Always try to do a ‘"warm’restart" if possible, to avoid the startup bubble of whole cluster restarting at same time but the Agents not being restarted.
- ‘Warm’ restart means restarting one Collector at a time until all Collectors in the Cluster have been restarted. Only then, restart the MOM. In between each EM restart, allow enough time for the rebalancing to kick in and maintain a stabilized cluster with even load.
- A ‘warm’ restart can be performed for simple restart without any upgrade and for most HF deployments.
- Also make sure that the metric/agent clamps on all Collectors are set high enough so that no Collector hits the clamp during the restart process.
If for whatever reason, a ‘cold’ restart cannot be avoided, try to restart all Collectors and MOM at same time to minimize the time period that only one or two Collectors/MOM is up and then swamped by all Agent connections.