init.d/vz*: unset io limits before stopping CTIn addition to removing cpulimit just before stopping a container,
now it also makes sense to remove its io and iops limits.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
init.d/vz-redhat: don't reset cpulimits for all CTsThis is a fix to commit 8fa8a6193. The intention was
to reset cpulimit just before stopping a CT.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
Add iolimit and iopslimitThis patch adds iolimit and iopslimit support. This functionality
is available for OpenVZ since kernel 2.6.32-042stab084.2.
Work was sponsored by GleSYS Internet Services AB (http://glesys.se/)
via a feature request form available at bit.ly/YtWwWN
* vzctl set: add --iolimit and --iopslimit
* bash_completion: add appropriate completions
* vzlist: add appropriate new fields
* vzctl(8): describe...
vzevent-stop: check for suspend/chkpntThis script check if the stop event was caused by vzctl
and exits in case it's so. Apparently, stop even can be
caused not only by vzctl stop and vzctl restart, but also
vzctl suspend and vzctl chkpnt, so let's check for those
as well.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vzctl create, vzctl exec: do skip fsckfsmount() function is called from vps_postcreate() and vps_run_script(),
and its last argument, fsck, is a binary flag passed on to libploop
directly, with value of 0 meaning to skip fsck, any other value to run fsck.
So, using SKIP_FSCK flag is a mistake here, we should just use 0.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vzctl console: hack to force redraw on reattachSince vzctl console is persistent, there might be a case when say vim
or bash is left on the console when we exit from it. The problem is
when we use vzctl console for the second time, nothing is re-drawn
despite the fact that we issue TIOCSWINSZ ioctl (which leads to SIGWINCH
being sent to an app running on that console). So second vzctl console
shows us empty screen. It would be nice to force...
vzoversell: addThis simple script aims in finding out whether a system is oversold
in terms of RAM and swap. In other words, whether the sum of RAM+swap
limits for all container exceeds available RAM and swap.
Note that overselling RAM is OK since swap is used, while overselling
RAM+swap is not, since this could lead to OOM killing some processes.
NOTE this version is only usable for VSwap containers. For n...
Fix quota on ploop for RHEL5 CTThis commit fixes two issues preventing in-container disk quota
to work inside a RHEL/CentOS 5 container with ploop layout.
First, ploop uses ext4 new quota format (vfsv0) which quota tools
should deal with. Support for this format appeared in quota-3.16,
so commit ebbf339 adds a check for it.
Red Hat backported this support to their quota tools, so in RHEL5
quota-3.13-4.el5 already supports ...
setver.sh: introduce build idIf we build a package for the second time without doing any git commits,
add a build id of .1 as a suffix to RPM release fields. Remove the suffix
once git release has changed.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vzlist(8): fix indentationBeancounter and disk quota suffixes description are not
related to -L option, so they should not be indented under it.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vzubc: don't show unlimited ubcs, add -vNow when most people use VSwap (or so I hope), there is no need
to show the beancounters that are set to unlimited. With this
patch, unlimited UBCs are not shown.
In non-VSwap mode, all UBCs are limited so they will still be
shown as before.
In VSwap mode, though, the output will be smaller thus simpler
to grasp.
Option -v|--verbose is added to switch vzubc to the old behavior,
i.e. show all...
init.d/vzeventd-redhat: switch to strict bash... and fix the following checkbashisms warnings:
possible bashism in etc/init.d/vzeventd-redhat.in line 55 ($EUID should
be "$(id -u)"):
[ "$EUID" != "0" ] && exit 4
possible bashism in etc/init.d/vzeventd-redhat.in line 66 ($"foo" should
be eval_gettext "foo"):
echo -n $"Starting $prog: "
possible bashism in etc/init.d/vzeventd-redhat.in line 75 ($EUID should
be "$(id -u)"):
[ "$EUID" != ...
init.d/vzeventd: set reboot_eventThe new RHEL6 kernel 042stab082.3, adds the following change:
* ext4: send vzevent on error (PSBM-20496, PSBM-21010)
This change makes ext4 kernel module depend on vzevent, which is
loaded as a dependency. Therefore, it is loaded with defaults,
including reboot_event=0, and then init.d/vzeventd warns about it:
/etc/init.d/vzeventd start
Checking vzevent kernel module ... ...
Revert "init.d/vzevent-redhat: support for RHEL4"There is no need to support RHEL4 kernel anymore, it was
frozen in November 2012.
This reverts commit 7d4c629f810fdb047298a1db442f0dff5c7adcb2.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vz-postinstall: don't add bridge params to sysctl.confThis enhances commit a7aaea0. Since the default for these
net.bridge.bridge-nf-call-* parameters are 1, it doesn't make
sense to add those, we only need to change them to 1 if they
are there and set to 0 (RHEL case).
This should fix the following error message for distros that do not
load bridge by default:
# sysctl -p
error: "net.bridge.bridge-nf-call-ip6tables" is an unknown key
error: "...
Don't modify sysctl.conf on installationAs suggested in the bug report linked below, installing vzctl rpm should
not modify /etc/sysctl.conf, plus there should be a way to totally
prevent it. Still, it's better to disable selinux when we install, i.e.
before a reboot.
So, what this patch does is:
1. Splits vz-postinstall functionality into 'sysctl' and 'selinux'
pieces.
2. Call vz-postinstall selinux from vzctl.rpm on a fresh insta...
fill_vswap_ub(): fix a segfaultCommit 044ff4e was setting oomguar and vmguar limits even if the
memory for those parameters were not allocated.
Let's just have separate barrier and limit in SET macro.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vzctl(8), ctid.conf(5): document vm_overcommitSame as in https://openvz.org/VSwap#Implicit_UBC_parameters
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
Don't save autocalculated secondary UBCsWhen in vswap mode we auto-calculate secondary UBCs,
we should not save them. So, we need a copy of ubc_param.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
[vswap] set calculated oomguar and vmguar limit to LONG_MAXWhen secondary UBC parameters are calculated in vswap mode based
on ram and swap, their barrier and limit are set to the same
value. In fact it's wrong, limit of oomguarpages and vmguarpages
should be set to unlimited.
From http://openvz.org/vmguarpages:
> The meaning of the limit for the vmguarpages parameter is
> unspecified in the current version and should be set to the maximal
> allowed ...
vzctl set: add --vm_overcommitAlso, add the new option to bash completion script.
This is an addition to commit 89fee19.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vswap: set secondary UBCs if unsetApparently, if some UBC parameters are not set in VSwap mode,
it makes sense to set those implicitly, based on 'ram' and 'swap'.
lockedpages = ram
oomguarpages = ram
vmguarpages = ram + swap
Also, if vm_overcommit is set:
privvmpages = (ram + swap) * vm_overcommit
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
Implement VM_OVERCOMMITThis patch adds VM_OVERCOMMIT configuration parameter
and --vm_overcommit option for vzctl set.
To be used by the next commit.
To be documented.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vzctl set --reset_ub: only allow for running CTOn a non-running container, --reset_ub doesn't make sense,
so print an error and return proper exit code.
Before this patch, this was silently ignored.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
parse(): add missing break statementThere was no bug since the next case statement consists of just break,
nevertheless let's add an explicit break.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
set_ublimit(): don't set unknown UBs to unlimThis basically reverts commit bfabefd. Apparently it was needed
for some very old 042* kernel, and was not taking into account
the fact that set_ublimit() is also called from the vzctl set
command line for a running container, when not all UBC parameters
are specified.
This fixes the following bug: when vzctl set --physpages is called
for a running container, it sets all the beancounters that ...
Kir Kolyshkin