Large spikes for throughput metrics at beginning of job

[moebiusband73]

This ticket was migrated from cc-backend: ClusterCockpit/cc-backend#86

There are spikes with impossible high values, especially for Likwid metrics, at the start of jobs.

Solution:
A quick fix would be to filter out values that are impossibly high, especially at the stat of Jobs.
On the long term the real cause for those spikes has to be investigated.

[TomTheBear]

The issue is caused by permission changes of the likwid.lock file. Only the owner of the lock file is allowed to use the LIKWID library. In order to get these permission changes early, the LikwidCollector uses fsnotify. The fsnotify callback did not only fire when actual owner changes happen but anytime a chown/chmod is executed on the file, even without any changes.

The fundamental problem is in the LIKWID library and the LIKWID team is working on it.