On 7/02/12 7:07 AM, Daniel Scheibe wrote:
Am 06.02.2012 00:21, schrieb Aristedes Maniatis:
On 6/02/12 9:48 AM, Daniel Scheibe wrote:
It's causing me an out of memory / permgen space exception since (from
what i can tell) Cayenne is keeping all the records im memory until
the commitChanges() call.
Am 05.02.2012 23:32, schrieb Aristedes Maniatis:
On 6/02/12 7:26 AM, Daniel Scheibe wrote:
Thanks Ari,
i'm trying to store information about the content of archives (ZIP and
RAR) inside a database table. They usually consist of a few hundred up
to 20-30k thousand entries and i would like to keep this inside a
transaction either suceeding or failing in a whole for a single
archive. Of course i could issue commits in between and process them
in chunks but its more work and harder to maintain the rollback of the
process if something goes wrong or fails during the import phase. As
previously mentioned i have it working with plain SQL statements but
would love to integrate Cayenne in between to get the full advantage
and convenience of an O/R mapper. Furthermore i'm using XADisk's
support for transactions on the filesystem level to keep it
synchronized with the database (also works out great with larger
transactions).
Appreciating your help! I will search the list again, maybe i've
missed some stuff that has been discussed already.
Cheers,
Daniel
Am 05.02.2012 13:28, schrieb Aristedes Maniatis:
On 5/02/12 10:11 PM, Daniel Scheibe wrote:
Hi guys,
i'm trying to figure out how Cayenne is handling transactions in
regards of when data is transferred to the database. I am trying to
get some large transactions (many small records or a few very large
records) working and so far haven't had luck. If i understand
correcly, Cayenne is holding my records in a cache and when i execute
the commitChanges() on the ObjectContext it starts to transfer all of
the data to the database (isolated in a real database
transaction). So
my question is (and please correct me if my previuos assumption is
wrong) how can I influence on this behaviour since i pretty much need
Cayenne to use the database transaction more directly. When i skip
Cayenne and issue SQL statements right through the JDBC connection
(ie.g. BEGIN TRANSACTION, UPDATE, UPDATE, UPDATE..., COMMIT
TRANSACTION) it works fine as with MySQL i'm only restricted to the
maximum allowed size of the InnoDB transaction logfile which is quite
sufficient for my purposes. On the other hand
i
completely understand that Cayenne would not know when i've finished
working with an Entity as there is no kind of "save" command on the
Entity itself.
Oh, this of course only goes for storing data inside the database,
not
for reading.
Maybe some of you guys with a bit more insight into the Cayenne
internals can help me out on this or point me to the right direction.
Thanks in advance and a refreshing weekend!
Cheers,
Daniel
Can you explain a little more what you are trying to do and what
problem
you are trying to avoid. Yes, there are ways to control the
transactions
through Cayenne (search this list for many previous thread discussing
this), but it sounds like you just need to commit your context at the
appropriate places in your code.
Ari
OK, good. Then what is the problem with using a single commit for the
whole batch?
Ari
Daniel,
Why do you post at the top of the emails when I've replied to the bottom?
Now that we actually know what your problem is that you are trying to
solve, perhaps someone can help. I believe the only approach is to take
control of the transactions yourself in order to be able to bypass the
normal behaviour of Cayenne with regard to committing and wrapping
several commits in one transaction. Then you can break up the Cayenne
commits into smaller chunks and wrap the whole thing in a database
transaction you manage yourself. I've never needed to do that myself,
but I believe it is possible.
They usually consist of a few hundred up to 20-30k thousand entries
If that means 20,000-30,000 entries, and the objects aren't too big, you
should be able to commit that in one go with a bit more RAM (yes,
Cayenne will keep the objects in memory). If you mean 20,000-30,000
thousand entries (that is, 20-30 million) then I think you are out of
the realm where an ORM is a good solution for your problem.
Ari
Hi Ari,
sorry my E-Mail client is used to quote from the top :) Thanks for you response, i tried to change my scenario a bit to get it right with Cayenne. Right know i'm inserting one record (Entity) with a byte buffer (mapped to a BLOB) and committing it right after through commitChanges() on the object context. The buffer is allocated once at the beginning of the loop (64kb) and reused for each new entity that is inserted. The problem here is that Cayenne is still eating up a lot of memory when creating let's say 1000 objects (1 object, then a commit, etc.) and causes an out of memory exception after some time. Is there any sort of a "write" cache that i could clear after doing a commit or something else? I don't see any reason why the memory consumption is growing and even if it caches something, everything that is committed should get GC'd at some point. I already isolated the problem, if i just remove the newObject/setData/commitChanges calls the process runs just fine. Maybe
i
can provide further information and some profiling results tomorrow.
I'm using Cayenne 3.0.2. and MySQL 5.
Btw: Are there any plans on integrating streams instead of byte[] buffers into
cayenne at some point in the future? MS-SQL Server has this FileStream data
type which looks promising but would require to channel the JDBC driver streams
support all the way up to the Entity level.
I can't speak for anyone else, but I'd have no interest in trying to fit this
approach into an ORM. Databases are not great places to store this type of data.
You may need to discard the Cayenne context between each batch that you
process. That should help with gc.
Ari
--
-------------------------->
Aristedes Maniatis
GPG fingerprint CBFB 84B4 738D 4E87 5E5C 5EFA EF6A 7D2E 3E49 102A
