[
https://issues.apache.org/jira/browse/CAMEL-8424?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Piotr Klimczak updated CAMEL-8424:
----------------------------------
Description:
I found that when we are using exactly same propagation policy bean in
different routes used together with "direct" component, then
TransactionErrorHandler always propagates current transaction even if our
policy is "PROPAGATION_REQUIRES_NEW".
The failing code is:
{code}
public TransactionErrorHandler(CamelContext camelContext, Processor output,
CamelLogger logger,
Processor redeliveryProcessor, RedeliveryPolicy redeliveryPolicy,
ExceptionPolicyStrategy exceptionPolicyStrategy,
TransactionTemplate transactionTemplate, Predicate retryWhile,
ScheduledExecutorService executorService,
LoggingLevel rollbackLoggingLevel) {
super(camelContext, output, logger, redeliveryProcessor,
redeliveryPolicy, null, null, false, retryWhile, executorService);
setExceptionPolicy(exceptionPolicyStrategy);
this.transactionTemplate = transactionTemplate;
this.rollbackLoggingLevel = rollbackLoggingLevel;
this.transactionKey =
ObjectHelper.getIdentityHashCode(transactionTemplate);
}
@Override
public void process(Exchange exchange) throws Exception {
// we have to run this synchronously as Spring Transaction does *not*
support
// using multiple threads to span a transaction
if (exchange.getUnitOfWork().isTransactedBy(transactionKey)) {
// already transacted by this transaction template
// so lets just let the error handler process it
processByErrorHandler(exchange);
} else {
// not yet wrapped in transaction so lets do that
// and then have it invoke the error handler from within that
transaction
processInTransaction(exchange);
}
}
{code}
So then for each policy there is a hash code created, which then is used to
verify whether current route is already transacted by this transaction policy.
This makes "PROPAGATION_REQUIRES_NEW" ignored when used with "direct" component.
So for example:
{code}
from("activemq:queue:start").routeId("route1)
.transacted("PROPAGATION_REQUIRES_NEW")
.setExchangePattern(ExchangePattern.InOnly)
.to("activemq:queue:result1")
.to("direct:route2")
.throwException(new RuntimeException("Expected!"));
from("direct:route2").routeId("route2)
.transacted("PROPAGATION_REQUIRES_NEW")
.setExchangePattern(ExchangePattern.InOnly)
.to("activemq:queue:result2");
{code}
The above route suppose to work in 2 different transactions, as our propagation
is REQUIRES_NEW. But due to hash code verification and optimisation, route2
will participate in same transaction as route1 instead of new.
This is rather buggy.
Will create pull request in minutes.
was:
I found that when we are using exactly same propagation policy bean in
different routes used together with "direct" component, then
TransactionErrorHandler always propagates current transaction even if our
policy is "PROPAGATION_REQUIRES_NEW".
The failing code is:
{code}
public TransactionErrorHandler(CamelContext camelContext, Processor output,
CamelLogger logger,
Processor redeliveryProcessor, RedeliveryPolicy redeliveryPolicy,
ExceptionPolicyStrategy exceptionPolicyStrategy,
TransactionTemplate transactionTemplate, Predicate retryWhile,
ScheduledExecutorService executorService,
LoggingLevel rollbackLoggingLevel) {
super(camelContext, output, logger, redeliveryProcessor,
redeliveryPolicy, null, null, false, retryWhile, executorService);
setExceptionPolicy(exceptionPolicyStrategy);
this.transactionTemplate = transactionTemplate;
this.rollbackLoggingLevel = rollbackLoggingLevel;
this.transactionKey =
ObjectHelper.getIdentityHashCode(transactionTemplate);
}
@Override
public void process(Exchange exchange) throws Exception {
// we have to run this synchronously as Spring Transaction does *not*
support
// using multiple threads to span a transaction
if (exchange.getUnitOfWork().isTransactedBy(transactionKey)) {
// already transacted by this transaction template
// so lets just let the error handler process it
processByErrorHandler(exchange);
} else {
// not yet wrapped in transaction so lets do that
// and then have it invoke the error handler from within that
transaction
processInTransaction(exchange);
}
}
{code}
So then for each policy there is a hash code created, which then is used to
verify whether current route is already transacted by this transaction policy.
