[FrameworkBundle] Improve performance of ControllerNameParser #20374

enumag · 2016-11-01T12:03:48Z

Q	A
Branch?	2.7
Bug fix?	yes
New feature?	no
BC breaks?	no
Deprecations?	no
Tests pass?	yes
Fixed tickets
License	MIT
Doc PR

Today I was searching for bottlenecks in my application using Blackfire. And among other things I found one in Symfony. Blackfire showed that Symfony\Bundle\FrameworkBundle\Controller\ControllerNameParser::findAlternative() was called almost 300 times which took 28 miliseconds.

It turns out that Symfony\Bundle\FrameworkBundle\Routing\DelegatingLoader::load() is calling ControllerNameParser::parse() without actually needing to do so because $controller is in the class::method notation already. ControllerNameParser threw an exception, DelegatingLoader caught and ignored it - that's ok. The problem is that generating the exception message took a lot of time because findAlternative is slow. In my case it called the levenshtein function over 5000 times which was completely useless because the exception is ignored anyway.

stloyd · 2016-11-01T12:19:50Z

src/Symfony/Bundle/FrameworkBundle/Controller/ControllerNameParser.php

+        // The second condition is for cases when this method is called with the
+        // class::method string instead of a:b:c. It's important to fail early in this
+        // case because calling findAlternative needlessly is bad for performance.
+        if (3 !== count($parts) || $parts[1] === '') {


Second check would be faster so IMO should be first ;)

Can't be first because $parts[1] might not exist.

Perhaps check [0] and [2] as well for emptiness? As those are also not a valid "a:b:c" controller string.

we prefer yoda style: '' === $parts[1]

nicolas-grekas · 2016-11-02T07:22:21Z

Just wondering: would it make sense to add this check in DelegatingLoader instead?
Something like:

--- a/src/Symfony/Bundle/FrameworkBundle/Routing/DelegatingLoader.php
+++ b/src/Symfony/Bundle/FrameworkBundle/Routing/DelegatingLoader.php
@@ -21,7 +21,7 @@ use Psr\Log\LoggerInterface;
  * DelegatingLoader delegates route loading to other loaders using a loader resolver.
  *
  * This implementation resolves the _controller attribute from the short notation
- * to the fully-qualified form (from a:b:c to class:method).
+ * to the fully-qualified form (from a:b:c to class::method).
  *
  * @author Fabien Potencier <[email protected]>
  */
@@ -85,7 +85,8 @@ class DelegatingLoader extends BaseDelegatingLoader
         $this->loading = false;

         foreach ($collection->all() as $route) {
-            if ($controller = $route->getDefault('_controller')) {
+            $controller = $route->getDefault('_controller');
+            if ($controller && false === strpos($controller, '::')) {
                 try {
                     $controller = $this->parser->parse($controller);
                 } catch (\InvalidArgumentException $e) {

enumag · 2016-11-02T08:32:55Z

@nicolas-grekas I can add that as well but the change in ControllerNameParser should be done regardless in my opinion.

enumag · 2016-11-05T08:44:28Z

@nicolas-grekas @ro0NL All fixed.

nicolas-grekas · 2016-11-05T09:42:43Z

👍

xabbuh · 2016-11-05T11:09:22Z

👍

ro0NL

👍

ro0NL · 2016-11-05T11:17:45Z

src/Symfony/Bundle/FrameworkBundle/Controller/ControllerNameParser.php

-        if (3 !== count($parts = explode(':', $controller))) {
+        $parts = explode(':', $controller);
+
+        // The second condition is for cases when this method is called with the


I think the exception message is already straightforward now? Imo. no need for this comment.

ro0NL · 2016-11-05T11:18:09Z

src/Symfony/Bundle/FrameworkBundle/Routing/DelegatingLoader.php

@@ -85,7 +85,8 @@ public function load($resource, $type = null)
        $this->loading = false;

        foreach ($collection->all() as $route) {
-            if ($controller = $route->getDefault('_controller')) {
+            $controller = $route->getDefault('_controller');
+            if ($controller && false === strpos($controller, '::')) {


Instead, maybe add a comment here about skipping FQCN's for performance.

enumag · 2016-11-06T07:11:47Z

@ro0NL Like this?

dunglas · 2016-11-10T15:35:08Z

src/Symfony/Bundle/FrameworkBundle/Routing/DelegatingLoader.php

-            if ($controller = $route->getDefault('_controller')) {
+            $controller = $route->getDefault('_controller');
+            // If $controller is in class::method notation already skip it for performance.
+            if ($controller && false === strpos($controller, '::')) {


What do you think about using a guard clause here to reduce the code's complexity?

if (!$controller || false !== strpos($controller, '::')) { continue; }

@dunglas Good idea. I'll change it.

dunglas · 2016-11-10T17:06:04Z

👍

enumag · 2016-11-19T10:28:54Z

I didn't notice the coding style error before. It's fixed now. Anything else?

nicolas-grekas · 2016-11-25T11:07:15Z

src/Symfony/Bundle/FrameworkBundle/Routing/DelegatingLoader.php

-                    // unable to optimize unknown notation
-                }
+            $controller = $route->getDefault('_controller');
+            // If $controller is in class::method notation already skip it for performance.


I know I suggested this change, but looking at it again, there is a potential BC break here since we have a new case where we won't call the parser. Let's be safe and revert (still keep the typo fix above). Most of the perf benefit in this PR doesn't come from this part of the patch.

