ted-ross commented on a change in pull request #925:
URL: https://github.com/apache/qpid-dispatch/pull/925#discussion_r527038602
##########
File path: src/adaptors/http2/http2_adaptor.c
##########
@@ -1308,9 +1305,28 @@ static void qdr_http_delivery_update(void *context,
qdr_delivery_t *dlv, uint64_
qdr_http2_stream_data_t* stream_data = qdr_delivery_get_context(dlv);
if (!stream_data)
return;
+
+ qdr_http2_connection_t *conn = stream_data->session_data->conn;
+
+ //
+ // DISPATCH-1849: In the case of large messages, the final DATA frame
arriving from the server may or may not
+ // contain the END_STREAM flag. In the cases when the final DATA frame
does not contain the END_STREAM flag,
+ // the router ends up forwarding all the data to the curl client without
sending the END_STREAM to the client. The END_STREAM does arrive from the server
+ // but not before the curl client closes the client connection after
receiving all the data. The curl client
+ // does not wait for the router to send an END_STREAM flag to close the
connection. The client connection closure
+ // triggers the link cleanup on the ingress connection, in turn freeing up
all deliveries and its peer deliveries.
+ // The peer delivery is released while it is still receiving the
END_STREAM frame and the router crashes. To solve this issue,
+ // the stream_data->in_dlv_released flag is set to true when the peer
delivery is released and this flag is
+ // check when performing further actions on the delivery. No action on the
peer delivery is performed
+ // if this flag is set because the delivery and its underlying message
have been freed.
Review comment:
Why don't you use the delivery ref-count mechanism to protect the
delivery rather than introducing an additional flag?
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
