cayenne-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Andrus Adamchik <>
Subject Re: Cayenne 4.1M1-snapshot upgrade experience
Date Mon, 18 Sep 2017 17:58:51 GMT
The old MergerFactory (now called MergerTokenFactory) was attached to DbAdapter and was not
injectable. Adapter acted as the factory of MergerFactory. So it was kind of auto-detected
by the virtue of the adapter being auto-detected. Though in a different place, it is still
auto-detected now.

In 4.0 merge token API was split away from the core runtime. It resides in a separate module
and DbAdapter no longer has access to it. To preserve the functionality (albeit not the API),
DbSyncModule creates a map of factories by adapter, and an injectable MergerTokenFactoryProvider
will give you the right factory for your adapter. The API is marginally harder than it was,
the functionality is the same. Should work fine with AutoAdapter too. E.g. if there's a PostgresAdapter
adapter working inside AutoAdapter, the factory will resolve to PostgresMergerTokenFactory
and so on. 

I suspect the reason why the code below caused the confusion though:

> MergerTokenFactory factory = runtime.getInjector().getInstance(MergerTokenFactory.class);
> assert factory instanceof MySQLMergerTokenFactory;

A binding for "MergerTokenFactory.class" is not the factory itself, but a _default/failover_
factory (i.e. the factory used when per-adapter lookup failed). Per-adapter factories are
stored elsewhere in a DI map (see for details). And the API to get a hold
of the factory is via MergerTokenFactoryProvider, as shown by Nikita. 

I guess one "fix" we may introduce is an extra DI "qualifier" to the default factory binding,
so that the code like yours above would simply fail instead of giving the wrong factory:

   binder.bind(Key.get(MergerTokenFactory.class, "failoverFactory").to(DefaultMergerTokenFactory.class);

In the meantime please use MergerTokenFactoryProvider API.


> On Sep 18, 2017, at 8:01 PM, John Huss <> wrote:
> I'm not sure what to make of your comment.  Previously the
> MergeTokenFactory was correctly auto-detected. Now it is not. I call that a
> regression.  If it was intentional, I think it was a poor choice. I have
> worked around the issue, but I recommend fixing it before this 4.1 gets
> released.
> On Thu, Sep 14, 2017 at 4:22 AM Nikita Timofeev <>
> wrote:
>> Ok, I think I found what is it :) See this task [1], if you don't want
>> to directly inject MergerTokenFactory you can use
>> MergerTokenFactoryProvider in your code like this (note that you will
>> need DbAdapter object):
>> @Inject
>> MergerTokenFactoryProvider factoryProvider;
>> // ...
>> MergerTokenFactory factory = factoryProvider.get(adapter);
>> [1]
>> On Wed, Sep 13, 2017 at 5:13 PM, John Huss <> wrote:
>>> On Wed, Sep 13, 2017 at 3:27 AM Nikita Timofeev <
>>> wrote:
>>>> Hi John,
>>>> Where exactly do you see this detection failure, is it Modeler DB
>>>> migration tool?
>>> This is in tomcat after the ServerRuntime is created from the builder.
>>> Here's a minimal example:
>>> ServerRuntime runtime = ServerRuntime.builder()
>>> .addConfigs("cayenne-MyProject.xml").build();
>>> MergerTokenFactory factory = runtime
>>> .getInjector().getInstance(MergerTokenFactory.class);
>>> assert factory instanceof MySQLMergerTokenFactory;
>>>> And how do you define your data source (via AutoAdapter or
>> MySQLAdapter)?
>>> I have custom DataSourceFactory specified in the cayenne project, but no
>>> override of the adapter (for MySQL), so it is using AutoAdapter.
>>>> On Tue, Sep 12, 2017 at 11:45 PM, John Huss <>
>> wrote:
>>>>> Also, the correct MergerTokenFactory for my DbAdapter is not longer
>>>>> automatically detected (both for Postgres and MySQL).  I had to add
>> this
>>>> to
>>>>> my ServerRuntime module to fix it.
>> binder.bind(MergerTokenFactory.class).to(PostgresMergerTokenFactory.class);
>>>>> On Thu, Aug 17, 2017 at 9:21 AM John Huss <>
>> wrote:
>>>>>> I just got done upgrading several projects to use the latest from
>>>>>> master. I am very excited about the memory usage improvements in
>>>>>> version.
>>>>>> I wanted to relay my experience along with some of the problems I
>> had in
>>>>>> case it would be helpful to others or allow this process to be
>> smoothed
>>>>>> out.  We were previously on a version of master from about two years
>>>> back.
>>>>>> I think most of these issues are fairly recent however.
>>>>>> - Can't build without running the tests (mvn clean install
>>>>>> -Dmaven.test.skip=true) at least once successfully due to missing
>> maven
>>>>>> dependency for cayenne.project (which is the second thing built).
>>>> don't
>>>>>> know why this depends on the tests.
>>>>>> - Can't successfully run tests without a newer Java 1.8 SDK. I had
>> JDK
>>>>>> 1.8.0 from 2014 (the first release?) that didn't work (mockito
>> errors).
>>>>>> Upgrading to the newest JDK fixed the problem.
>>>>>> - new ServerRuntime(...) no longer works because my custom DbAdapter
>>>>>> subclass would not get the new ValueObjectTypeRegistry parameter
>>>> injected
>>>>>> automatically. Switching to ServerRuntime.builder() fixed the
>> problem.
>>>>>> - CayenneFilter chooses a different name for the DataDomain when
>>>> load
>>>>>> multiple projects now (is the name the constant "cayenne"?). I use
>> this
>>>>>> name in a lot of places to set DB connection information so I don't
>>>> want to
>>>>>> change it. Unfortunately CayenneFilter does not provide a way to
>>>> override
>>>>>> the DataDomain's name, so I had to just stop using CayenneFilter.
>> Using
>>>>>> runtime.getDataDomain.setName("abc") doesn't work because the DB
>>>>>> already been connected to at that point. CayenneFilter should take
>>>>>> init-param to set the name instead.
>>>>>> - Cayenne project .xml files need to be upgraded by running the
>> Modeler
>>>>>> and re-saving them first.
>>>>>> - I had previously set the PK attribute's generation method to
>> Generated
>>>>>> for me (using Postgres). Since Postgres didn't used to support this
>> with
>>>>>> Cayenne, this used to fall back to the default behavior
>>>>>> "Cayenne-generated". Now that this IS supported everything fails
>> since
>>>> my
>>>>>> database isn't set up to do this yet. I switched these back to the
>>>> default
>>>>>> "Cayenne-generated" method of generating PKs. Not a Cayenne problem,
>> but
>>>>>> something to be aware of.
>>>>>> - The SQL logger's name has changed from CommonsJdbcEventLogger to
>> just
>>>>>> JdbcEventLogger, so I had to update places where I was changing the
>>>> logging
>>>>>> level (mostly in properties files).
>>>>>> John
>>>> --
>>>> Best regards,
>>>> Nikita Timofeev
>> --
>> Best regards,
>> Nikita Timofeev

View raw message