systemd vs. autofs/nfs
by David A. De Graaf
systemd and autofs/nfs are at war and have been ever since systemd
appeared.
Specifically, if machine A has an open connection to machine B
and B goes down or become inaccessible, then A cannot shutdown.
A's shutdown sequence hangs, waiting for B to respond to an unmount
command, which will not/cannot happen.
WAITING WILL NOT HELP.
Only a dirty disconnect will work; there's no way to avoid it.
Here's what I've tried:
- Precede 'shutdown' or 'reboot' or 'systemctl reboot' with
umount -fl -t nfs <each nfs-mounted-filesystem>
(You might expect a forced lazy umount to do the job.
It doesn't work; umount just hangs)
- Change nfs mount option to 'soft' instead of 'hard'.
Edit /etc/nfsmount.conf to say:
Hard=False
Soft=True
(This did change the autofs/nfs mount option, but didn't improve
the tolerance for a dirty disconnect. Shutdown still hangs.)
- Reduce systemd's timeout values from 90 to 10 sec;
edit /etc/systemd/system.conf to say:
DefaultTimeoutStartSec=10s
DefaultTimeoutStopSec=10s
(During shutdown, systemd displays two times; the time it has waited
for an unmount event, and the time limit to wait. Incomprehensibly,
when the time limit is reached, it bumps it up and continues to wait.
Changing the default timeout values seems to accelerate this recycling,
but doesn't do anything for the actual time limit.)
- Change the nfs timeout from 600 (default) to 20 decisec to reduce
the time nfs waits for any response, including umount, before
declaring a timeout. Edit /etc/nfsmount.conf to say:
# Timeo=600
Timeo=20
(The option displayed by the mount command did change, but it did not
help the shutdown hangup.)
- That leaves ONLY TWO WAYS to shutdown machine A, both ugly:
SysReq R E I S U B - which forces a reboot NOW
Hold the Power button - which just stops everything.
This is NOT NEW. Systemd has had this design defect from day one.
SysV never hung on shutdown. It Just Knew when a dirty disconnect
was the only viable way and handled it.
How many more years must we wait?
Have I overlooked something obvious? Is there a way to make systemd
perform the simple function 'shutdown' smoothly, reliably and quickly?
If anyone knows how, I would love to hear it.
--
David A. De Graaf DATIX, Inc. Hendersonville, NC
dad(a)datix.us www.datix.us
"Tolerance becomes a crime when applied to evil."
-- Thomas Mann
7 years, 10 months
gsl-1 and gsl-2
by Patrick Dupre
Hello,
It seems that I need to have both installed gsl-1 and gsl-2 on
a fc24 machine.
How can I install gsl-1 into a "special" directory ?
Can I use rpmbuild ?
Thank for your help.
Regards.
===========================================================================
Patrick DUPRÉ | | email: pdupre(a)gmx.com
Laboratoire de Physico-Chimie de l'Atmosphère | |
Université du Littoral-Côte d'Opale | |
Tel. (33)-(0)3 28 23 76 12 | | Fax: 03 28 65 82 44
189A, avenue Maurice Schumann | | 59140 Dunkerque, France
===========================================================================
7 years, 10 months
gsl-1 and gsl-2
by Patrick Dupre
Hello,
Is it possible to install gsl-1 and gsl-2 with fc24 ?
