Re: Cannot using Geronimo to execute bean-managed transaction with oracle transaction more than once, but Glassfish does
If just reading code of Geronimo , it's hard to understand sometimes. after building a debug-able environment and tracking the source code of Geronimo at run-time, things will be clearly. -origin- Sender: ext2 [mailto:x...@tongtech.com] Date: 2010/1/11 14:57 Receiver: user@geronimo.apache.org Subject: Re: Cannot using Geronimo to execute bean-managed transaction with oracle transaction more than once, but Glassfish does >>I see that ManagedXAConnection will set autoCommit to false in >>localTransactionStart(). So you mean Geronimo is enlisting the resouce >>before it starts the local transaction? >>I see ManagedXAConnection also sets autoCommit to true in >>localTransactionCommit() and localTransactionRollback(), so it's not >>only the cleanup() method. While using JTA , the Geronimo do nothing with localTransaction; The localTransaction only works when using in non-jta transaction. No matter the database 's driver support xa or no-xa, the data source can be configured to use in jta transaction or in No-jta transaction; -origin- 发件人: Jack Cai [mailto:greensi...@gmail.com] 发送时间: 2010年1月11日 11:42 收件人: user@geronimo.apache.org 主题: Re: Cannot using Geronimo to execute bean-managed transaction with oracle transaction more than once, but Glassfish does I see that ManagedXAConnection will set autoCommit to false in localTransactionStart(). So you mean Geronimo is enlisting the resouce before it starts the local transaction? I see ManagedXAConnection also sets autoCommit to true in localTransactionCommit() and localTransactionRollback(), so it's not only the cleanup() method. -Jack On Thu, Jan 7, 2010 at 8:42 PM, ext2 wrote: > Hi David Jeck: > > If the tranql's oracle wrapper is changed, the oracle will works well. And > user doesn't need write to a un-usual program at application level; > > Although I am still not agree to the Geronimo 's current action, but anyway > it's just another design strategy: now the database vendor has some trivial > difference , so the middle-ware (application-server) should has the > responsible to shield the difference and always give a consistence > appearance to the end-user; That's the strategy I finally decide to use in > my work (it's not a app-server, but still provide jta features); > > Best regard > Ext2 > -Origin----- > Sender: David Jencks [mailto:david_jen...@yahoo.com] > Date: 2010-1-7 17:23 > Receiver: user@geronimo.apache.org > Subject: Re: Cannot using Geronimo to execute bean-managed transaction with > oracle transaction more than once, but Glassfish does > > > On Jan 6, 2010, at 11:21 PM, ext2 wrote: > >> Hi, David Jeck: >> >> Sorry , I forget to said >>> or, does the connection have to have auto commit set to true or set >>> to >>> false (which one)? >> At the point of JTA's view, set auto commit has no other means, so >> we need >> not affect it and read it while using JTA; this works for most >> database in >> Geronimo, But oracle whose usage is different, only because of a >> bug of >> tranql. At application level ,we can only to resole the bug in a un- >> usual >> way. > > As I said, I haven't read the spec in several years, but I think this > is a bug in oracle. My recollection is that the autocommit state is > supposed to be ignored when enlisting in a JTA transaction. There > are other oracle bugs related to autocommit, for instance the spec > clearly says that (using local transactions) if autocommit is false, > setting it to true should commit pending work, but oracle does not. > >> >> Although setting auto commit has no other means in JTA, but this >> doesn't >> means the application-server has no responsible to check it. > > My recollection of the spec is that the app server has no need to > check the autocommit state since enlisting in a jta transaction should > cause the autocommit state to be ignored no matter what it is. > >> And that is >> what I doest agree to Geronimo 's current action; > > We can probably fix this in the tranql oracle wrapper. Geronimo > doesn't even know that it's dealing with a jdbc driver, so it has no > knowledge of the autocommit property's existence. > > thanks > david jencks > >> >> -Origin- >> Sender: David Jencks [mailto:david_jen...@yahoo.com] >> Date: 2010-1-7 2:03 >> Receiver: user@geronimo.apache.org >> Subject: Re: Cannot using Geronimo to execute bean-managed >> transaction with >> oracle transaction more than once, but Glassfish does >> >> >> On Jan 6, 2010, at 3:05 AM, ext2 wrote: >> >>> T
Re: Cannot using Geronimo to execute bean-managed transaction with oracle transaction more than once, but Glassfish does
