Hi Willem. I did some experiment on CompensationMessageHandler.
In first case, I send a delayed TxCompensatedEvent to simulate async
TxCompensated notification:
@Override
public void onReceive(final String globalTxId, final String localTxId,
final String parentTxId, final String compensationMethod,
Object... payloads) {
context.apply(globalTxId, localTxId, compensationMethod, payloads);
new Thread(new Runnable() {
@Override
public void run() {
try {
Thread.sleep(10000L);
} catch (InterruptedException e) {
e.printStackTrace();
}
sender.send(new TxCompensatedEvent(globalTxId, localTxId,
parentTxId, compensationMethod));
}
}).start();
}
I watched command table and found that compensation command status
PENDING -> DONE transition.
I also watched txevent table found SagaEnded correctly:
In second case, I just remove TxCompensatedEvent sending code:
@Override
public void onReceive(final String globalTxId, final String localTxId,
final String parentTxId, final String compensationMethod,
Object... payloads) {
context.apply(globalTxId, localTxId, compensationMethod, payloads);
}
In command table compensation command remains PENDING and in txevent
table no SagaEndedEvent happens.
Zheng Feng <[email protected]> 于2019年7月24日周三 下午4:07写道：
>
> So the grpc could support the async invoking ?
>
> Willem Jiang <[email protected]> 于2019年7月24日周三 下午2:31写道：
>
> > Hi Daniel,
> >
> > If you take a look at the Omega side, the CompensationMessageHandler
> > is called in onNext() method of GrpcCompensateStreamObserver.
> > If there is something wrong with CompensationMessageHandler
> > onReceive() method, the Stream could be broken and Alpha can now about
> > it immediately. (You can write a simple test case to verify it).
> >
> > Willem Jiang
> >
> > Twitter: willemjiang
> > Weibo: 姜宁willem
> >
> > On Wed, Jul 24, 2019 at 12:27 PM Daniel Qian <[email protected]>
> > wrote:
> > >
> > > Thanks for sharing your thoughts, Willem. But I didn't really get your
> > > point.
> > >
> > > I read the code and found that compensation action works like this:
> > >
> > > 1. Alpha GrpcOmegaCallback send GrpcCompensateCommand to Omega
> > > 2. Omega CompensationMessageHandler handle the command, execute
> > > compensation logic and send TxCompensatedEvent to Alpha
> > >
> > > In this progress, Alpha doesn't wait for compensation success message
> > > after sending compensation command, it just returns. So the progress is
> > > already async.
> > >
> > > If what I said above is correct, please read the following, if not,
> > please
> > > correct me.
> > >
> > > As to the current implementation, it's possible that compensation command
> > > is sent but no compensation success message received. I think what you
> > > said is Alpha should have a mechanism to handle compensation timeout,
> > > retry or just log it. But I think this an improvement should be done on
> > the
> > > Alpha side. Am I right?
> > >
> > > In my opinion, to support async compensation, the only thing need to do
> > is
> > > providing a way to let user code send TxCompensatedEvent. Sure I don't
> > mean
> > > that we should expose the underly communication,
> > > GrpcSagaClientMessageSender
> > > for example, to user code. Providing a encapsulated interface will be
> > > better,
> > > just like what I mentioned before.
> > >
> > > Zheng Feng <[email protected]> 于2019年7月23日周二 下午2:10写道：
> > > >
> > > > OK, I understand and with my prev experiences with the JTA and XA, it
> > > looks
> > > > similar with the XA.recovery() to get the in-doubt transactions. So the
> > > > omega side or the application have to keep the records of the pending
> > > > transactions, is it right ?
> > > >
> > > > Willem Jiang <[email protected]> 于2019年7月22日周一 下午11:51写道：
> > > >
> > > > > Yes, We can kick off the recovery process or status verify process in
> > > > > Alpha if the distribute transaction is in suspend status, but in
> > this
> > > > > case we need Alpha talk to customer Application to tell if we need to
> > > > > do the recovery again or just do some auditing there. It depends on
> > > > > the customer's configuration.
> > > > >
> > > > > Willem Jiang
> > > > >
> > > > > Twitter: willemjiang
> > > > > Weibo: 姜宁willem
> > > > >
> > > > > On Mon, Jul 22, 2019 at 11:38 PM Zheng Feng <[email protected]>
> > wrote:
> > > > > >
> > > > > > Hi Willem,
> > > > > >
> > > > > > In the term of providing some further processing extension, you
> > mean
> > > that
> > > > > > the customer could help the state machine to recovery from the
> > > suspension
> > > > > > states ?
> > > > > >
> > > > > > Willem Jiang <[email protected]> 于2019年7月22日周一 下午11:28写道：
> > > > > >
> > > > > > > If we provide the async compensation call, we need to do lots of
> > > thing
> > > > > > > on Alpha and Omega side. My suggestion we just use provide sync
> > way
> > > to
> > > > > > > get the feedback of compensation immediately to keep the design
> > > > > > > simpler.
> > > > > > >
> > > > > > > As we are reimplementing the Alpha with state machine, there are
> > > some
> > > > > > > suspension state which could be caused by the timeout or the
> > missing
> > > > > > > event message.
