Yeah, We’ve plugins that are in mixed mode and add txn-close hook through C API.
If the plugin is just using CPP API, then yes you are right - it’d not crash but leak. > On May 18, 2020, at 6:15 PM, Walt Karas <wka...@verizonmedia.com.invalid> > wrote: > > Hmm I think if you forgot to call 'Transaction::addPlugin()' you would just > leak the plugin object, I don't think it would cause the crash you saw. > >> On Mon, May 18, 2020 at 6:38 PM Sudheer Vinukonda >> <sudheervinuko...@yahoo.com.invalid> wrote: >> >> We recently ran into a core dump issue due to one of our >> TransactionPlugin's continuations being called back with TXN_CLOSE event. >> The current CPPAPI's TransactionPlugin does not expect a call back on >> TXN_CLOSE, and asserts on it. This is because the expectation seems to be >> that all the continuations for TransactionPlugins attached to the >> Transaction are destroyed in the global TXN_CLOSE hook. However, this only >> works based on the plugin table in the Transaction object, which gets >> populated based on `Transaction::addPlugin()`. Turns out, some of our >> plugins do not call `Transaction::addPlugin()` and just create a >> `TransactionPlugin()` and register hooks to it resulting in them getting >> called back for TXN_CLOSE event as well. >> Long story short, it didn't particularly seem unreasonable to add support >> to TXN_CLOSE event in TransactionPlugin, since the current API does not >> seem to mandate calling `Transaction.addPlugin()`. >> >> https://github.com/apache/trafficserver/pull/6800 >> >> Please let me know if there are any concerns/comments.