I see that ManagedXAConnection will set autoCommit to false in localTransactionStart(). So you mean Geronimo is enlisting the resouce before it starts the local transaction? I see ManagedXAConnection also sets autoCommit to true in localTransactionCommit() and localTransactionRollback(), so it's not only the cleanup() method. -Jack On Thu, Jan 7, 2010 at 8:42 PM, ext2 wrote: > Hi David Jeck: > > If the tranql's oracle wrapper is changed, the oracle will works well. And > user doesn't need write to a un-usual program at application level; > > Although I am still not agree to the Geronimo 's current action, but anyway > it's just another design strategy: now the database vendor has some trivial > difference , so the middle-ware (application-server) should has the > responsible to shield the difference and always give a consistence > appearance to the end-user; That's the strategy I finally decide to use in > my work (it's not a app-server, but still provide jta features); > > Best regard > Ext2 > -Origin- > Sender: David Jencks [mailto:david_jen...@yahoo.com] > Date: 2010-1-7 17:23 > Receiver: user@geronimo.apache.org > Subject: Re: Cannot using Geronimo to execute bean-managed transaction with > oracle transaction more than once, but Glassfish does > > > On Jan 6, 2010, at 11:21 PM, ext2 wrote: > >> Hi, David Jeck: >> >> Sorry , I forget to said >>> or, does the connection have to have auto commit set to true or set >>> to >>> false (which one)? >> At the point of JTA's view, set auto commit has no other means, so >> we need >> not affect it and read it while using JTA; this works for most >> database in >> Geronimo, But oracle whose usage is different, only because of a >> bug of >> tranql. At application level ,we can only to resole the bug in a un- >> usual >> way. > > As I said, I haven't read the spec in several years, but I think this > is a bug in oracle. My recollection is that the autocommit state is > supposed to be ignored when enlisting in a JTA transaction. There > are other oracle bugs related to autocommit, for instance the spec > clearly says that (using local transactions) if autocommit is false, > setting it to true should commit pending work, but oracle does not. > >> >> Although setting auto commit has no other means in JTA, but this >> doesn't >> means the application-server has no responsible to check it. > > My recollection of the spec is that the app server has no need to > check the autocommit state since enlisting in a jta transaction should > cause the autocommit state to be ignored no matter what it is. > >> And that is >> what I doest agree to Geronimo 's current action; > > We can probably fix this in the tranql oracle wrapper. Geronimo > doesn't even know that it's dealing with a jdbc driver, so it has no > knowledge of the autocommit property's existence. > > thanks > david jencks > >> >> -Origin- >> Sender: David Jencks [mailto:david_jen...@yahoo.com] >> Date: 2010-1-7 2:03 >> Receiver: user@geronimo.apache.org >> Subject: Re: Cannot using Geronimo to execute bean-managed >> transaction with >> oracle transaction more than once, but Glassfish does >> >> >> On Jan 6, 2010, at 3:05 AM, ext2 wrote: >> >>> The problem is caused by tranql-commons-connector is not compatible >>> with >>> Oracle 9i 's XA API. The details things happened as following >>> description. >>> >>> When the XA transaction finished, the ManagedXAConnection(tranql) >>> will be >>> cleanup and put back to the pool. While cleanup the >>> ManagedXAConnection, >>> tranql reset the associated physical connection 's auto commit to >>> true. >>> Because at this time, Transaction Manager has finished the XA >>> transaction, >>> so reset the commit to true will acceptable by Oralce's XA driver; >>> >>> So unfortunately this will cause the XA resource cannot be enlist >>> into >>> transaction next time. and you can only execute the xa-transaction >>> only >>> once; >> >> It's been a long time since I looked into this, but if I remember the >> spec correctly oracle's driver is not spec compliant here. I'm not >> sure I understand exactly what sequences of operations are allowed by >> oracle and which are not. >> >> if you call setAutocommit on a connection, either true or false, does >> that prevent the associated managed conne
Re: Cannot using Geronimo to execute bean-managed transaction with oracle transaction more than once, but Glassfish does
