buffer: fix Blob.stream() leaking source buffer

Blob.prototype.stream() registered a wakeup callback on the
underlying source's start() and never released it. The strong
Reader::wakeup_ handle kept the reader -- and through it the
blob's DataQueue and backing store -- reachable as a GC root, so
the source buffer leaked on every stream() call. On Node 26+,
streaming a 1 MiB blob 300 times retained ~300 MiB in
process.memoryUsage().arrayBuffers while the V8 heap stayed small.

Register the wakeup lazily in pull() and clear it on every terminal
or idle path (EOS, error, cancel, backpressure), mirroring the
cleanup already done by the async iterator path. The strong handle
now only lives while a pull is in flight, so the reader and its
backing store become collectable once the stream finishes, errors,
is cancelled, or goes idle under backpressure.

Fixes: #63574

Signed-off-by: semimikoh <ejffjeosms@gmail.com>
PR-URL: #63577
Fixes: #63574
Reviewed-By: James M Snell <jasnell@gmail.com>

Author: semimikoh

lib/internal/blob.js

@@ -473,30 +473,40 @@ function createBlobReaderStream(reader) {
       // There really should only be one read at a time so using an
       // array here is purely defensive.
       this.pendingPulls = [];
-      // Register a wakeup callback that the C++ side can invoke
+      // Lazily register a wakeup callback that the C++ side can invoke
       // when new data is available after a STATUS_BLOCK.
-      reader.setWakeup(() => {
+      this.wakeup = () => {
         if (this.pendingPulls.length > 0) {
           this.readNext(c);
         }
-      });
+      };
     },
     pull(c) {
       const { promise, resolve, reject } = PromiseWithResolvers();
+      if (this.pendingPulls.length === 0) {
+        reader.setWakeup(this.wakeup);
+      }
       this.pendingPulls.push({ resolve, reject });
       this.readNext(c);
       return promise;
     },
+    clearWakeupIfIdle() {
+      if (this.pendingPulls.length === 0) {
+        reader.setWakeup(undefined);
+      }
+    },
     readNext(c) {
       reader.pull((status, buffer) => {
         // If pendingPulls is empty here, the stream had to have
         // been canceled, and we don't really care about the result.
         // We can simply exit.
         if (this.pendingPulls.length === 0) {
+          reader.setWakeup(undefined);
           return;
         }
         if (status === 0) {
           // EOS
+          reader.setWakeup(undefined);
           c.close();
           // This is to signal the end for byob readers
           // see https://streams.spec.whatwg.org/#example-rbs-pull
@@ -508,6 +518,7 @@ function createBlobReaderStream(reader) {
           // The read could fail for many different reasons when reading
           // from a non-memory resident blob part (e.g. file-backed blob).
           // The error details the system error code.
+          reader.setWakeup(undefined);
           const error =
             lazyDOMException('The blob could not be read',
                              'NotReadableError');
@@ -517,7 +528,7 @@ function createBlobReaderStream(reader) {
           return;
         } else if (status === 2) {
           // STATUS_BLOCK: No data available yet. The wakeup callback
-          // registered in start() will re-invoke readNext when data
+          // registered in pull() will re-invoke readNext when data
           // arrives.
           return;
         }
@@ -537,6 +548,7 @@ function createBlobReaderStream(reader) {
             if (this.pendingPulls.length !== 0) {
               const pending = this.pendingPulls.shift();
               pending.resolve();
+              this.clearWakeupIfIdle();
             }
             return;
           }
@@ -545,6 +557,7 @@ function createBlobReaderStream(reader) {
       });
     },
     cancel(reason) {
+      reader.setWakeup(undefined);
       // Reject any currently pending pulls here.
       for (const pending of this.pendingPulls) {
         pending.reject(reason);

test/parallel/test-blob-stream-gc.js

@@ -0,0 +1,56 @@
+// Flags: --expose-gc --no-concurrent-array-buffer-sweeping
+'use strict';
+
+const common = require('../common');
+const assert = require('assert');
+const { setImmediate: setImmediatePromise } = require('timers/promises');
+
+const MiB = 1024 * 1024;
+const iterations = 64;
+const maxRetained = 16 * MiB;
+
+async function collectArrayBuffers() {
+  for (let i = 0; i < 3; i++) {
+    global.gc();
+    await setImmediatePromise();
+  }
+}
+
+async function assertNoBlobStreamRetention(name, fn) {
+  const buffer = Buffer.alloc(MiB);
+
+  await collectArrayBuffers();
+  const before = process.memoryUsage().arrayBuffers;
+
+  for (let i = 0; i < iterations; i++) {
+    await fn(buffer);
+  }
+
+  await collectArrayBuffers();
+  const retained = process.memoryUsage().arrayBuffers - before;
+
+  assert(
+    retained < maxRetained,
+    `${name} retained ${retained} bytes in arrayBuffers`,
+  );
+}
+
+(async () => {
+  await assertNoBlobStreamRetention('unused Blob streams',
+                                    common.mustCall(async (buffer) => {
+                                      new Blob([buffer]).stream();
+                                    }, iterations));
+
+  await assertNoBlobStreamRetention('cancelled Blob streams',
+                                    common.mustCall(async (buffer) => {
+                                      await new Blob([buffer]).stream()
+                                        .cancel();
+                                    }, iterations));
+
+  await assertNoBlobStreamRetention('drained Blob streams',
+                                    common.mustCall(async (buffer) => {
+                                      await new Response(
+                                        new Blob([buffer]).stream(),
+                                      ).arrayBuffer();
+                                    }, iterations));
+})().then(common.mustCall());