[Bug 237720] tcpip network stack seized for six hours after large high-throughput file transfer

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=237720

            Bug ID: 237720
           Summary: tcpip network stack seized for six hours after large
                    high-throughput file transfer
           Product: Base System
           Version: 12.0-RELEASE
          Hardware: amd64
                OS: Any
            Status: New
          Severity: Affects Only Me
          Priority: ---
         Component: kern
          Assignee: bugs at FreeBSD.org
          Reporter: freebsd at t.lastninja.net

I have a server running FreeBSD 12.0-RELEASE-p3 which has been upgraded over
the years, originally starting its life as FreeBSD 8. The firewall (pf/ipfw)
has never been enabled.

I recently sftp'd a 5GB file from another system to my server at 40-70MB/s and
the transfer completed. Shortly after (about 1-5 minutes) it appears the entire
tcpip network stack seized and several hours later unseized. I lost
connectivity and was unable to SSH to the server. After several hours, the
server came back online. Looking at the logs, I came across several lines like
this:

May  3 09:16:08 redactedhostname sendmail[20886]: x42NFHn5020886:
SYSERR(redacteduser): timeout writing message to [127.0.0.1]

These were of course cron jobs that were failing and so mail was trying to be
sent. Going by the logs, the networking stack seized for 6 hours.

I have experienced this problem before on the same server:
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=218653 . However it is
notable that since that bug has been filed, the server was upgraded with new
hardware. I am now using the igb driver for "Intel 82576 Gigabit Network
Connection" PCI card. However given that connectivity to the loopback went
down, I don't think this issue is related to hardware or a specific NIC driver.

-- 
You are receiving this mail because:
You are the assignee for the bug.