Hi David Jeck: If the tranql's oracle wrapper is changed, the oracle will works well. And user doesn't need write to a un-usual program at application level; Although I am still not agree to the Geronimo 's current action, but anyway it's just another design strategy: now the database vendor has some trivial difference , so the middle-ware (application-server) should has the responsible to shield the difference and always give a consistence appearance to the end-user; That's the strategy I finally decide to use in my work (it's not a app-server, but still provide jta features); Best regard Ext2 -Origin- Sender: David Jencks [mailto:david_jen...@yahoo.com] Date: 2010-1-7 17:23 Receiver: user@geronimo.apache.org Subject: Re: Cannot using Geronimo to execute bean-managed transaction with oracle transaction more than once, but Glassfish does On Jan 6, 2010, at 11:21 PM, ext2 wrote: > Hi, David Jeck: > > Sorry , I forget to said >> or, does the connection have to have auto commit set to true or set >> to >> false (which one)? > At the point of JTA's view, set auto commit has no other means, so > we need > not affect it and read it while using JTA; this works for most > database in > Geronimo, But oracle whose usage is different, only because of a > bug of > tranql. At application level ,we can only to resole the bug in a un- > usual > way. As I said, I haven't read the spec in several years, but I think this is a bug in oracle. My recollection is that the autocommit state is supposed to be ignored when enlisting in a JTA transaction. There are other oracle bugs related to autocommit, for instance the spec clearly says that (using local transactions) if autocommit is false, setting it to true should commit pending work, but oracle does not. > > Although setting auto commit has no other means in JTA, but this > doesn't > means the application-server has no responsible to check it. My recollection of the spec is that the app server has no need to check the autocommit state since enlisting in a jta transaction should cause the autocommit state to be ignored no matter what it is. > And that is > what I doest agree to Geronimo 's current action; We can probably fix this in the tranql oracle wrapper. Geronimo doesn't even know that it's dealing with a jdbc driver, so it has no knowledge of the autocommit property's existence. thanks david jencks > > -----Origin- > Sender: David Jencks [mailto:david_jen...@yahoo.com] > Date: 2010-1-7 2:03 > Receiver: user@geronimo.apache.org > Subject: Re: Cannot using Geronimo to execute bean-managed > transaction with > oracle transaction more than once, but Glassfish does > > > On Jan 6, 2010, at 3:05 AM, ext2 wrote: > >> The problem is caused by tranql-commons-connector is not compatible >> with >> Oracle 9i 's XA API. The details things happened as following >> description. >> >> When the XA transaction finished, the ManagedXAConnection(tranql) >> will be >> cleanup and put back to the pool. While cleanup the >> ManagedXAConnection, >> tranql reset the associated physical connection 's auto commit to >> true. >> Because at this time, Transaction Manager has finished the XA >> transaction, >> so reset the commit to true will acceptable by Oralce's XA driver; >> >> So unfortunately this will cause the XA resource cannot be enlist >> into >> transaction next time. and you can only execute the xa-transaction >> only >> once; > > It's been a long time since I looked into this, but if I remember the > spec correctly oracle's driver is not spec compliant here. I'm not > sure I understand exactly what sequences of operations are allowed by > oracle and which are not. > > if you call setAutocommit on a connection, either true or false, does > that prevent the associated managed connection from ever being > enlisted in an XA transaction? > > or, does the connection have to have autocommit set to true or set to > false (which one)? > > I don't have an easy way to set up oracle here to investigate this for > myself, I hope I'm asking my questions clearly enough to be > understood. > > thanks > david jencks > > >> >> -Original - >> Sender: xuhongbo [mailto:x...@tongtech.com] >> Date: 2009-12-30 17:15 >> Receiver: user@geronimo.apache.org >> Subject: Re: Reply: Cannot using Geronimo to execute bean-managed >> transaction with oracle transaction more than once, but Glassfish >> does >> >> Hi Jack Cai: >> >> I have tried, the "tranql-connector-oracl
Re: Cannot using Geronimo to execute bean-managed transaction with oracle transaction more than once, but Glassfish does
