currently open in the file cache. Multiple references can be held for the same file.",
+		}),
+
+		totalOpenFileReferences: f.NewCounterVec(prometheus.CounterOpts{
+			Namespace: "coderd",
+			Subsystem: subsystem,
+			Name:      "open_file_refs_total",
+			Help:      "The total number of file references ever opened in the file cache. The 'hit' label indicates if the file was loaded from the cache.",
+		}, []string{"hit"}),
+	}
 }
 
 // Cache persists the files for template versions, and is used by dynamic
@@ -106,18 +107,22 @@ type cacheMetrics struct {
 	totalCacheSize   prometheus.Counter
 }
 
+type cacheEntry struct {
+	// refCount must only be accessed while the cacheEntry lock is held.
+	lock     sync.Mutex
+	refCount int
+	value    *lazy.ValueWithError[CacheEntryValue]
+
+	close func()
+	purge func()
+}
+
 type CacheEntryValue struct {
 	fs.FS
 	Object rbac.Object
 	Size   int64
 }
 
-type cacheEntry struct {
-	// refCount must only be accessed while the Cache lock is held.
-	refCount int
-	value    *lazy.ValueWithError[CacheEntryValue]
-}
-
 var _ fs.FS = (*CloseFS)(nil)
 
 // CloseFS is a wrapper around fs.FS that implements io.Closer. The Close()
@@ -142,93 +147,122 @@ func (c *Cache) Acquire(ctx context.Context, db database.Store, fileID uuid.UUID
 	// mutex has been released, or we would continue to hold the lock until the
 	// entire file has been fetched, which may be slow, and would prevent other
 	// files from being fetched in parallel.
-	it, err := c.prepare(ctx, db, fileID).Load()
+	e := c.prepare(db, fileID)
+	ev, err := e.value.Load()
 	if err != nil {
-		c.release(fileID)
+		e.close()
+		e.purge()
+		return nil, err
+	}
+
+	// We always run the fetch under a system context and actor, so we need to check the caller's
+	// context manually before returning.
+
+	// Check if the caller's context was canceled
+	if err := ctx.Err(); err != nil {
+		e.close()
 		return nil, err
 	}
 
+	// Check that the caller is authorized to access the file
 	subject, ok := dbauthz.ActorFromContext(ctx)
 	if !ok {
+		e.close()
 		return nil, dbauthz.ErrNoActor
 	}
-	// Always check the caller can actually read the file.
-	if err := c.authz.Authorize(ctx, subject, policy.ActionRead, it.Object); err != nil {
-		c.release(fileID)
+	if err := c.authz.Authorize(ctx, subject, policy.ActionRead, ev.Object); err != nil {
+		e.close()
 		return nil, err
 	}
 
-	var once sync.Once
+	var closeOnce sync.Once
 	return &CloseFS{
-		FS: it.FS,
+		FS: ev.FS,
 		close: func() {
 			// sync.Once makes the Close() idempotent, so we can call it
 			// multiple times without worrying about double-releasing.
-			once.Do(func() { c.release(fileID) })
+			closeOnce.Do(func() {
+				e.close()
+			})
 		},
 	}, nil
 }
 
-func (c *Cache) prepare(ctx context.Context, db database.Store, fileID uuid.UUID) *lazy.ValueWithError[CacheEntryValue] {
+func (c *Cache) prepare(db database.Store, fileID uuid.UUID) *cacheEntry {
 	c.lock.Lock()
 	defer c.lock.Unlock()
 
 	hitLabel := "true"
 	entry, ok := c.data[fileID]
 	if !ok {
-		value := lazy.NewWithError(func() (CacheEntryValue, error) {
-			val, err := fetch(ctx, db, fileID)
+		hitLabel = "false"
+
+		var purgeOnce sync.Once
+		entry = &cacheEntry{
+			refCount: 0,
+			value: lazy.NewWithError(func() (CacheEntryValue, error) {
+				val, err := fetch(db, fileID)
+				if err != nil {
+					return val, err
+				}
 
-			// Always add to the cache size the bytes of the file loaded.
-			if err == nil {
+				// Add the size of the file to the cache size metrics.
 				c.currentCacheSize.Add(float64(val.Size))
 				c.totalCacheSize.Add(float64(val.Size))
-			}
 
-			return val, err
-		})
+				return val, err
+			}),
 
-		entry = &cacheEntry{
-			value:    value,
-			refCount: 0,
+			close: func() {
+				entry.lock.Lock()
+				defer entry.lock.Unlock()
+
+				entry.refCount--
+				c.currentOpenFileReferences.Dec()
+				if entry.refCount > 0 {
+					return
+				}
+
+				entry.purge()
+			},
+
+			purge: func() {
+				purgeOnce.Do(func() {
+					c.purge(fileID)
+				})
+			},
 		}
 		c.data[fileID] = entry
+
 		c.currentOpenFiles.Inc()
 		c.totalOpenedFiles.Inc()
-		hitLabel = "false"
 	}
 
