[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
[
List Home]
|
Re: [eclipselink-dev] questions while running JPA JUnit tests
|
Hi Tom,
I just received a reply from the Symfoware team.
It's not a detailed reply regarding the behaviours of each category in
your e-mail, but it does seem to explain the last lock errors I saw.
In my question to them I had described that I saw lock errors in step 4
of the following sequence:
1. Obtain a connection
2. (re)create tables (DROP TABLE, CREATE TABLE)
3. start transaction (con.setAutoCommit())
access tables (INSERT, UPDATE, DELETE)
commit transaction (con.commit())
4. recreate tables created at 2. (DROP TABLE, CREATE TABLE)
5. access tables as in 3.
My translation of his reply is that Symfoware has a concept of a
"session", and resources accessed from within a session cannot be
dropped, by design, until the session has been closed. He says that
could be the reason an error occurred at step 4.
He adds that there is no effect with the other categories of statements
(referring to the four categories in your e-mail below).
Does this make sense to you?
Regarding the hang I saw in getConnection, he asked more information, so
that issue might be unrelated.
Thanks,
Dies
Tom Ware wrote:
Hi Dies,
Sorry not to have gotten back to you at the end of the week last week
- last week things were very busy around here.
I am hesitant to go down the "static session" path until we have fully
explored the transaction route. Building a "static session" into our
schema generator in a reasonable way will be fairly involved.
I think we need to figure out the details of what we are allowed to do
and what we are not allowed to do and then figure out where the test
framework causes issues with these assumptions.
As I understand the information you have provided in the past, there
are several categories of statements that have different rules in
Symfoware:
1. CREATE TABLE
2. DROP TABLE
3. DML (INSERT, UPDATE, DELETE)
4. Temporary tables
For each of these there are several possible behaviors
1. No Effect on permissions for later statements
2. Transactional - If wrapped in a transaction, does not affect
permissions after transaction is committed. Can exist in the same
transaction as other types of statments.
3. Transactional (exclusive) - same as above, but needs its own transaction
4. Connection-Local - Later statements that affect the same table must
be in the same connection
5. Connection must be closed before later statement can affect the same
table
A definitive answer about what constraints we are trying to meet for
each of these scenarios will help us. Is there a good way to get that.
When we have a definitive answer, we can so some analysis of the test
issues you are seeing to see if there are easy ways we can make the test
framework behave in the way Symfoware expects.
A few other comments:
- Andrei Ilitchev send a link about temp tables in Symfoware that
contained some keywords like: "LOCAL TEMPORARY". It is hard for us to
understand alot of the docs we find for symfoware in the internet since
they seem to be in Japanese. Is that keyword a valid one. It seems
counter-intuitive to allow something called a temporary table and then
not be allowed to create it and insert into in the same transaction.
- Is it the first access to a temporary table that fails, or does the
first one succeed and a later one fails?
- TableSpace: Do all tables need a tablespace, or just temporary
tables? Are you also getting an issue with our normal table creation?
- You mentioned earlier you tried to experiment with the connection pool
settings. Remind me what your results are when you set the settings in
the test persistence xmls to have a MIN and MAX of 1.
- Update and Delete all statements are defined in the JPA 1.0
specification in the JPQL. e.g. Delete from Employee e where e.salary >
100000
- The types of statements that require Temporary tables are fairly
limited. (Andrei can probably give a better comment, but I think they
are the same ones that we rewrite to use a subquery and that were
causing issues in your test because the subquery refers to the same
table as the outer query.) If we log this issue as a limitation, I do
not think this would get in the way of certification.
-Tom
Dies Koper wrote:
Hi Tom,
> I am glad you are starting to get some success.
I might have reached the limit of this success already.
I can do a full run again (no hangs), with a success rate of 57.90%
(highest so far!), 56 failures and 495 errors.
The main errors/failures I see are due to:
- table locked issue (still)
- temporary table issue
I'm not sure why, but I see a certain table (for example CMP3_DEPT)
being dropped/created and successfully used, then it tries to drop and
recreate it, but the drop failed because the table's locked by another
user. This exception then prevents other tables from being created,
leading to other failures (like table not found).
From the log I can see the create and drop statements were executed
in transactions. Even the DML to it seems to all use transactions.