On Jan 6, 2010, at 11:21 PM, ext2 wrote: Hi, David Jeck: Sorry , I forget to said or, does the connection have to have auto commit set to true or set to false (which one)? At the point of JTA's view, set auto commit has no other means, so we need not affect it and read it while using JTA; this works for most database in Geronimo, But oracle whose usage is different, only because of a bug of tranql. At application level ,we can only to resole the bug in a un- usual way. As I said, I haven't read the spec in several years, but I think this is a bug in oracle. My recollection is that the autocommit state is supposed to be ignored when enlisting in a JTA transaction. There are other oracle bugs related to autocommit, for instance the spec clearly says that (using local transactions) if autocommit is false, setting it to true should commit pending work, but oracle does not. Although setting auto commit has no other means in JTA, but this doesn't means the application-server has no responsible to check it. My recollection of the spec is that the app server has no need to check the autocommit state since enlisting in a jta transaction should cause the autocommit state to be ignored no matter what it is. And that is what I doest agree to Geronimo 's current action; We can probably fix this in the tranql oracle wrapper. Geronimo doesn't even know that it's dealing with a jdbc driver, so it has no knowledge of the autocommit property's existence. thanks david jencks -Origin- Sender: David Jencks [mailto:david_jen...@yahoo.com] Date: 2010-1-7 2:03 Receiver: user@geronimo.apache.org Subject: Re: Cannot using Geronimo to execute bean-managed transaction with oracle transaction more than once, but Glassfish does On Jan 6, 2010, at 3:05 AM, ext2 wrote: The problem is caused by tranql-commons-connector is not compatible with Oracle 9i 's XA API. The details things happened as following description. When the XA transaction finished, the ManagedXAConnection(tranql) will be cleanup and put back to the pool. While cleanup the ManagedXAConnection, tranql reset the associated physical connection 's auto commit to true. Because at this time, Transaction Manager has finished the XA transaction, so reset the commit to true will acceptable by Oralce's XA driver; So unfortunately this will cause the XA resource cannot be enlist into transaction next time. and you can only execute the xa-transaction only once; It's been a long time since I looked into this, but if I remember the spec correctly oracle's driver is not spec compliant here. I'm not sure I understand exactly what sequences of operations are allowed by oracle and which are not. if you call setAutocommit on a connection, either true or false, does that prevent the associated managed connection from ever being enlisted in an XA transaction? or, does the connection have to have autocommit set to true or set to false (which one)? I don't have an easy way to set up oracle here to investigate this for myself, I hope I'm asking my questions clearly enough to be understood. thanks david jencks -Original - Sender: xuhongbo [mailto:x...@tongtech.com] Date: 2009-12-30 17:15 Receiver: user@geronimo.apache.org Subject: Re: Reply: Cannot using Geronimo to execute bean-managed transaction with oracle transaction more than once, but Glassfish does Hi Jack Cai: I have tried, the "tranql-connector-oracle-local" is ok, but it doesn't use xa transaction in jta, and use a faked Local-XAResource instead of real oracle-xa-resource; I am sorry to mis-understand your means and give the run-time class name in my previous reply; The error occurred program is just using " tranql-connector-oracle-xa" Additionally, I have try another Mysql database and using " tranql-connector-mysql-xa" do real xa transaction. It works well. So my mind changed, maybe there is something not compatible with oracle 9i database; In my original mail, I have post a very simple program which use the Geronimo Transaction Manager and Oracle XA API directly, this works well; Because tranql resource adaptor is a very simple wrapper , Geronimo does additional things to wrap the database connection (etc control pooling, xa-resource wrap, xa-resource cache for transaction-manager ...) , so I am wondering if there is some other un-excepted database operation has been done and cause this problem? for convenience I post the simple program again. If we only concern database operation, does this simple program done exactly like the Geronimo done ? Or it doesn't , Geronimo do additional things... maybe the difference will be the real reason cause the problem; I have tracked at runtime, but unfortunately has not find some difference yet... Thanks a lot xuhongbo -origin- sender: Jack Cai [mail
Re: Cannot using Geronimo to execute bean-managed transaction with oracle transaction more than once, but Glassfish does
