diff --git a/pool.go b/pool.go
index acbcc33..a413e4d 100644
--- a/pool.go
+++ b/pool.go
@@ -311,32 +311,79 @@ func (p *Pool[T]) Acquire(ctx context.Context) (*Resource[T], error) {
 			p.destructWG.Add(1)
 			p.cond.L.Unlock()
 
-			value, err := p.constructResourceValue(ctx)
-			p.cond.L.Lock()
-			if err != nil {
-				p.allResources = removeResource(p.allResources, res)
-				p.destructWG.Done()
+			// we create the resource in the background because the constructor might
+			// outlive the context and we want to continue constructing it as long as
+			// necessary but the acquire should be cancelled when the context is cancelled
+			// see: https://github.com/jackc/pgx/issues/1287 and https://github.com/jackc/pgx/issues/1259
+			constructErrCh := make(chan error)
+			go func() {
+				value, err := p.constructResourceValue(ctx)
+				p.cond.L.Lock()
+				if err != nil {
+					p.allResources = removeResource(p.allResources, res)
+					p.destructWG.Done()
 
+					select {
+					case <-ctx.Done():
+						if err == ctx.Err() {
+							p.canceledAcquireCount += 1
+						}
+					default:
+					}
+
+					p.cond.L.Unlock()
+					p.cond.Signal()
+
+					// try to notify the caller that we failed
+					select {
+					case constructErrCh <- err:
+					default:
+					}
+					return
+				}
+				res.value = value
+
+				// check the context now so we don't increment the metrics when the caller
+				// has already been cancelled
 				select {
 				case <-ctx.Done():
-					if err == ctx.Err() {
-						p.canceledAcquireCount += 1
-					}
+					p.cond.L.Unlock()
 				default:
+					// assume that we will acquire it
+					res.status = resourceStatusAcquired
+					// we have to increment these BEFORE the next select because otherwise
+					// they could run after Acquire has returned and that will mess up
+					// tests but this also means that these could be incremented even if
+					// the acquire times out, but we just checked that so the chances are
+					// slim
+					p.emptyAcquireCount += 1
+					p.acquireCount += 1
+					p.acquireDuration += time.Duration(nanotime() - startNano)
+					p.cond.L.Unlock()
+					// we don't call Signal here we didn't change any of the resopurce pools
 				}
 
-				p.cond.L.Unlock()
-				p.cond.Signal()
-				return nil, err
-			}
+				select {
+				case constructErrCh <- nil:
+				default:
+					// since we couldn't send the constructed resource to the acquire
+					// function that means the caller has stopped waiting and we should
+					// just put this resource back in the pool
+					p.releaseAcquiredResource(res, res.lastUsedNano)
+				}
+			}()
 
-			res.value = value
-			res.status = resourceStatusAcquired
-			p.emptyAcquireCount += 1
-			p.acquireCount += 1
-			p.acquireDuration += time.Duration(nanotime() - startNano)
-			p.cond.L.Unlock()
-			return res, nil
+			select {
+			case <-ctx.Done():
+				return nil, ctx.Err()
+			case err := <-constructErrCh:
+				if err != nil {
+					return nil, err
+				}
+				// we don't call signal here because we didn't change the resource pools
+				// at all so waking anything else up won't help
+				return res, nil
+			}
 		}
 
 		if ctx.Done() == nil {
@@ -355,8 +402,8 @@ func (p *Pool[T]) Acquire(ctx context.Context) (*Resource[T], error) {
 				// do anything with it. Another goroutine might be waiting.
 				go func() {
 					<-waitChan
-					p.cond.Signal()
 					p.cond.L.Unlock()
+					p.cond.Signal()
 				}()
 
 				p.cond.L.Lock()
diff --git a/pool_test.go b/pool_test.go
index 5a0f645..d7e82b7 100644
--- a/pool_test.go
+++ b/pool_test.go
@@ -133,6 +133,32 @@ func TestPoolAcquireReturnsErrorFromFailedResourceCreate(t *testing.T) {
 	assert.Nil(t, res)
 }
 
+func TestPoolAcquireCreatesResourceRespectingContext(t *testing.T) {
+	var cancel func()
+	constructor := func(ctx context.Context) (int, error) {
+		cancel()
+		// sleep to give a chance for the acquire to recognize it's cancelled
+		time.Sleep(10 * time.Millisecond)
+		return 1, nil
+	}
+	pool := puddle.NewPool(constructor, stubDestructor, 1)
+	defer pool.Close()
+
+	var ctx context.Context
+	ctx, cancel = context.WithCancel(context.Background())
+	defer cancel()
+	_, err := pool.Acquire(ctx)
+	assert.ErrorIs(t, err, context.Canceled)
+
+	// wait for the constructor to sleep and then for the resource to be added back
+	// to the idle pool
+	time.Sleep(100 * time.Millisecond)
+
+	stat := pool.Stat()
+	assert.EqualValues(t, 1, stat.IdleResources())
+	assert.EqualValues(t, 1, stat.TotalResources())
+}
+
 func TestPoolAcquireReusesResources(t *testing.T) {
 	constructor, createCounter := createConstructor()
 	pool := puddle.NewPool(constructor, stubDestructor, 10)
@@ -544,7 +570,12 @@ func TestPoolStatResources(t *testing.T) {
 
 func TestPoolStatSuccessfulAcquireCounters(t *testing.T) {
 	constructor, _ := createConstructor()
-	pool := puddle.NewPool(constructor, stubDestructor, 1)
+	sleepConstructor := func(ctx context.Context) (int, error) {
+		// sleep to make sure we don't fail the AcquireDuration test
+		time.Sleep(time.Nanosecond)
+		return constructor(ctx)
+	}
+	pool := puddle.NewPool(sleepConstructor, stubDestructor, 1)
 	defer pool.Close()
 
 	res, err := pool.Acquire(context.Background())