[HttpFoundation] Fix DbalSessionHandler #10720

Tobion · 2014-04-16T15:19:49Z

Q	A
Bug fix?	yes
New feature?	no
BC breaks?	no
Deprecations?	no
Tests pass?	yes
Fixed tickets	-
License	MIT

This is basically the same as #10652 for the DbalSessionHandler.

First commit fixes fix DbalSessionHandler for high concurrency, interface compliance, compatibility with all drivers (oci8, mysqli, pdo with mysql, sqlsrv, sqlite).
Second commit updates phpdoc of SessionHandlerInterface and unifies parameters of all handlers according to interface (so inheritdoc actually makes sense).

Tobion · 2014-04-16T15:23:36Z

src/Symfony/Bridge/Doctrine/HttpFoundation/DbalSessionHandler.php

        try {
-            $this->con->executeQuery("DELETE FROM {$this->tableName} WHERE sess_id = :id", array(


executeQuery is not part of Doctrine\DBAL\Driver\Connection (only in Doctrine\DBAL\Connection) so this wasn't working without wrapper

the expectation of the class was actually Doctrine\DBAL\Connection, which is the public facade of Doctrine.
Given the class is throwing a fatal error if usign the internal connection instead of the public class anyway, we can fix the typehint without breaking BC. It will simplify your implementation a lot

Maybe, but typehinting the interface instead of the wrapper makes sense. And I already found a solution now which is only relevant for the MERGE SQL.

not in this case. Doctrine\DBAL\Driver\Connection is the interface for internal implementations for each driver. Doctrine\DBAL\Connection is the public API of DBAL. The fact that Doctrine\DBAL\Connection implements Doctrine\DBAL\Driver\Connection is a legacy mistake which cannot be fixed until Doctrine 3.0 because of BC. It totally makes sense to depend on the public API rather than on the internal one

In that case the doctrine phpdoc needs an update. Driver connection should be marked as internal and the description of Connection is also irritating

A wrapper around a Doctrine\DBAL\Driver\Connection that adds features like events, ...

agreed. reported as http://www.doctrine-project.org/jira/browse/DBAL-867

Btw, Symfony\Component\Security\Acl\Dbal\AclProvider and Symfony\Component\Security\Acl\Dbal\MutableAclProvider also use the wrong connection typehint.

Fixed in #10723

…ace compliance, compatibility with all drivers (oci8, mysqli, pdo with mysql, sqlsrv, sqlite)

…arameters of all handlers according to interface

fabpot · 2014-04-16T16:17:57Z

src/Symfony/Bridge/Doctrine/HttpFoundation/DbalSessionHandler.php

+                    "WHEN MATCHED THEN UPDATE SET $this->dataCol = :data;";
+            case $platform instanceof SqlitePlatform || 'sqlite' === $pdoDriver:
+                return "INSERT OR REPLACE INTO $this->table ($this->idCol, $this->dataCol, $this->timeCol) VALUES (:id, :data, :time)";
+        }


PostgreSQL is missing here... like in the other PR but I totally forgot to mention it back then. The problem is that PostgreSQL does not support upserts but there are some workarounds.

see https://github.com/mitsuhiko/lucumr/blob/master/2014/2/16/a-case-for-upserts.rst for more info.

Yes I already read that when researching for upserts in each database engine. And this article says that postgresql does not have upserts and wishes there was one. So I'm not sure what you expect here?
The workaround, that I already implemented as fallback, is to use DELETE -> INSERT

You could use a custom database function, but that will require someone manual work trough.
http://koo.fi/blog/2014/01/04/upsert-methods-for-postgresql/

For PostgreSQL 9.1 you can use a writable CTE.
http://stackoverflow.com/questions/1109061/insert-on-duplicate-update-in-postgresql/8702291#8702291

I wonder if checking if the id exists is faster then deleting it and then directly inserting it.
Knowing that id is unique (indexed) it should not be to slow.