Hi, David Jeck: Sorry , I forget to said >or, does the connection have to have auto commit set to true or set to >false (which one)? At the point of JTA's view, set auto commit has no other means, so we need not affect it and read it while using JTA; this works for most database in Geronimo, But oracle whose usage is different, only because of a bug of tranql. At application level ,we can only to resole the bug in a un-usual way. Although setting auto commit has no other means in JTA, but this doesn't means the application-server has no responsible to check it. And that is what I doest agree to Geronimo 's current action; -Origin- Sender: David Jencks [mailto:david_jen...@yahoo.com] Date: 2010-1-7 2:03 Receiver: user@geronimo.apache.org Subject: Re: Cannot using Geronimo to execute bean-managed transaction with oracle transaction more than once, but Glassfish does On Jan 6, 2010, at 3:05 AM, ext2 wrote: > The problem is caused by tranql-commons-connector is not compatible > with > Oracle 9i 's XA API. The details things happened as following > description. > > When the XA transaction finished, the ManagedXAConnection(tranql) > will be > cleanup and put back to the pool. While cleanup the > ManagedXAConnection, > tranql reset the associated physical connection 's auto commit to > true. > Because at this time, Transaction Manager has finished the XA > transaction, > so reset the commit to true will acceptable by Oralce's XA driver; > > So unfortunately this will cause the XA resource cannot be enlist > into > transaction next time. and you can only execute the xa-transaction > only > once; It's been a long time since I looked into this, but if I remember the spec correctly oracle's driver is not spec compliant here. I'm not sure I understand exactly what sequences of operations are allowed by oracle and which are not. if you call setAutocommit on a connection, either true or false, does that prevent the associated managed connection from ever being enlisted in an XA transaction? or, does the connection have to have autocommit set to true or set to false (which one)? I don't have an easy way to set up oracle here to investigate this for myself, I hope I'm asking my questions clearly enough to be understood. thanks david jencks > > -Original - > Sender: xuhongbo [mailto:x...@tongtech.com] > Date: 2009-12-30 17:15 > Receiver: user@geronimo.apache.org > Subject: Re: Reply: Cannot using Geronimo to execute bean-managed > transaction with oracle transaction more than once, but Glassfish does > > Hi Jack Cai: > > I have tried, the "tranql-connector-oracle-local" is ok, but it > doesn't use > xa transaction in jta, and use a faked Local-XAResource instead of > real > oracle-xa-resource; > > I am sorry to mis-understand your means and give the run-time class > name in > my previous reply; The error occurred program is just using " > tranql-connector-oracle-xa" > > Additionally, I have try another Mysql database and using " > tranql-connector-mysql-xa" do real xa transaction. It works well. > > So my mind changed, maybe there is something not compatible with > oracle 9i > database; In my original mail, I have post a very simple program > which use > the Geronimo Transaction Manager and Oracle XA API directly, this > works > well; > > Because tranql resource adaptor is a very simple wrapper , Geronimo > does > additional things to wrap the database connection (etc control > pooling, > xa-resource wrap, xa-resource cache for transaction-manager ...) , > so I am > wondering if there is some other un-excepted database operation has > been > done and cause this problem? for convenience I post the simple program > again. > > If we only concern database operation, does this simple program done > exactly like the Geronimo done ? Or it doesn't , Geronimo do > additional > things... maybe the difference will be the real reason cause the > problem; > I have tracked at runtime, but unfortunately has not find some > difference > yet... > > Thanks a lot > xuhongbo > > > -origin- > sender: Jack Cai [mailto:greensi...@gmail.com] > date: 2009/12/30 11:45 > receiver: user@geronimo.apache.org > subject: Re: Reply: Cannot using Geronimo to execute bean-managed > transaction with oracle transaction more than once, but Glassfish does > > Can you try to use the tranql-connector-oracle-xa or > tranql-connector-oracle-local to do the test? > > -Jack > > On Wed, Dec 30, 2009 at 11:26 AM, xuhongbo wrote: >> >>>> In the future it would be great if you could
Re: Cannot using Geronimo to execute bean-managed transaction with oracle transaction more than once, but Glassfish does