Thank for your help
===========================================================================
Patrick DUPRÉ | | email: pdupre(a)gmx.com
Laboratoire de Physico-Chimie de l'Atmosphère | |
Université du Littoral-Côte d'Opale | |
Tel. (33)-(0)3 28 23 76 12 | | Fax: 03 28 65 82 44
189A, avenue Maurice Schumann | | 59140 Dunkerque, France
===========================================================================
7 years, 10 months
Simultaneous DHCP and static IP address with NetworkManager
by Clayton Louden
Hi,
I've got a NIC that gets one address via dhcp and another one statically assigned. This is done with Fedora's legacy network service (systemctl start network) and these config files
/etc/sysconfig/network-scripts/ifcfg-enp0s25:
HWADDR=XX:XX:XX:XX:XX:XX
TYPE=Ethernet
BOOTPROTO=dhcp
DEFROUTE=yes
PEERDNS=yes
PEERROUTES=yes
IPV4_FAILURE_FATAL=no
IPV6INIT=yes
IPV6_AUTOCONF=yes
IPV6_DEFROUTE=yes
IPV6_PEERDNS=yes
IPV6_PEERROUTES=yes
IPV6_FAILURE_FATAL=no
NAME=enp0s25
ONBOOT=yes
and
/etc/sysconfig/network-scripts/ifcfg-enp0s25:0:
TYPE=Ethernet
BOOTPROTO=none
IPADDR=192.168.99.4
PREFIX=24
DEFROUTE=yes
IPV4_FAILURE_FATAL=no
IPV6INIT=no
NAME=static-ip
ONBOOT=yes
which works just fine:
# ip addr show
...
2: enp0s25: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc fq_codel state UP group default qlen 1000
link/ether xx:xx:xx:xx:xx:xx brd ff:ff:ff:ff:ff:ff
inet 192.168.1.68/24 brd 192.168.1.255 scope global dynamic enp0s25
valid_lft 6887sec preferred_lft 6887sec
inet 192.168.99.4/24 brd 192.168.99.255 scope global enp0s25:0
valid_lft forever preferred_lft forever
inet6 fe80::2ad2:44ff:fe7d:c66d/64 scope link
valid_lft forever preferred_lft forever
How can I achieve the same with NetworkManager? If I just disable the legacy network service and start NetworkManager with the same set of config files, I'll just get an IP address from the dhcp server (it seems the ifcfg-enp0s25:0 file is simply ignored)
Any pointers are greatly appreciated!
7 years, 10 months
fedup F20 to F24
by JD
Is it advisable? What sort of problems could the user encounter?
Reason for asking: Many locally compiled programs (non-fedora sources)
that compile and build (via rpmbuild) just find under F20, but will not
compile on another machine running F24 (too many dependencies).
P.S: For fun, Just wondering about this: since OS can run both 64 and 32
bit apps (Installation is 64 bit),
and the apps link with the libs in /usr/lib64 or /usr/lib; then,
why not be able to build and
run apps which require libs of older fedora releases?
for example: have /f20/usr/lib and /f20/usr/lib64 on a system
of higher or lower OS release version.
Yes, it would requires a lot of disk space for all those libs of
that specific release, but
disk space is cheap and plenty. I hope some savvy and inspired
developer(s) can pick up on
this idea :) :) It would help keep any fedora machine up to snuff
for multiple releases.
Cheers!
JD
7 years, 10 months
Date change on reboot
by Stephen Davies
My F22 randomly crashes to a frozen state that requires a power cycle to restart.
When this happens, the time stamp on log files such as /var/log/messages
changes for a while before reverting to the correct date and time.
Today,the sequence was:
Jul 24 10:55:55 mustang rsyslogd: [origin software="rsyslogd"
swVersion="8.8.0" x-pid="853" x-info="http://www.rsyslog.com"] start
Jul 15 11:19:20 mustang systemd: Stopped target Default.
Jul 15 11:19:25 mustang systemd: Stopping Default.
followed by some 17,000 lines with Jul 15 through to:
Jul 15 19:28:11 mustang named[1221]: client 122.61.143.143#39431 (isc.org):
query (cache) 'isc.org/ANY/IN' denied
Jul 24 10:56:14 mustang rsyslogd-2177: imjournal: begin to drop messages due
to rate-limiting
Sun Jul 24 10:56:16 ACST 2016 rc.fw
Jul 24 11:06:11 mustang rsyslogd-2177: imjournal: 2971956 messages lost due to
rate-limiting
A similar thing happens in /var/log/secure (though with fewer Jul 15 entries).
The Jul 15 dates have been appearing at/after the last two or three crashes.
Even though the Jul 15 entries look to span over eight hours, the actual time
span was only a minute or so.
Any suggestions as to what is happening here?
Cheers and thanks,
Stephen
7 years, 10 months