Edit. To bad there is no clear answer on the performance question.
http://stackoverflow.com/questions/4901475/using-postgres-sql-what-is-the-fastest-transaction-to-emulate-an-upsert-for-a-s

Edit2. Ah wait! you can update and then check the affected-rows count to see if any changes have happened.
But its vulnerable for race conditions. Is this really an issue for the session-id?

update -> affected rows = 0 -> insert does not work well when the row did not change (two simultaneoes writes with same data) or. then the insert will create duplicate key again. it's also not safe against race conditions in high concurrency (even in serializable isolation level, mysql will throw a deadlock when using two such parallel transactions, same for delete -> insert though)
one could also catch exceptions on duplicate key and ignore them. but it's not easily possible to do this correcly for all databases because each one has different subclasses of 23xxx sql state.

@Tobion DBAL 2.5 is throwing a specific exception for constraint violations, so this will help

http://www.depesz.com/2012/06/10/why-is-upsert-so-complicated/ is an importand read for postgresql.

@stof but there are many different types of constraint violations and we only need to ignore one of it

fabpot · 2014-04-18T20:55:06Z

Thank you @Tobion.

This PR was merged into the 2.3 branch. Discussion ---------- [HttpFoundation] Fix DbalSessionHandler | Q | A | ------------- | --- | Bug fix? | yes | New feature? | no | BC breaks? | no | Deprecations? | no | Tests pass? | yes | Fixed tickets | - | License | MIT This is basically the same as #10652 for the DbalSessionHandler. - First commit fixes fix DbalSessionHandler for high concurrency, interface compliance, compatibility with all drivers (oci8, mysqli, pdo with mysql, sqlsrv, sqlite). - Second commit updates phpdoc of SessionHandlerInterface and unifies parameters of all handlers according to interface (so inheritdoc actually makes sense). Commits ------- 524bf84 [HttpFoundation] update phpdoc of SessionHandlerInterface and unify parameters of all handlers according to interface ccdfbe6 [Doctrine Bridge] fix DbalSessionHandler for high concurrency, interface compliance, compatibility with all drivers (oci8, mysqli, pdo with mysql, sqlsrv, sqlite)

This PR was merged into the 2.3 branch. Discussion ---------- [Doctrine Bridge] simplify session handler | Q | A | ------------- | --- | Bug fix? | yes | New feature? | no | BC breaks? | no because the driver connection could not be used anyway before | Deprecations? | no | Tests pass? | yes | Fixed tickets | - | License | MIT Use main connection as discussed in #10720 Commits ------- 6518b74 [Doctrine Bridge] simplify session handler by using main connection

Tobion reviewed Apr 16, 2014
View reviewed changes

Tobion added 2 commits April 16, 2014 17:38

[Doctrine Bridge] fix DbalSessionHandler for high concurrency, interf…

ccdfbe6

…ace compliance, compatibility with all drivers (oci8, mysqli, pdo with mysql, sqlsrv, sqlite)

[HttpFoundation] update phpdoc of SessionHandlerInterface and unify p…

524bf84

…arameters of all handlers according to interface

Tobion changed the title ~~Fix doctrine session~~ [HttpFoundation] Fix DbalSessionHandler Apr 16, 2014

fabpot reviewed Apr 16, 2014
View reviewed changes

fabpot merged commit 524bf84 into symfony:2.3 Apr 18, 2014

Tobion deleted the fix-doctrine-session branch April 18, 2014 23:48

Tobion mentioned this pull request Apr 19, 2014

[Doctrine Bridge] simplify session handler #10743

Merged

doctrinebot mentioned this pull request Dec 7, 2015

DBAL-867: Doctrine\DBAL\Driver\Connection should be marked as an internal interface doctrine/dbal#2105

Closed

		try {
		$this->con->executeQuery("DELETE FROM {$this->tableName} WHERE sess_id = :id", array(

Uh oh!

[HttpFoundation] Fix DbalSessionHandler #10720

[HttpFoundation] Fix DbalSessionHandler #10720

Uh oh!

Conversation

Tobion commented Apr 16, 2014

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

fabpot commented Apr 18, 2014

Uh oh!

Uh oh!