>It's been a long time since I looked into this, but if I remember the >spec correctly oracle's driver is not spec compliant here. I'm not >sure I understand exactly what sequences of operations are allowed by >oracle and which are not. I am not sure about the spec you said is which one, If you means X/Open, in my test result until now, I haven't find anything which is not compliant. If you means JTA spec, the sequence is true different, but this not the oracle driver itself caused but tranql caused.At following , I illustrate it more details. The right sequence for oracle program in Geronimo must be : 1)Get connection from datasource 2) Set connection auto commit to false; //you must reset the auto-commit status to false. As tranql will force the connection auto-commit to true when JTA finished and cause enlist error; 3)usertransaction.begin ...do some thing... 4)Usertransaction end (commit or rollback) 5)Connection.close() // step 4,5 's sequence doesn't matter. But any how when the 2 step both finished. The Managed Connection will be clean up to release registered listener and prepare for put back into database pool. But unfortunately , tranql will force auto-commit status to true when cleanup; >if you call setAutocommit on a connection, either true or false, does >that prevent the associated managed connection from ever being >enlisted in an XA transaction? Change the wrapped connection's auto-commit status will affect the underlying physical connection's auto-commit status; but if we only concern about oracle-xa , it will cause different exception occurred, and it's not real-reason cause enlist error; (real reason is tranql's cleanup, I have mentioned, you can see source code of tranql : org.tranql.connector.jdbc. ManagedXAConnection.cleanup()); But if we concern other database, the Geronimo 's current action for set auto-commit maybe cause other bad things: if end-user mis-used set auto-commit, the program may or may not be work right, all dependency on database vendor; for example the two database oracle and mysql I have test. While using oracle, set auto-commit to false is always acceptable , but when setting it to true , oracle driver will check : if the xa-resource is enlist in global transaction, oracle will throw error; if xa-resource is not enlist in global transaction, oracle will accept it ; But before enlist oracle xa resource, we must make sure it's auto commit must be false; While mysql doesn't care it, that's to say: not matter set to true or false, no matter in or outside jta transaction. It doesn't care. Things will always be right; >or, does the connection have to have auto commit set to true or set to >false (which one)? It's purely dependent on the database vendor , if Geronimo doesn't change; By the way , I thing the Geronimo and tranql should do additional things to enhance JTA using XA: 1: When using XA Datasource in JTA Transaction, if end-user set auto commit status , Geronimo must always check it itself, but not directly dependent on database driver , that is : if set to true, Geronimo will always report error, if set to false, Geronimo will always omit; 2) Tranql 's oracle xa driver should changed to make sure the underlying connection's auto-commit status must be false before enlist XAResource, or never change the auto-commit status and left it be default value -Origin----- Sender: David Jencks [mailto:david_jen...@yahoo.com] Date: 2010-1-7 2:03 Receiver: user@geronimo.apache.org Subject: Re: Cannot using Geronimo to execute bean-managed transaction with oracle transaction more than once, but Glassfish does On Jan 6, 2010, at 3:05 AM, ext2 wrote: > The problem is caused by tranql-commons-connector is not compatible > with > Oracle 9i 's XA API. The details things happened as following > description. > > When the XA transaction finished, the ManagedXAConnection(tranql) > will be > cleanup and put back to the pool. While cleanup the > ManagedXAConnection, > tranql reset the associated physical connection 's auto commit to > true. > Because at this time, Transaction Manager has finished the XA > transaction, > so reset the commit to true will acceptable by Oralce's XA driver; > > So unfortunately this will cause the XA resource cannot be enlist > into > transaction next time. and you can only execute the xa-transaction > only > once; It's been a long time since I looked into this, but if I remember the spec correctly oracle's driver is not spec compliant here. I'm not sure I understand exactly what sequences of operations are allowed by oracle and which are not. if you call setAutocommit on a connection, either true or false, does that prevent the associated managed connection from ever being
Re: Cannot using Geronimo to execute bean-managed transaction with oracle transaction more than once, but Glassfish does