> > > > > > > I had a long talk with ZhangLei, we are agree that we could leave
> > > the
> > > > > > > status check or recovery to the customer by providing some
> > further
> > > > > > > processing extension, as there are too many detail things in
> > > customer
> > > > > > > code to think about.
> > > > > > >
> > > > > > > Anyway, it's my pleasure to have this kind of discussion with the
> > > > > > > team, it's the beauty of Open Source project development, we are
> > > > > > > tackling the interesting problem by working together from
> > different
> > > > > > > company :)
> > > > > > >
> > > > > > > Willem Jiang
> > > > > > >
> > > > > > > Twitter: willemjiang
> > > > > > > Weibo: 姜宁willem
> > > > > > >
> > > > > > > On Mon, Jul 22, 2019 at 5:14 PM Zheng Feng <[email protected]>
> > > wrote:
> > > > > > > >
> > > > > > > > In term of the compensation method, I think I had discussed
> > with
> > > > > Willem
> > > > > > > > before that it could introduce the @Status annotation for the
> > > alpha
> > > > > > > server
> > > > > > > > to query the compensation status. When the compensation method
> > is
> > > > > async
> > > > > > > > which means it can not response immediately, it would return
> > the
> > > > > > > > COMPENSATING result and the alpha server could query the
> > @Status
> > > > > method
> > > > > > > to
> > > > > > > > check the compensation status, if this method returns
> > > COMPENSATE_OK,
> > > > > the
> > > > > > > > alpha server will mark the local transaction is compensated
> > > otherwise
> > > > > > > will
> > > > > > > > mark it with compensate_failed.
> > > > > > > >
> > > > > > > > Daniel Qian <[email protected]> 于2019年7月21日周日 下午8:37写道：
> > > > > > > >
> > > > > > > > > I rethink the idea I proposed, yes, provide low level apis
> > is a
> > > bad
> > > > > > > idea,
> > > > > > > > > and I also don't suggest that let user code use
> > > omega-xxx-transport
> > > > > > > api.
> > > > > > > > >
> > > > > > > > > I think the essential issues are three:
> > > > > > > > >
> > > > > > > > > 1. How to pass tx context info across threads
> > > > > > > > > 2. How to asynchronously tell Alpha that Saga is ended or
> > > > > aborted,
> > > > > > > which
> > > > > > > > > means not triggered on @SagaStart method returns.
> > > > > > > > > 3. How to asynchronously tell Alpha that LocalTx is ended
> > or
> > > > > > > aborted,
> > > > > > > > > which means not triggered on @Compensable method returns.
> > > > > > > > >
> > > > > > > > > I think we can keep using @SagaStart @Compensable for the
> > > > > > > XXXStartedEvent,
> > > > > > > > > and provide a helper to manually end/abort Saga/LocalTx.
> > Thanks
> > > > > for PR
> > > > > > > #506
> > > > > > > > > (SCB-1385) we can use TransactionContext to achieve that.
> > Below
> > > is
> > > > > a
> > > > > > > code
> > > > > > > > > sample:
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > @SagaStart(async=true)
> > > > > > > > > public void foo() {
> > > > > > > > > TransactionContext txContext =
> > > > > OmegaContext.getTransactionContext();
> > > > > > > > > someAsyncCall()
> > > > > > > > > .onSuccess(Callback() {
> > > > > > > > > omega.endSaga(txContext);
> > > > > > > > > })
> > > > > > > > > .onException(Callback() {
> > > > > > > > > omega.abortSaga(txContext);
> > > > > > > > > })
> > > > > > > > > }
> > > > > > > > >
> > > > > > > > > @Compensable(async=true, compensationMethod="rollbackBar")
> > > > > > > > > public void bar() {
> > > > > > > > > TransactionContext txContext =
> > > > > OmegaContext.getTransactionContext();
> > > > > > > > > someAsyncCall()
> > > > > > > > > .onSuccess(Callback() {
> > > > > > > > > omega.endTx(txContext);
> > > > > > > > > })
> > > > > > > > > .onException(Callback() {
> > > > > > > > > omega.abortTx(txContext);
> > > > > > > > > })
> > > > > > > > > }
> > > > > > > > >
> > > > > > > > > The async attribute on @SagaStart and @Compensable prevents
> > > > > > > Saga/LocalTx
> > > > > > > > > ended when method returns.
> > > > > > > > > TransactionContext object can be passed around safely because
> > > it's
> > > > > > > > > immutable.
> > > > > > > > > What I have not considered clearly is that how to deal with
> > > > > > > compensation
> > > > > > > > > method if it's also async.
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > Willem Jiang <[email protected]> 于2019年7月20日周六
> > 下午10:30写道：
> > > > > > > > > >
> > > > > > > > > > Yeah, I agree we need provide some low level API for the
> > user
> > > to
> > > > > > > pass.
> > > > > > > > > > In the recent change of SCB-1386, I introduce the
> > > > > TransactionContext
> > > > > > > > > > object which holds the reference of GID and LID, we may add
> > > some
> > > > > > > other
> > > > > > > > > > transaction context information there too.
