Details
-
Task
-
Status: Resolved
-
Critical
-
Resolution: Fixed
-
2.0.0
-
None
Description
https://issues.apache.org/jira/browse/AMBARI-8885 fixes a problem where agents required a restart in order for the alert framework to pickup configuration changes.
Initially, alerts were designed so that they only received the parameters they requested. A single alert definition might only need a couple of values (such as hdfs-site/foo) as opposed to the entire configuration structure.
However, it seems like we're doing a lot of extra work to ensure that these values are kept current; alert definitions need to be rescheduled when a configuration changes so that its cached value can be updated.
It would be good to investigate (and implement if plausible) a way to store the entire configuration structure in memory and have it accessible to alerts (and any other part of the agent framework). This would allow us to:
- Remove the code that caches values in the alert jobs before they are schedule
- Remove the code the restarts jobs on configuration changes
- Have an up-to-date configuration structure that is easily accessible without the need to parse any files on-disk
The major concern would be that keeping such a large structure in memory could cause some kind of performance degradation or memory overhead that would be unacceptable.
Attachments
Attachments
Issue Links
- links to