This makes "PROPAGATION_REQUIRES_NEW" ignored when used with "direct" component.
So fo example:
{code}
from("activemq:queue:start").routeId("route1)
.transacted("PROPAGATION_REQUIRES_NEW")
.setExchangePattern(ExchangePattern.InOnly)
.to("activemq:queue:result1")
.to("direct:route2")
.throwException(new RuntimeException("Expected!"));
from("direct:route2").routeId("route2)
.transacted("PROPAGATION_REQUIRES_NEW")
.setExchangePattern(ExchangePattern.InOnly)
.to("activemq:queue:result2");
{code}
The above route suppose to work in 2 different transactions, as our propagation
is REQUIRES_NEW. But due to hash code verification and optimisation, route2
will participate in same transaction as route1 instead of new.
This is rather buggy.
Will create pull request in minutes.
> Transaction propagated ignoring REQUIRES_NEW when using direct component
> ------------------------------------------------------------------------
>
> Key: CAMEL-8424
> URL: https://issues.apache.org/jira/browse/CAMEL-8424
> Project: Camel
> Issue Type: Bug
> Components: camel-core
> Affects Versions: 2.10.7, 2.11.4, 2.13.3, 2.14.1, 2.15.0
> Reporter: Piotr Klimczak
>
> I found that when we are using exactly same propagation policy bean in
> different routes used together with "direct" component, then
> TransactionErrorHandler always propagates current transaction even if our
> policy is "PROPAGATION_REQUIRES_NEW".
> The failing code is:
> {code}
> public TransactionErrorHandler(CamelContext camelContext, Processor
> output, CamelLogger logger,
> Processor redeliveryProcessor, RedeliveryPolicy redeliveryPolicy,
> ExceptionPolicyStrategy exceptionPolicyStrategy,
> TransactionTemplate transactionTemplate, Predicate retryWhile,
> ScheduledExecutorService executorService,
> LoggingLevel rollbackLoggingLevel) {
> super(camelContext, output, logger, redeliveryProcessor,
> redeliveryPolicy, null, null, false, retryWhile, executorService);
> setExceptionPolicy(exceptionPolicyStrategy);
> this.transactionTemplate = transactionTemplate;
> this.rollbackLoggingLevel = rollbackLoggingLevel;
> this.transactionKey =
> ObjectHelper.getIdentityHashCode(transactionTemplate);
> }
> @Override
> public void process(Exchange exchange) throws Exception {
> // we have to run this synchronously as Spring Transaction does *not*
> support
> // using multiple threads to span a transaction
> if (exchange.getUnitOfWork().isTransactedBy(transactionKey)) {
> // already transacted by this transaction template
> // so lets just let the error handler process it
> processByErrorHandler(exchange);
> } else {
> // not yet wrapped in transaction so lets do that
> // and then have it invoke the error handler from within that
> transaction
> processInTransaction(exchange);
> }
> }
> {code}
> So then for each policy there is a hash code created, which then is used to
> verify whether current route is already transacted by this transaction policy.
> This makes "PROPAGATION_REQUIRES_NEW" ignored when used with "direct"
> component.
> So for example:
> {code}
> from("activemq:queue:start").routeId("route1)
> .transacted("PROPAGATION_REQUIRES_NEW")
> .setExchangePattern(ExchangePattern.InOnly)
> .to("activemq:queue:result1")
> .to("direct:route2")
> .throwException(new RuntimeException("Expected!"));
> from("direct:route2").routeId("route2)
> .transacted("PROPAGATION_REQUIRES_NEW")
> .setExchangePattern(ExchangePattern.InOnly)
> .to("activemq:queue:result2");
> {code}
> The above route suppose to work in 2 different transactions, as our
> propagation is REQUIRES_NEW. But due to hash code verification and
> optimisation, route2 will participate in same transaction as route1 instead
> of new.
> This is rather buggy.
> Will create pull request in minutes.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
