You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
If a single Linux host is running multiple system-level containers (e. g. systemd-nspawn, not docker) and both the host and the containers are running Netdata with streaming, a significant chunk of metrics will be duplicated between the instances, severely impacting totals on the "central" Netdata dashboard.
For example, if a 32G RAM host is running 10 containers and all of those run Netdata (in addition to the host itself), then the total RAM usage graph will show 352G.
Description
It would be useful to have a configuration template or a global switch that changes all relevant defaults that would disable collection of all metrics that do not make sense to collect the container (assuming that the container host is also collecting those).
Or, in other words, a method to disable collection of all metrics that are not namespaced in Linux (such as RAM usage metrics, CPU usage or interrupts statistics).
Importance
nice to have
Value proposition
For container hosts running Netdata both outside and inside of containers:
Reduced resource usage due to not collecting duplicated metrics
Improved usability of the "totals" dashboard if streaming is used
Proposed implementation
All metrics that are being collected in Netdata would need to be investigated and separated into one of two categories: "namespaced" and "not namespaced". For example, network interface statistics are namespaced, while kernel RAM usage and CPU interrupts statistics are inherently not namespaced.
Then, either a configuration file template should be provided that disables the collection of all "not namespaced" statistics, or a single global switch that changes the defaults for all "not namespaced" statistics to "off".
The text was updated successfully, but these errors were encountered:
intelfx
changed the title
[Feat]: configuration template or "container mode" switch to disable collection of non-namespaced metrics
[Feat]: "container mode" switch or config template to disable collection of non-namespaced metrics
Mar 9, 2024
Problem
If a single Linux host is running multiple system-level containers (e. g. systemd-nspawn, not docker) and both the host and the containers are running Netdata with streaming, a significant chunk of metrics will be duplicated between the instances, severely impacting totals on the "central" Netdata dashboard.
For example, if a 32G RAM host is running 10 containers and all of those run Netdata (in addition to the host itself), then the total RAM usage graph will show 352G.
Description
It would be useful to have a configuration template or a global switch that changes all relevant defaults that would disable collection of all metrics that do not make sense to collect the container (assuming that the container host is also collecting those).
Or, in other words, a method to disable collection of all metrics that are not namespaced in Linux (such as RAM usage metrics, CPU usage or interrupts statistics).
Importance
nice to have
Value proposition
For container hosts running Netdata both outside and inside of containers:
Proposed implementation
All metrics that are being collected in Netdata would need to be investigated and separated into one of two categories: "namespaced" and "not namespaced". For example, network interface statistics are namespaced, while kernel RAM usage and CPU interrupts statistics are inherently not namespaced.
Then, either a configuration file template should be provided that disables the collection of all "not namespaced" statistics, or a single global switch that changes the defaults for all "not namespaced" statistics to "off".
The text was updated successfully, but these errors were encountered: