Drop support for `EntityManager#flush($something)`

[Ocramius]

EntityManager#flush($something) is a problematic operation, since it leads to consumers potentially storing incomplete in-memory graphs into the DB.

In general, flush($something) should be used as an optimization, but dealing with it from within the ORM has only led to bugs and massive headaches so far, whereas keeping a smaller UnitOfWork is trivial from the consumer perspective.

Therefore, I suggest dropping this ability completely, while consumers should instead approach batch-processing problems with something like:

• explicit change tracking policy
• use a TableDataGateway approach (TableDataGateway interface for fetching/saving entities without UnitOfWork interactions #5550)
• use EntityManager#clear() more often, especially in combination with a READ_WRITE second level cache region
• create new EntityManager instances, and keep them small

[chalasr]

👍 I too often seen changes made on the associations of an entity not took into account because of the use of EntityManager#flush($entity).
On the other hand it's often scaring to don't specify the entity, because some other entities may be unexpectedly impacted, even if having entities in invalid states should never occur...

create new EntityManager instances, and keep them small

Not always easy e.g. in symfony where the manager is default a shared service which is injected everywhere, that is imho the cause of lots of issues.

use EntityManager#clear() more often

Looks like a good solution, it's more explicit what you're doing.

[Jean85]

In my team, we use a lot of this, to avoid flushing something else that got fetched from the DB and I don't want to change right now. This is an issue when the code base is large and there are a lot of (Symfony) events and listeners, and it's hard to keep track of everything. I agree that clear() is the right solution with batch operations, and in fact we use that a lot, since it has the important benefit of freeing memory.

I'm interested in the suggestions for the alternative approaches, I will look more into it on Monday with more code at hand.

[pablomoreno61]

I've had unexpected entities persisted because of not using $entity inside the flush operation, so nowadays I use it in all cases.

Your alternatives look interesting, I will investigate them

[Ocramius]

The EntityManager#flush() method makes no guarantees about what ia being
flushed: it just reduces computed changesets

On 5 Nov 2016 15:55, "Pablo Moreno" notifications@github.com wrote:

I've had unexpected entities persisted because of not using $entity inside
the flush operation, so nowadays I use it in all cases.

Your alternatives look interesting, I will investigate them

—
You are receiving this because you were assigned.
Reply to this email directly, view it on GitHub
#6118 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AAJakPHQ2PM2AzqBJWmUaGtZiVTq7P1Uks5q7JjVgaJpZM4KqNo8
.

[baptistemanson]

We use it in a ZF2 daemon listening to a queue. It is the only way we found to get rid of all the in memory objects to retrieve them from the db regularly.

[yannickl88]

We do the same as @Jean85 within our company. So I'm really interested in the alternatives. A way to create a "sub-entity manager" (same connection, same config, new UnitOfWork) would pretty much solve the issue since each listener can use its own entity manager. Thus it cannot influence the other state.

[alcaeus]

I'm interested in the suggestions for the alternative approaches, I will look more into it on Monday with more code at hand.

The most obvious one would be to use explicit change tracking instead of the default implicit change tracking. In a nutshell, when you use explicit change tracking and you call flush(), no entities will be updated. Every entity you want to be updated has to be marked for update with persist() (as if it were a new entity). Note that change tracking can be specified on a per-entity basis, so you could actually store some entities the old-fashioned way.

When it comes to event subscribers that write entities to the database, you should most likely use a separate EntityManager. Even when calling flush($entity), the UnitOfWork would still write new documents to the database. All you're doing is circumventing the implicit change tracking mentioned above.

I'm all for removing this - the comments in this thread as well as on Twitter show quite well that many developers misunderstand what the feature does.

[drola]

After this change happens we need to make sure that flush() method still checks its argument count and throws an exception if there is an argument given. Ignoring the argument silently will lead in hard to catch bugs in people's codebases.

[Ocramius]

Very good point, Matjaž!

On 6 Nov 2016 11:10, "Matjaž Drolc" notifications@github.com wrote:

After this change happens we need to make sure that flush() method still
checks its argument count and throws an exception if there is an argument
given. Ignoring the argument silently will lead in hard to catch bugs in
people's codebases.

—
You are receiving this because you were assigned.
Reply to this email directly, view it on GitHub
#6118 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AAJakJHB1IQUKYahmukladdiW5pI54tCks5q7aeFgaJpZM4KqNo8
.

[sstok]

@yannickl88 If I'm not mistaken you "can" configure multiple entity-managers in a Symfony application, but I have not seen many developers use it. It would make sense to give each bounded context it's own EntityManager rather then sharing this with other contexts.