I don't know if it means anything to you, but the logged numeric id of
the connection when it failed was the same as when the table was
successfully created earlier, but the id of the connection used for
the DML was different.
I'd like to try the other idea you had with a static session. Could
you give more information?
I suppose I can swap it in in the locations where I now begin/commit
the transactions?
Can this static session use (unpooled) connections that can be closed
at "commit" time?
Did you take a look at any of the other calls to
shouldWriteToDatabase()? Will they require transactions as well?
Yes I did.
They don't require transactions for Symfoware because they relate to
creating/dropping constraints using ALTER TABLE (a syntax Symfoware
does not support) and to altering sequences (again, not supported by
Symfoware).
With global temporary tables implemented, creation of the table goes
fine, but the following INSERT fails because the table is "locked".
I'm not sure if I can (or should) put this table's create statement
in a transaction, I suppose this whole query could already be
running in a transaction.
I am surprised that Delete All and Update All queries are not running
in transactions already. Maybe we should try to isolate a sequence of
Note that a "Delete All" consists of the following statements (at
least in the tests that I saw failing):
1. creation of the global table
2. insert some rows
3. select some rows
4. drop the global table (or delete * from the table, depending on
whether local temporary tables are used).
All of this might be in one transaction, but the issue with Symfoware
was that DDL and DML can't be in the same transaction, so 1. needs to
run in a transaction and commit, then 2. and 3. in a new transaction,
then 4. (even putting just 1. and 4. in transactions does not seem to
help as 2. and 3. lock the table)
These statements are created and added to a Vector, and I'm not sure
where and how these statements are executed (so I could add the
demarcations between them). I am hoping Andrei will have a good
suggestion. Or a better solution.
tests that cause the failure... (i.e. Does the first Delete all or
Update All fail? Or, is the temporary table already created by
another Delete all or Update all? Do we see the initial create
demarked by transactions? Is there an error in the initial create
that causes issues?)
The temporary table seems to have been created successfully. Here is
the relevant part of the log:
[junit] [EL Finer]:
ServerSession(14864562)--Thread(Thread[main,5,main])--client acquired
[junit] [EL Finest]:
UnitOfWork(31975400)--Thread(Thread[main,5,main])--Execute query
DeleteAllQuery(referenceClass=Project sql="DELETE FROM TL_CMP3_PROJECT")
[junit] [EL Finer]:
ClientSession(2776693)--Connection(14563222)--Thread(Thread[main,5,main])--begin
transaction
[junit] [EL Fine]:
ClientSession(2776693)--Connection(14563222)--Thread(Thread[main,5,main])--CREATE
GLOBAL TEMPORARY TABLE TL_CMP3_PROJECT (PROJ_ID INTEGER NOT NULL,
PROJ_TYPE VARCHAR(255), DESCRIP VARCHAR(255), PROJ_NAME VARCHAR(255),
VERSION INTEGER, LEADER_ID INTEGER, PRIMARY KEY (PROJ_ID)) ON TESTDB01 10
[junit] [EL Fine]:
ClientSession(2776693)--Connection(14563222)--Thread(Thread[main,5,main])--INSERT
INTO TL_CMP3_PROJECT (PROJ_ID) SELECT PROJ_ID FROM CMP3_PROJECT WHERE
(PROJ_NAME = testUpdateAllProjects)
[junit] [EL Fine]:
ClientSession(2776693)--Connection(14563222)--Thread(Thread[main,5,main])--DELETE
FROM TL_CMP3_PROJECT
[junit] [EL Warning]:
UnitOfWork(31975400)--Thread(Thread[main,5,main])--Local Exception Stack:
[junit] Exception [EclipseLink-4002] (Eclipse Persistence Services
- 2.0.0.qualifier): org.eclipse.persistence.exceptions.DatabaseException
[junit] Internal Exception: java.sql.SQLException: [SymfoWARE ODBC
Driver][SymfoWARE Server] JYP2091E Table "TL_CMP3_PROJECT" of schema
"DEVELOP" being used exclusively by another user.
