[
https://issues.apache.org/jira/browse/CASSANDRA-9909?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14644785#comment-14644785
]
Jason Brown commented on CASSANDRA-9909:
----------------------------------------
The reason why we reload the seeds is that the nodes can be dynamic - think
ec2, where nodes can come and go. So, proposal #1 is no good (a list that does
not change during the lifetime of the process). For proposal #3, you assume the
seeds are being written out to a file, which is not always the case (netflix's
Priam app provides an implementation of SeedProvider that calls over to Priam
via http on localhost, which keeps a cached notion of the seeds for that node).
Thus, proposal #2 is probably the best.
All this being said, DD.getSeeds() is called incredibly infrequently; in fact,
almost never. If you want to add a timer background task to reload the seeds
from the SeedProvider, I suspect you will actually end up doing more work.
> Configuration is loaded too often during runtime
> ------------------------------------------------
>
> Key: CASSANDRA-9909
> URL: https://issues.apache.org/jira/browse/CASSANDRA-9909
> Project: Cassandra
> Issue Type: Improvement
> Reporter: Robert Stupp
> Priority: Minor
>
> Each call to {{SimpleSeedProvider.getSeeds()}} via
> {{DatabaseDescriptor.getSeeds()}} loads the configuration file from disk.
> This is unnecessary in the vast majority of calls from {{Gossiper}} and
> {{StorageService}}.
> Proposal:
> * Instantiate {{ConfigurationLoader}} once during init of DD (not every time
> in {{loadConfig()}}}
> * Only load configuration once per time interval
> * Only load configuration if config file has changed (file modification
> timestamp) - if applicable (URL resolves to a file)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
