James Peach created TS-4928:
-------------------------------
Summary: transactions should not destroy sessions
Key: TS-4928
URL: https://issues.apache.org/jira/browse/TS-4928
Project: Traffic Server
Issue Type: Bug
Components: Core
Reporter: James Peach
In {{Http1ClientSession}} we find the following code:
{noformat}
void
Http1ClientTransaction::transaction_done()
{
current_reader = NULL;
// If the parent session is not in the closed state, the destroy will not
occur.
if (parent) {
parent->destroy();
}
}
{noformat}
the model, as I understand it, is that sessions own transactions, so it is
quite unexpected for the transaction to reach up an kill its parent. It is a
very surprising side-effect of {{transaction_done}} and means that this can
only be reliably used in the specific context of the calling code.
Additionally, why isn't the parent cleared in {{destroy()}}? If it must be
NULL, we should have an assertion.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