[junit] Error Code: -2091
[junit] Call: INSERT INTO TL_CMP3_PROJECT (PROJ_ID) SELECT PROJ_ID
FROM CMP3_PROJECT WHERE (PROJ_NAME = 'testUpdateAllProjects')
[junit] Query: DeleteAllQuery(referenceClass=Project sql="DELETE
FROM TL_CMP3_PROJECT")
[junit] at
org.eclipse.persistence.exceptions.DatabaseException.sqlException(DatabaseException.java:333)
Another issue with Symfoware's temporary tables is that the table
space name for the table must be specified at creation time. It does
not default to the table space that creation of normal tables
default to. So we'd need a way for the user to define this table space.
There is a table passed into the writeCreateTempTableSql() method.
Can we use the table space from that table?
I would like it to use that table's table space, but I don't know how
to find out dynamically what table space that is using. (I am
currently hard-coding it for now)
Symfoware V10 (coming out soon) has a new "DEFAULT TABLESPACE" option
that I can specify. I'm not sure how it works, but there is also a new
CREATE DEFAULT TABLESPACE DDL statement, that I suppose it works with.
So I could specify that and inform the user to prepare a default
tablespace, but in that case I need to explain in the Wiki which kind
of JPA functions require it.
Also, except for the table space I need to specify the number of
concurrent users that will use it. Is my understanding correct that
these tables are used within a transaction, so they will never be
accessed by multiple users?
In what cases are updateAll/deleteAll required exactly? Are they
related to JPA 2.0 functions? Are they required for JPA 1.0
functionality too?
UpdateAll and DeleteAll are both JPA 1.0 features and covered in the
JPA 1.0 TCK.
The spec does not use this terminology. Is it equivalent to an UPDATE
or DELETE statement with no WHERE clause? Or an UPDATE/DELETE
statement for an entity which has a particular relationship with other
entities?
As I see no clean solution I wonder what the impact is of making
this a limitation for this platform for now.
Have you looked into Andrei's suggestion of Local temporary tables.
Yes, unfortunately it did not help.
From an EclipseLink point of view, that is a possibility. It is
just a matter of figuring out what that means to you from a TCK point
of view?
No, more of a JPA1.0 /TLE point of view.
I've based this Symfoware platform class on the Symfoware platform my
team developed for TLE. I cannot remember encountering this issue at
the time. If there is a chance the same problem would occur on TLE
(using JPA 1.0 functionality, not TLE/Toplink's additional stuff), I'd
like to know the impact to Symfoware users of that implementation.
But also from a JPA 1.0/2.0 (=TCK?) point of view: to what extend is
the platform still usable without this function? If it only affects a
certain type of update/delete queries, and the range of functions that
is affected can be determined and clearly explained (from a user's
point of view), then I'd like to consider giving up on this and focus
on other functions.
Unless it is a condition to ever get the Symfoware platform graduated
from incubation (is it?), I'm not thinking of TCK certification at
this time.
If I no longer run into locking issues, what will the final solution
look like? Can I add a method createObjectsInTransactions() to the
DB platforms, defaulting to false of course, true for Symfoware,
that begins/commits/rolls back transactions in the locations I
described above?
That is likely the best way to address this. We could tweak the
method name a bit, and of course, check the other calls for
shouldWriteToDatabase() to see if they needed to be transactional as
well.
Ah, you don't like the method name. ;)
Well lucky for us the final solution will be different.
I won't propose createObjectsInStaticSession(). ;-p
Thanks,
Dies
Dies Koper wrote:
Hi Tom,
I am trying to enclose DDL calls with transactions. Inside
schemaManager.createSequences() there are mixed DDL and DML (select
and inserts on the sequence table/object), so just I'm moving the
transaction calls deeper into the call stack.
The locking error I get now is from the SELECT statement on table
CMP3_ENTITYB_SEQ in the following call in SchemaManager#createObject.
databaseObjectDefinition.createOnDatabase(getSession());
I did put a getSession().beginTransaction(); before this call, so
I'm not sure yet what the problem is. I'll investigate a bit more
and let you know.
Thanks,
Dies
The idea is that we would add calls that begin and commit
transactions around table creation calls.
Lets see if it works before we design the final solution:
- Find the
org.eclipse.persistence.tools.schemaframework.TableCreator.replaceTables(session,
schemaManager) method
- Add transactional boundaries
public void replac