On Jan 6, 2010, at 3:05 AM, ext2 wrote: The problem is caused by tranql-commons-connector is not compatible with Oracle 9i 's XA API. The details things happened as following description. When the XA transaction finished, the ManagedXAConnection(tranql) will be cleanup and put back to the pool. While cleanup the ManagedXAConnection, tranql reset the associated physical connection 's auto commit to true. Because at this time, Transaction Manager has finished the XA transaction, so reset the commit to true will acceptable by Oralce's XA driver; So unfortunately this will cause the XA resource cannot be enlist into transaction next time. and you can only execute the xa-transaction only once; It's been a long time since I looked into this, but if I remember the spec correctly oracle's driver is not spec compliant here. I'm not sure I understand exactly what sequences of operations are allowed by oracle and which are not. if you call setAutocommit on a connection, either true or false, does that prevent the associated managed connection from ever being enlisted in an XA transaction? or, does the connection have to have autocommit set to true or set to false (which one)? I don't have an easy way to set up oracle here to investigate this for myself, I hope I'm asking my questions clearly enough to be understood. thanks david jencks -Original - Sender: xuhongbo [mailto:x...@tongtech.com] Date: 2009-12-30 17:15 Receiver: user@geronimo.apache.org Subject: Re: Reply: Cannot using Geronimo to execute bean-managed transaction with oracle transaction more than once, but Glassfish does Hi Jack Cai: I have tried, the "tranql-connector-oracle-local" is ok, but it doesn’t use xa transaction in jta, and use a faked Local-XAResource instead of real oracle-xa-resource; I am sorry to mis-understand your means and give the run-time class name in my previous reply; The error occurred program is just using " tranql-connector-oracle-xa" Additionally, I have try another Mysql database and using " tranql-connector-mysql-xa" do real xa transaction. It works well. So my mind changed, maybe there is something not compatible with oracle 9i database; In my original mail, I have post a very simple program which use the Geronimo Transaction Manager and Oracle XA API directly, this works well; Because tranql resource adaptor is a very simple wrapper , Geronimo does additional things to wrap the database connection (etc control pooling, xa-resource wrap, xa-resource cache for transaction-manager ...) , so I am wondering if there is some other un-excepted database operation has been done and cause this problem? for convenience I post the simple program again. If we only concern database operation, does this simple program done exactly like the Geronimo done ? Or it doesn't , Geronimo do additional things... maybe the difference will be the real reason cause the problem; I have tracked at runtime, but unfortunately has not find some difference yet... Thanks a lot xuhongbo -origin- sender: Jack Cai [mailto:greensi...@gmail.com] date: 2009/12/30 11:45 receiver: user@geronimo.apache.org subject: Re: Reply: Cannot using Geronimo to execute bean-managed transaction with oracle transaction more than once, but Glassfish does Can you try to use the tranql-connector-oracle-xa or tranql-connector-oracle-local to do the test? -Jack On Wed, Dec 30, 2009 at 11:26 AM, xuhongbo wrote: In the future it would be great if you could only post to one mailing list. Thanks, I know I'm not sure what is wrong yet, however you should never try to set the autocommit state of a connection that is enlisted in a jta transaction. Enlisting and delisting the XAConnection will result in the autocommit being dealt with properly. The value from getAutoCommit may or may not be meaningful in a jta transaction. Outside a jta transaction the autocommit state defaults to true. I'm quite surprised you didn't get a more informative error. Yes you are right, auto commit has no means for jta connection and cannot be set to true; here I just set auto commit to false, so a jta connection should just omit it; But the surprise thing is if I doesn't affect auto commit state(in the program, just comment the statement "setAutocommit(false)"), when execute database operation, a "ORA-02089: COMMIT ..." exception will be throwed by oracle's database driver; it looks like the Geronimo does a wrong things "commit on the connection when execute database operation"; and this should only occurs on non-jta connection, because only no-jta connection will set auto commit default to true. One important piece of information that I don't see is which tranql wrapper you used to deploy your datasource. The datasource is org.tranql.connector.jdbc.DataSource. And it use a a managed-connection factory " org.tranql.connector.oracle.XAMCF" to ope
Re: Cannot using Geronimo to execute bean-managed transaction with oracle transaction more than once, but Glassfish does
