Public bug reported:
SRU Justification:
possible race cleaning ct conns too early.
* Explain the bug(s)
There is a possible race between updating established conn timeout
and initial timeout expiring.
* brief explanation of fixes
in normal flow, established conns gets thier timeout extended to a day and
the GC keeps extending them if timeout is below half a day.
but it looks like there is a possible race that the intial timeout expired
before that timeout was extended.
this is a temporary fix and still being investigated how to avoid this
potential race on the timeout.
* How to test
huge amount of traffic, we used ixia traffic generator. during traffic can dump
conntrack
using "conntrack -L" or "cat /proc/net/nf_conntrack" this sometimes helps
reproduce faster.
after conns offloaded stop the traffic and wait few seconds. now check if all
conntrack conns
still offloaded. we noticed sometimes few conns are missing.
* What it could break.
ct conns expiring too soon. so sometimes idle sessions starting traffic again
would have some
first packets not offloaded and conntrack would reoffload the ct conn.
** Affects: linux-bluefield (Ubuntu)
Importance: Undecided
Status: New
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/1929458
Title:
CT: check offload bit on table dump
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux-bluefield/+bug/1929458/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
