srt: [BUG] srt-live-transmit sender stuck, repeatedly
Describe the bug We are running srt-live-transmit in PULL mode and have a customer connecting to a password protected stream. The past 2 weeks we have been plagued by disconnects and the inability of our customer to reconnect (or at least receive data).
This is then only resolved by us restarting the srt-live-transmit application
The following logmessage at the start of the loss period suggests this is triggered by a ack pkt with a sequence number of -1
17:36:41.963749/SRT:RcvQ:w1*E:SRT.gg: @352733066:ATTACK/IPE: incoming ack seq -1 exceeds current 1863400545 by 284083101!
The code for that logmessage can be found from line 8145 in srtcore/core.cpp.
To Reproduce Steps to reproduce the behavior: -setup a password protected srt-live-transmit in listening mode that sends an mpegts stream -connect to it (customer srt: “haivision srt 1.4”) -mix in invalid seq’s in the ackstream (assuming this is the cause) -see the srt connection breakdown
Expected behavior Invalid sequence data should not destroy a srt session in such a way that a restart of the sending application is required.
About this issue
- Original URL
- State: closed
- Created 4 years ago
- Comments: 15
Ok, I found a potential problem. Looxlike the code in the old UDT didn’t predict the situation that the loss list can be modified by withdrawing loss records, while this can happen in SRT due to TLPKTDROP, and the data in the pcap show that it indeed does happen.