+	entry.lock.Lock()
+	defer entry.lock.Unlock()
 	c.currentOpenFileReferences.Inc()
 	c.totalOpenFileReferences.WithLabelValues(hitLabel).Inc()
 	entry.refCount++
-	return entry.value
+	return entry
 }
 
-// release decrements the reference count for the given fileID, and frees the
-// backing data if there are no further references being held.
-//
-// release should only be called after a successful call to Acquire using the Release()
-// method on the returned *CloseFS.
-func (c *Cache) release(fileID uuid.UUID) {
+// purge immediately removes an entry from the cache, even if it has open references.
+// It should only be called from the `close` function in a `cacheEntry`.
+func (c *Cache) purge(fileID uuid.UUID) {
 	c.lock.Lock()
 	defer c.lock.Unlock()
 
 	entry, ok := c.data[fileID]
 	if !ok {
-		// If we land here, it's almost certainly because a bug already happened,
-		// and we're freeing something that's already been freed, or we're calling
-		// this function with an incorrect ID. Should this function return an error?
-		return
-	}
-
-	c.currentOpenFileReferences.Dec()
-	entry.refCount--
-	if entry.refCount > 0 {
+		// If we land here, it's probably because of a fetch attempt that
+		// resulted in an error, and got purged already. It may also be an
+		// erroneous extra close, but we can't really distinguish between those
+		// two cases currently.
 		return
 	}
 
+	// Purge the file from the cache.
 	c.currentOpenFiles.Dec()
-
 	ev, err := entry.value.Load()
 	if err == nil {
 		c.currentCacheSize.Add(-1 * float64(ev.Size))
@@ -246,11 +280,18 @@ func (c *Cache) Count() int {
 	return len(c.data)
 }
 
-func fetch(ctx context.Context, store database.Store, fileID uuid.UUID) (CacheEntryValue, error) {
-	// Make sure the read does not fail due to authorization issues.
-	// Authz is checked on the Acquire call, so this is safe.
+func fetch(store database.Store, fileID uuid.UUID) (CacheEntryValue, error) {
+	// Because many callers can be waiting on the same file fetch concurrently, we
+	// want to prevent any failures that would cause them all to receive errors
+	// because the caller who initiated the fetch would fail.
+	// - We always run the fetch with an uncancelable context, and then check
+	//   context cancellation for each acquirer afterwards.
+	// - We always run the fetch as a system user, and then check authorization
+	//   for each acquirer afterwards.
+	// This prevents a canceled context or an unauthorized user from "holding up
+	// the queue".
 	//nolint:gocritic
-	file, err := store.GetFileByID(dbauthz.AsFileReader(ctx), fileID)
+	file, err := store.GetFileByID(dbauthz.AsFileReader(context.Background()), fileID)
 	if err != nil {
 		return CacheEntryValue{}, xerrors.Errorf("failed to read file from database: %w", err)
 	}

diff --git a/coderd/files/cache_test.go b/coderd/files/cache_test.go
@@ -8,6 +8,7 @@ import (
 
 	"github.com/google/uuid"
 	"github.com/prometheus/client_golang/prometheus"
+	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 	"go.uber.org/mock/gomock"
 	"golang.org/x/sync/errgroup"
@@ -26,6 +27,46 @@ import (
 	"github.com/coder/coder/v2/testutil"
 )
 
+// TestCancelledFetch runs 2 Acquire calls. The first fails with a ctx.Canceled
+// error. The second call should ignore the first error and try to fetch the file
+// again, which should succeed.
+func TestCancelledFetch(t *testing.T) {
+	t.Parallel()
+
+	fileID := uuid.New()
+	dbM := dbmock.NewMockStore(gomock.NewController(t))
+
+	// First call should fail
+	dbM.EXPECT().GetFileByID(gomock.Any(), gomock.Any()).DoAndReturn(func(mTx context.Context, fileID uuid.UUID) (database.File, error) {
+		// Wait long enough for the second call to be queued up.
+		return database.File{}, context.Canceled
+	})
+
+	// Second call should succeed
+	dbM.EXPECT().GetFileByID(gomock.Any(), gomock.Any()).DoAndReturn(func(mTx context.Context, fileID uuid.UUID) (database.File, error) {
+		return database.File{
+			ID:   fileID,
+			Data: make([]byte, 100),
+		}, nil
+	})
+
+	//nolint:gocritic // Unit testing
+	ctx := dbauthz.AsFileReader(testutil.Context(t, testutil.WaitShort))
+	cache := files.New(prometheus.NewRegistry(), &coderdtest.FakeAuthorizer{})
+
+	// First call that will fail
+	_, err := cache.Acquire(ctx, dbM, fileID)
+	assert.ErrorIs(t, err, context.Canceled)
+
+	// Second call, that should succeed
+	// Wait until the first goroutine has started
+	fs, err := cache.Acquire(ctx, dbM, fileID)
+	assert.NoError(t, err)
+	if fs != nil {
+		fs.Close()
+	}
+}
+
 // nolint:paralleltest,tparallel // Serially testing is easier
 func TestCacheRBAC(t *testing.T) {
 	t.Parallel()