The problem is caused by tranql-commons-connector is not compatible with Oracle 9i 's XA API. The details things happened as following description. When the XA transaction finished, the ManagedXAConnection(tranql) will be cleanup and put back to the pool. While cleanup the ManagedXAConnection, tranql reset the associated physical connection 's auto commit to true. Because at this time, Transaction Manager has finished the XA transaction, so reset the commit to true will acceptable by Oralce's XA driver; So unfortunately this will cause the XA resource cannot be enlist into transaction next time. and you can only execute the xa-transaction only once; -Original - Sender: xuhongbo [mailto:x...@tongtech.com] Date: 2009-12-30 17:15 Receiver: user@geronimo.apache.org Subject: Re: Reply: Cannot using Geronimo to execute bean-managed transaction with oracle transaction more than once, but Glassfish does Hi Jack Cai: I have tried, the "tranql-connector-oracle-local" is ok, but it doesnt use xa transaction in jta, and use a faked Local-XAResource instead of real oracle-xa-resource; I am sorry to mis-understand your means and give the run-time class name in my previous reply; The error occurred program is just using " tranql-connector-oracle-xa" Additionally, I have try another Mysql database and using " tranql-connector-mysql-xa" do real xa transaction. It works well. So my mind changed, maybe there is something not compatible with oracle 9i database; In my original mail, I have post a very simple program which use the Geronimo Transaction Manager and Oracle XA API directly, this works well; Because tranql resource adaptor is a very simple wrapper , Geronimo does additional things to wrap the database connection (etc control pooling, xa-resource wrap, xa-resource cache for transaction-manager ...) , so I am wondering if there is some other un-excepted database operation has been done and cause this problem? for convenience I post the simple program again. If we only concern database operation, does this simple program done exactly like the Geronimo done ? Or it doesn't , Geronimo do additional things... maybe the difference will be the real reason cause the problem; I have tracked at runtime, but unfortunately has not find some difference yet... Thanks a lot xuhongbo -origin- sender: Jack Cai [mailto:greensi...@gmail.com] date: 2009/12/30 11:45 receiver: user@geronimo.apache.org subject: Re: Reply: Cannot using Geronimo to execute bean-managed transaction with oracle transaction more than once, but Glassfish does Can you try to use the tranql-connector-oracle-xa or tranql-connector-oracle-local to do the test? -Jack On Wed, Dec 30, 2009 at 11:26 AM, xuhongbo wrote: > >>>In the future it would be great if you could only post to one mailing >>>list. > Thanks, I know > >>>I'm not sure what is wrong yet, however you should never try to set >>>the autocommit state of a connection that is enlisted in a jta >>>transaction. Enlisting and delisting the XAConnection will result in >>>the autocommit being dealt with properly. The value from >>>getAutoCommit may or may not be meaningful in a jta transaction. >>>Outside a jta transaction the autocommit state defaults to true. I'm >>>quite surprised you didn't get a more informative error. > > Yes you are right, auto commit has no means for jta connection and cannot be > set to true; here I just set auto commit to false, so a jta connection > should just omit it; > But the surprise thing is if I doesn't affect auto commit state(in the > program, just comment the statement "setAutocommit(false)"), when execute > database operation, a "ORA-02089: COMMIT ..." exception will be throwed by > oracle's database driver; it looks like the Geronimo does a wrong things > "commit on the connection when execute database operation"; and this should > only occurs on non-jta connection, because only no-jta connection will set > auto commit default to true. > >>>One important piece of information that I don't see is which tranql >>>wrapper you used to deploy your datasource. > > The datasource is org.tranql.connector.jdbc.DataSource. And it use a a > managed-connection factory " org.tranql.connector.oracle.XAMCF" to open > connection; And the managed-connection factory use a oracle's xa datasource > (oracle.jdbc.xa.client.OracleXADataSource); > > By the way , I haven't ever post a trivial problem I have meet when I deploy > the Oracle-XA data source. The trivial thing is: I must change the deploy > plan created by Geronimo's web manage console tools, delete the empty > property "TNSEntryName" and manually deploy it; because this is the oracle9i > database driver's question --- "a empty string value(not a null value) set > to TNSEntryName" will cause oracle9i's database driver to omit the other > property (etc serverName, serviceName ...) and cannot establish a connect ; > I thinks this should have no means to the transaction commit failure;