@nicolas-grekas Understood, done.

stof · 2016-11-25T11:41:26Z

👍 for this

And should we apply it to older branches too ? We did it in the past for some performance improvements.

On a side note, this will not improve the performance in prod, except for the cache warmup, as the routing loader is not called once anymore the matcher and generator are cached.

nicolas-grekas · 2016-11-25T12:12:23Z

Thank you @enumag.

…ser (enumag) This PR was merged into the 2.7 branch. Discussion ---------- [FrameworkBundle] Improve performance of ControllerNameParser | Q | A | ------------- | --- | Branch? | 2.7 | Bug fix? | yes | New feature? | no | BC breaks? | no | Deprecations? | no | Tests pass? | yes | Fixed tickets | | License | MIT | Doc PR | Today I was searching for bottlenecks in my application using Blackfire. And among other things I found one in Symfony. Blackfire showed that `Symfony\Bundle\FrameworkBundle\Controller\ControllerNameParser::findAlternative()` was called almost 300 times which took 28 miliseconds. It turns out that `Symfony\Bundle\FrameworkBundle\Routing\DelegatingLoader::load()` is calling `ControllerNameParser::parse()` without actually needing to do so because `$controller` is in the class::method notation already. `ControllerNameParser` threw an exception, DelegatingLoader caught and ignored it - that's ok. The problem is that generating the exception message took a lot of time because findAlternative is slow. In my case it called the levenshtein function over 5000 times which was completely useless because the exception is ignored anyway. Commits ------- cf333f3 [FrameworkBundle] Improve performance of ControllerNameParser

fabpot · 2016-11-25T18:33:49Z

This one should not have been merged in 2.7. It does not fix a bug. This should have been merged in master instead (not even 3.2 at this stage).

carsonbot added Status: Needs Review FrameworkBundle Bug labels Nov 1, 2016

stloyd suggested changes Nov 1, 2016

View reviewed changes

enumag force-pushed the patch-18 branch from f9b9774 to ac284c3 Compare November 5, 2016 08:43

ro0NL reviewed Nov 5, 2016

View reviewed changes

enumag force-pushed the patch-18 branch from ac284c3 to 08cc54f Compare November 6, 2016 07:11

javiereguiluz added this to the 3.2 milestone Nov 7, 2016

dunglas reviewed Nov 10, 2016

View reviewed changes

enumag force-pushed the patch-18 branch from 08cc54f to e571f99 Compare November 10, 2016 21:56

fabpot removed this from the 3.2 milestone Nov 16, 2016

enumag force-pushed the patch-18 branch from e571f99 to a588222 Compare November 19, 2016 10:24

nicolas-grekas added Status: Reviewed and removed Status: Needs Review labels Nov 24, 2016

nicolas-grekas reviewed Nov 25, 2016

View reviewed changes

[FrameworkBundle] Improve performance of ControllerNameParser

cf333f3

enumag force-pushed the patch-18 branch from a588222 to cf333f3 Compare November 25, 2016 11:34

nicolas-grekas closed this Nov 25, 2016

fabpot mentioned this pull request Nov 27, 2016

Release v3.2.0-RC2 #20649

Merged

This was referenced Dec 13, 2016

Release v2.7.22 #20890

Merged

Release v2.8.15 #20892

Merged

Release v3.1.8 #20895

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[FrameworkBundle] Improve performance of ControllerNameParser #20374

[FrameworkBundle] Improve performance of ControllerNameParser #20374

enumag commented Nov 1, 2016 •

edited

Loading

stloyd Nov 1, 2016

enumag Nov 1, 2016

ro0NL Nov 1, 2016 •

edited

Loading

nicolas-grekas Nov 2, 2016

nicolas-grekas commented Nov 2, 2016

enumag commented Nov 2, 2016

enumag commented Nov 5, 2016

nicolas-grekas commented Nov 5, 2016

xabbuh commented Nov 5, 2016

ro0NL left a comment

ro0NL Nov 5, 2016

ro0NL Nov 5, 2016

enumag commented Nov 6, 2016

dunglas Nov 10, 2016 •

edited

Loading

enumag Nov 10, 2016

enumag Nov 10, 2016

dunglas commented Nov 10, 2016

enumag commented Nov 19, 2016 •

edited

Loading

nicolas-grekas Nov 25, 2016

enumag Nov 25, 2016

stof commented Nov 25, 2016

nicolas-grekas commented Nov 25, 2016

fabpot commented Nov 25, 2016

[FrameworkBundle] Improve performance of ControllerNameParser #20374

[FrameworkBundle] Improve performance of ControllerNameParser #20374

Conversation

enumag commented Nov 1, 2016 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ro0NL Nov 1, 2016 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

nicolas-grekas commented Nov 2, 2016

enumag commented Nov 2, 2016

enumag commented Nov 5, 2016

nicolas-grekas commented Nov 5, 2016

xabbuh commented Nov 5, 2016

ro0NL left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

enumag commented Nov 6, 2016

dunglas Nov 10, 2016 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dunglas commented Nov 10, 2016

enumag commented Nov 19, 2016 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

stof commented Nov 25, 2016

nicolas-grekas commented Nov 25, 2016

fabpot commented Nov 25, 2016

enumag commented Nov 1, 2016 •

edited

Loading

ro0NL Nov 1, 2016 •

edited

Loading

dunglas Nov 10, 2016 •

edited

Loading

enumag commented Nov 19, 2016 •

edited

Loading