> > > > > > > > > > If the sub transaction is happened in other JVM, we need to
> > > pass
> > > > > the
> > > > > > > > > > TxContext across the JVM with help of omega-xxx-transport.
> > > > > > > > > >
> > > > > > > > > > We already have some internal API to send the message from
> > > Omega
> > > > > to
> > > > > > > > > > Alpha, I prefer to use annotation instead of expose low
> > level
> > > > > API to
> > > > > > > > > > the user.
> > > > > > > > > >
> > > > > > > > > > Willem Jiang
> > > > > > > > > >
> > > > > > > > > > Twitter: willemjiang
> > > > > > > > > > Weibo: 姜宁willem
> > > > > > > > > >
> > > > > > > > > > On Sat, Jul 20, 2019 at 9:50 PM Daniel Qian <
> > > > > [email protected]>
> > > > > > > > > wrote:
> > > > > > > > > > >
> > > > > > > > > > > After look into SCB-163, SCB-1385 and SCB-1386 I have
> > some
> > > > > > > thoughts on
> > > > > > > > > Saga
> > > > > > > > > > > involved in async invocation.
> > > > > > > > > > > Current implementation is basically based on sync
> > > invocation,
> > > > > > > there are
> > > > > > > > > > > some assumption:
> > > > > > > > > > >
> > > > > > > > > > > 1. When @SagaStart method returns, the Saga finished.
> > > > > > > > > > > 2. When @Compensable method returns/throws exception,
> > the
> > > > > Local
> > > > > > > Tx
> > > > > > > > > > > succeeds/failed.
> > > > > > > > > > > 3. When compensationMethod returns, the Local Tx is
> > > > > compensated.
> > > > > > > > > > >
> > > > > > > > > > > Even if considering what SCB-100 provided:
> > > > > > > > > > >
> > > > > > > > > > > 1. Add @OmegaContextAware annotation enabling
> > > > > > > > > > > java.util.concurrent.Executor inject OmegaConext into
> > > > > threads it
> > > > > > > > > > > manages/spawns
> > > > > > > > > > > 2. Make OmegaContext use InheritableThreadLocal field
> > let
> > > > > child
> > > > > > > > > thread
> > > > > > > > > > > inherit parent thread's Local Tx info
> > > > > > > > > > >
> > > > > > > > > > > There are still some limitations:
> > > > > > > > > > >
> > > > > > > > > > > 1. @OmegaContextAware is only viable if you use spring
> > > > > framework
> > > > > > > > > > > 2. @OmegaContextAware and OmegaContext's
> > > > > InheritableThreadLocal
> > > > > > > > > field
> > > > > > > > > > > assuming that the calling thread or initator thread
> > has
> > > > > Local Tx
> > > > > > > > > info.
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > What if user code use producer-consumer pattern in which
> > > > > > > > > > > InheritableThreadLocal can't work?
> > > > > > > > > > > What if user code use a thread scheduling library which
> > we
> > > > > cannot
> > > > > > > use
> > > > > > > > > > > @OmegaContextAware,RxJava and Reactor, for example?
> > > > > > > > > > > I think we could provide some low-level APIs that user
> > code
> > > can
> > > > > > > manualy
> > > > > > > > > > > starts/ends Saga and Local Tx, something like below:
> > > > > > > > > > >
> > > > > > > > > > > TxContext context = omega.startSaga();
> > > > > > > > > > > TxContext subTxContext = omega.startTx(TxContext
> > > > > parentTxContext);
> > > > > > > > > > > omega.endTx(TxContext);
> > > > > > > > > > > omega.abortTx(TxContext);
> > > > > > > > > > > omega.abortSaga(TxContext);
> > > > > > > > > > > omega.endSaga(TxContext);
> > > > > > > > > > >
> > > > > > > > > > > TxContext is just a immutable dto like this:
> > > > > > > > > > >
> > > > > > > > > > > public class TxContext {
> > > > > > > > > > > private final String globalTxId;
> > > > > > > > > > > private final String localTxId;
> > > > > > > > > > > }
> > > > > > > > > > >
> > > > > > > > > > > Above is a just a rough idea. So any thoughts?
> > > > > > > > > > > --
> > > > > > > > > > > Daniel Qian
> > > > > > > > > > >
> > > > > > > > > > > 博客：https://segmentfault.com/u/chanjarster
> > > > > > > > > > > github：https://github.com/chanjarster
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > --
> > > > > > > > > Daniel Qian
> > > > > > > > >
> > > > > > > > > 博客：https://segmentfault.com/u/chanjarster
> > > > > > > > > github：https://github.com/chanjarster
> > > > > > > > >
> > > > > > >
> > > > >
> > >
> > >
> > >
> > > --
> > > Daniel Qian
> > >
> > > 博客：https://segmentfault.com/u/chanjarster
> > > github：https://github.com/chanjarster
> >
--
Daniel Qian
博客：https://segmentfault.com/u/chanjarster
github：https://github.com/chanjarster
