[ https://issues.apache.org/jira/browse/CASSANDRA-9018?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14377851#comment-14377851 ]
Maxim Podkolzine commented on CASSANDRA-9018: --------------------------------------------- I see, thanks for clarification. Here's I find misleading here: - The two settings are obviously conflicting. I personally find gc_grace_seconds option a better choice, because it gives a sense of control: I can decide the expiration of my data, per table on top of that. - More general: there are several notions dealing with the same problem: grace period, snapshot and backup. In such cases conflicts often arise. - Strong recommended settings ("auto_snapshot: true") require an operator to monitor the disk space. It seems more convenient if Cassandra takes care of its own space. - The log doesn't give enough information about what's going on. Can advise the proper settings for our case: it's perfectly normal that the data is dropped, and it should be kept _just in case_ for some period? When needed we do a full backup ourselves. Are there any potential problems with "auto_snapshot: false"? > Dropped keyspace is not collected > --------------------------------- > > Key: CASSANDRA-9018 > URL: https://issues.apache.org/jira/browse/CASSANDRA-9018 > Project: Cassandra > Issue Type: Bug > Reporter: Maxim Podkolzine > Attachments: cassandra-log.zip > > > As far as I understand when a keyspace is dropped, the data is marked as > tombstone. We expect that after the grace period (all tables are created with > gc_grace_seconds=7200), this data is automatically removed during the > compaction process, which means that keyspace no longer takes any space on > disk. > This is not happening (not after 2 or 24 hours). The log keeps saying "No > files to compact for user defined compaction", keyspace files remain on disk. > It's not clear whether Cassandra is still waiting for certain event, or > decided not to collect the data. > Is there any setting that I missed? Any clues to figure out from the log, > what's the current state. -- This message was sent by Atlassian JIRA (v6.3.4#6332)