Cache and reuse the scraper HTTP client (#1855)

* Add Cookies directly to the request

Rather than maintaining a cookie jar on a one-shot HTTP client, maintain
the jar ourselves: make a new jar, then use it to select the right
cookies.

The cookies are set on the request rather than on the client. This will
retain the current behavior as we are always throwing the client away
after each use.

This patch enables the lifting of the http client as well over time.

* Introduce a cached scraper HTTP client

The scraper cache is augmented with an *http.Client. These are safe for
concurrent use, so the pointer can safely be passed around. Push this
into scraper configurations where applicable, next to the txnManagers.

When we issue a loadUrl request, do so on the cached *http.Client,
which will reuse existing idle connections in the client if any are
present.

* Set MaxIdleConnsPerHost. Closes #1850

We allow for up to 8 idle connections to a single host. This should
make concurrent operation toward the same host reuse connections, even
for sizeable concurrency.

The number isn't bumped excessively high. We should probably limit
concurrency toward a single site anyway, since we'll be able to overrun
a site with queries quite easily if we have many concurrent goroutines
issuing requests at the same time.

* Reinstate driverOptions / useCDP check

Use DeMorgan's laws to invert the logic and exit early. Fixes tests
breaking.

* Documentation fixup.

* Use the scraper http.Client when fetching images

Fold image fetchers onto the cached scraper http.Client as well. This
makes the scraper have a single http.Client cache for all its
operations.

Thread the client upwards to the relevant attachment points: either the
cache, or a stash_box instance, which is extended to include a pointer
to the client.

Style roughly follows that of txnManagers.

* Use the same http Client as the GraphQL client use

Rather than using http.DefaultClient, use the same client as the
GraphQL client use in the stash_box subsystem. This localizes the
client used in the subsystem into the constructing New.. call.

* Hoist HTTP client construction

Create a function for initializaing the HTTP Client we use. While here
hoist magic numbers into constants. Introduce a proper static redirect
error and use it in the client code as well.

* Reinstate printCookies

This is a debugging function, and it might still come in handy in the
future at some point.

* Nitpick comment.

* Minor tidy

Co-authored-by: WithoutPants <53250216+WithoutPants@users.noreply.github.com>

pkg/scraper/stashbox/stash_box.go

@@ -7,7 +7,6 @@ import (
 	"net/http"
 	"strconv"
 	"strings"
-	"time"
 
 	"github.com/Yamashou/gqlgenc/client"
 
@@ -18,10 +17,6 @@ import (
 	"github.com/stashapp/stash/pkg/utils"
 )
 
-// Timeout to get the image. Includes transfer time. May want to make this
-// configurable at some point.
-const imageGetTimeout = time.Second * 30
-
 // Client represents the client interface to a stash-box server instance.
 type Client struct {
 	client     *graphql.Client
@@ -44,6 +39,10 @@ func NewClient(box models.StashBox, txnManager models.TransactionManager) *Clien
 	}
 }
 
+func (c Client) getHTTPClient() *http.Client {
+	return c.client.Client.Client
+}
+
 // QueryStashBoxScene queries stash-box for scenes using a query string.
 func (c Client) QueryStashBoxScene(ctx context.Context, queryStr string) ([]*models.ScrapedScene, error) {
 	scenes, err := c.client.SearchScene(ctx, queryStr)
@@ -55,7 +54,7 @@ func (c Client) QueryStashBoxScene(ctx context.Context, queryStr string) ([]*mod
 
 	var ret []*models.ScrapedScene
 	for _, s := range sceneFragments {
-		ss, err := sceneFragmentToScrapedScene(context.TODO(), c.txnManager, s)
+		ss, err := sceneFragmentToScrapedScene(context.TODO(), c.getHTTPClient(), c.txnManager, s)
 		if err != nil {
 			return nil, err
 		}
@@ -201,7 +200,7 @@ func (c Client) findStashBoxScenesByFingerprints(ctx context.Context, fingerprin
 		sceneFragments := scenes.FindScenesByFingerprints
 
 		for _, s := range sceneFragments {
-			ss, err := sceneFragmentToScrapedScene(ctx, c.txnManager, s)
+			ss, err := sceneFragmentToScrapedScene(ctx, c.getHTTPClient(), c.txnManager, s)
 			if err != nil {
 				return nil, err
 			}
@@ -509,11 +508,7 @@ func formatBodyModifications(m []*graphql.BodyModificationFragment) *string {
 	return &ret
 }
 
-func fetchImage(ctx context.Context, url string) (*string, error) {
-	client := &http.Client{
-		Timeout: imageGetTimeout,
-	}
-
+func fetchImage(ctx context.Context, client *http.Client, url string) (*string, error) {
 	req, err := http.NewRequestWithContext(ctx, http.MethodGet, url, nil)
 	if err != nil {
 		return nil, err
@@ -595,8 +590,8 @@ func performerFragmentToScrapedScenePerformer(p graphql.PerformerFragment) *mode
 	return sp
 }
 
-func getFirstImage(ctx context.Context, images []*graphql.ImageFragment) *string {
-	ret, err := fetchImage(ctx, images[0].URL)
+func getFirstImage(ctx context.Context, client *http.Client, images []*graphql.ImageFragment) *string {
+	ret, err := fetchImage(ctx, client, images[0].URL)
 	if err != nil {
 		logger.Warnf("Error fetching image %s: %s", images[0].URL, err.Error())
 	}
@@ -617,7 +612,7 @@ func getFingerprints(scene *graphql.SceneFragment) []*models.StashBoxFingerprint
 	return fingerprints
 }
 
-func sceneFragmentToScrapedScene(ctx context.Context, txnManager models.TransactionManager, s *graphql.SceneFragment) (*models.ScrapedScene, error) {
+func sceneFragmentToScrapedScene(ctx context.Context, client *http.Client, txnManager models.TransactionManager, s *graphql.SceneFragment) (*models.ScrapedScene, error) {
 	stashID := s.ID
 	ss := &models.ScrapedScene{
 		Title:        s.Title,
@@ -634,7 +629,7 @@ func sceneFragmentToScrapedScene(ctx context.Context, txnManager models.Transact
 	if len(s.Images) > 0 {
 		// TODO - #454 code sorts images by aspect ratio according to a wanted
 		// orientation. I'm just grabbing the first for now
-		ss.Image = getFirstImage(ctx, s.Images)
+		ss.Image = getFirstImage(ctx, client, s.Images)
 	}
 
 	if err := txnManager.WithReadTxn(ctx, func(r models.ReaderRepository) error {