Since pve-container commit
c48a25452dccca37b3915e49b7618f6880aeafb1
the code to get the cpuset controller path lives in pve-commons PVE::CGroup.
Use that and improve the logging in case some error happens in the future.
Such an error will only be logged once per pvestatd run,
so it does not spam the log.
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
This uses the newly introduced PVE::LXC::CGroup's
cpuset_controller_path() method to find the controller path,
so we need to depend on the newer pve-container package.
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
pvestatd will check if the KVM version has changed using
kvm_user_version (which automatically clears its cache if QEMU/KVM
updates), and if it has, query supported CPU flags and broadcast them as
key-value pairs to the cluster.
If detection fails, we clear the kv-store and set up a delay (120s), to not
try again too quickly.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Commit 0dd73a7fec (statd: refactor update_node_status) changed $target
in pvestatd's auto_balloning sub into a variable:
my $target = int($res->{$vmid});
but then uses it in a string as a parameter to the $log function:
$log->("BALLOON $vmid to $target (%d)\n", $target - $current);
This surprisingly causes the variable to be incorrectly converted into a
JSON string by perl's to_json (called in QMPClient after mon_cmd):
{"value":"1234"}
instead of
{"value":1234}
which causes QEMU to report the parameter as invalid:
"Invalid parameter type for 'value', expected: integer"
This behaviour is made even trickier, since $target internally is still
considered more of an 'int' (although that's a weak claim in perl
anyway), showing up without quotes in Dumper et. al. - but the perldoc
for to_json scheds some light:
simple scalars
Simple Perl scalars (any scalar that is not a reference) are the
most difficult objects to encode: this module will encode undefined
scalars as JSON "null" values, scalars that have last been used in a
string context before encoding as JSON strings, and anything else as
number value
So coerce to_json to treat $target as an integer by using it as one and
everything is fine again.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
it seems that we have a reference leak or the like somewhere in the
(graphite?) status plugin, while the recent transaction based update
mechanism made it slightly better, it's still bad with a lot of VMs..
Until we can track that down, or abandon perl for good, avoid to
frequent restarts by allowing statd to grow 15 MB of memory usage
after initial calibration (it's memory usage at the 10th cycle)
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
For now it only handles the plugin registration and the two recently
integrated helpers.
But, this is a prepartation to move the external metrics server
update mechanic from a stateless always-newly-connect-send-disconnect
to a statefull transaction based mechanis; see later patches
keep the PVE::Status::Plugin use in pvestatd, as we read the cfs
hosted status.cfg there, and the parser is defined by the common
status plugin base module.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
include the version as string and as parts, as we do the split
already. Also include the build commit, so if we re-release a ceph
version, we can differ here too.
Use node as key, to make the new entry a bit more general, could be
easily expanded with other infos, if required.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
and use the broadcast when a service is added/removed
we will use 'get_cluster_service' in the future when we generate a list
of services of a specific type
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
We're already limiting CPUs to lxc/cpuset.effective_cpus,
so let's use the highest cpuid from that set as a maximum to
initialize the container count array.
This allows filtering by node in InfluxDB queries, so the statistics
of all virtual guests on a specific nodes can be queried.
While for InfluxDB this is only a tag which does changes where the
data is stored, Graphite - our other status plugin - has no such
mechanics available. If we would add it to the object hierarchy,
e.g.: "qemu.$vmid.$nodename" a migration of a VM would result in two
different datasets.
So avoid breaking setups and omit it for Graphite for now.
Suggested-by: Daniel1108 <danielgallegosanchez@gmail.com>
CC: Daniel1108 <danielgallegosanchez@gmail.com>
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
