In this case when deciding whether this method is right for you you nee to consider performance requirements. This is where streaming topology comes in handy. By allowing logs to leave the node as quickly as possible you can reuce the risk of production application failures. And the bravest ones can increase the maximum number of open files for a process using sysctl sysctl w fs.filemax However we do not recommend this approach. Case No Explosion Prometheus While running on a node Vector collects pod logs and exports metrics such as the number of log lines collecte the number of errors encountere and others.
Duplicates parameter
However many metrics have a label file which is a b for exporters restarting pods on a node results in more and more metrics. That is Vector continues to provide metrics for pods that are no longer in the cluster which is a common problem for all exporters. To solve this problem Pakistan WhatsApp Number List we remove the extra file labels using relabeling rules metric_relabel_configs regex file action labeldrop After this Prometheus functione normally. But after some time we encountere another problem. Vector consume more and more memory to store all the metrics and eventually it ran out.
Logs it also augments
To fix this we use a global option in Vector expire_metric_secs If you set it to for example seconds Vector will check every minute to see if it is still collecting data from Brazil Phone Number List these pods. If not exporting metrics for these files will stop. Although this solution was effective it also affecte other metrics such as the Vector component error metric. Lets look at an example. As you can see from the graph below three errors were initially recorde then four more.
