If on bootup one of our VMs is locked by an backup we safely can
assume that this backup job does not run anymore and that the lock
has no reason anymore and just hinders uptime of services.
As at this time we (the node) have quorum so we may safely assume
that we have a consistent view of the cluster and all our VMs really
belong to us. We just need to ensure that we do not run into an
automatic backup jobs, so execute our code with VZDumps lock or
timeout.
Log in the Task and Sys log that we removed the lock, so that an
admin easily sees that there may be need for cleaning leftovers from
an interrupted backup.
Addresses bug #1024
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
small refactoring in get_filtered_vmlist: save a VMs config in its
own subhash to avoid collisions with other data which we want to save
in the vmid list, for now this is only `type` but in the next patch
I want to save also the class
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
extjs cannot "convert" and id from other fields, so the ids in the
diffstore and the realstore are different and we re-add every element on
every update
to mitigate this, we generate the id (which is "uid:hostname") in the
backend, and simply use it in the frontend
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
since bash does not play well with utf8 encoded input,
eg. pressing ALTGR + M on an english intl keyboard layout int novnc
does not enter the mu symbol but makes bash think you pressed alt+5,
we should use login without authorization when logging in as root
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
use the new keeplocale parameter from run_command and do not delete the
LANG and LANGUAGE variable for the vncshell
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
instead of using get_start_stop_list in a rather hacky way use the
new get_filtered_vmlist method and adapt the loop, as now only one
level is required.
most changes are just an indent shift left as we lost an unnecessary
loop level
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
we used get_start_stop_list for assembling a list of VMIDs for
stopall, startall and migrateall.
While get_start_stop_list did already some filtering we did some
more in the specific startall and stopall calls.
Add a new private helper which does all the filtering and just
assemble the startup order in the get_start_stop_list method.
Move the template and ha managed filtering in the new
get_filtered_vmlist method, this allows us to remove the filtering
from the API calls them self and we do not load the vms config twice
(it was cached, but nonetheless unnecessary)
get_filtered_vmlist is also intended to be used by migrateall in a
next patch
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Instead, pass the HTTP server as last argument to the page formater,
so that we can call $server->create_auth_cookie().
Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>
this is a simple filter which allows us to limit the actions to specific
vmids
this makes it much simpler to start/stop/migrate a range of vms
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
this adds a hastate field to all vms/ct which have ha enabled
we will use this for showing the error state in the tree (in the webgui)
and for the cluster dashboard (to count the error state guests)
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
instead of using
'/dev/${real_dev}'
we use the devpath property directly
also we skip the smart check in the cleanup
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
instead of getting all disks, only get the info
from the one we get as parameter
and use the 'devname' value for the
ceph commands instead of the parameter itself
(this fixes the cciss!cXdY cciss/cXdY mismatch)
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
As some Makefiles in sub directories do not implement the distclean
target, namely:
PVE/Service/Makefile
PVE/CLI/Makefile
This target is broken.
As all other implementations just redirect to the 'clean' target I
do not implement the missing ones but rather remove all such
targets. Keep it just in the top level directory, for consistence
sake with other pve repos, and redirect it there directly to the
clean target.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
removes the check for dir or nfs storage,
because a few lines below we check for
the content type vztmpl
which should only be allowed when we have a
location where we put templates, thus we should
be able to download them there
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
With the new calculation $pe->{maxcpu} was used before being
initialized to zero. Moving the initialization up.
Additionally setting $pe->{cpu} to $entry->{cpu} if maxcpu
is not set seems pointless as with its factor (maxcpu)
initialized to zero it is cancelled out anyway.
we only added the % of the vms in a pool
which lead to wrong results
e.g. having a pool with 3 vms with 4 cores each and a
cpu usage of 50% each (2 cores at 100%)
lead to :
vm1 50%
vm2 50%
vm3 50%
pool 150%
instead we new calculate the percentage for the whole pool
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
also introduces a force parameter to this call
if force is true, the call destroys the ceph pool
even when it is use
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
VZDump->new() dies when a tmpdir or dumpdir is configured
but does not exist. At this point the error is not being
reported via email.
This also moves the instantiation of VZDump into the worker
since new() can now call sendmail() on error.
Additionally rather than only showing a single error if both
tmpdir and dumpdir don't exist, both are included in the
message.
Refactored and now using PVE::QemuConfig and PVE::LXC::Config
Moved the next if.. statements into the corresponding branches
Signed-off-by: Caspar Smit <casparsmit@supernas.eu>
Reviewed-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
allow the since/until parameters which journalctl understands
We use a really simple regex to check a valit date (time) stamp.
This can be done as this API call gets mainly used by the GUI where
we have full controll over what we send and additional journalctl
has already a good timestamp parser which can handle that.
Do not only allow root@pam to admin ceph server as some user do not
want to allow root logins and users with the Sys.Modify permission
should be able to modify ceph related stuff.
We use basically the following permissions:
Sys.Modify:
for any delete, add, modify action (POST, PUT, DELETE)
Sys.Audit and Datastore.Audit:
for any status/information view action (GET)
Sys.Log:
for viewing the Ceph log (was already implemented)
We have two exceptions creating and destroying osds. Those may only
be done by 'root@pam' for security reasons.
Also show users with any of those capabilities the ceph tab in the
web GUI.
Addresses bug#818
If set limit the maximal worker count to the new datacenter.cfg
setting 'max_workers'.
For stopall we prefer this over the cpu count if it's set.
For migrateall we prefer the parameter but allow now to ommit
the parameter and then we use the new setting if set.
if both are not set we throw an error.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Previously we waited 10 seconds, in PVE 4 system sometimes need more
time to initialize all services so the didn't execute the startall
command even if a few seconds later cfs quorum was etablished.
This is a background process started from the pve-manager init
script, thus waiting doesn't interferes with other processes, so
wait long enough.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
The check used was completely wrong and so the startall and stopall
API calls touched HA managed service, which they shouldn't.
Using the vm_is_ha_managed call from the HA stack fixes that.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
When adding the new LXC config format we forgot to adapt the
get_start_stop_list and do s/pve.startup/onboot/
This fixes that qemu VMs were started always before LXC CTs,
irregardless of the defined order.
This also let's us simplify the surrounding code and reduce code
reuse.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
This fix is essential of cause json field changes in the ceph api.
They change the field size in pool to max_avail
and the field total_size in stats to total_max_avail
Added an option to disable a backup job while preserving its
settings. When a job is disabled a '#' is added in front of the vzdump
command in the vzdump.cron file. So the cron job still fires,
but only a comment gets executed. If we would comment out the cron job
a bit more changes to the parser would be needed.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
* do not unlink $pidfile inside stop_running_backups to avoid race conditions
* use pve UPID to save pid (see PVE::Tools::upid_decode)
* allow to pass -stop parameter to nomal backup job
* simple return 'OK' instead of calling exit() inside API call.
* rename stop_all_backups to stop_running_backups
This fix include a new function of vzdump.
Now you can call vzdump -stop and the backup will be aborted.
Also if the pve-manager init script stop the process, vzdump -stop will called.
Signed-off-by: Wolfgang Link <wolfgang@linksystems.org>
Suspend/resume support for VMs has been in the PVE2 API for some time,
but even though vzctl supports suspend/resume (what they call checkpoint/
restore), the API doesn't yet support suspend/resume for CTs. This patch
adds that support.
Signed-off-by: Dan Hunsaker <danhunsaker@gmail.com>
- remove command from index (so that pvesh does not list it)
- use POST instead of GET
- correctly check permissions
- return value with status codes and data (allow expections)
The requirement to have VM.Allocate it non-obvious and confusing
if the VM already exists. If the user can backup, he should also be able
to restore that backup.
Changes since V2:
- swap in / out / transmit / receive again
Changes since V1:
- new return format (use an arrayref instead of a hash to be JS compatible)
- swap in / out / transmit / receive
Signed-off-by: Stefan Priebe <s.priebe@profihost.ag>
