[
https://issues.apache.org/jira/browse/CASSANDRA-10547?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15004449#comment-15004449
]
Susan Perkins edited comment on CASSANDRA-10547 at 11/13/15 6:37 PM:
---------------------------------------------------------------------
* We are just upgrading from 2.0.13 to 2.1 now;
* Each list has about 80 doubles
* I think there is one tombstone for each update, so 80 doubles per updated
list.
In the past, we didn't check the data before updating, we just wrote whether it
was changed or not, so the entire list would be updated, potentially with the
same data. We recently updated the code to check whether there were any
changes. And, it no longer updates: If anything differs it deletes the entire
partition and writes a new one.
was (Author: sperkins):
* We are just upgrading from 2.0 to 2.1 now;
* Each list has about 80 doubles
* I think there is one tombstone for each update, so 80 doubles per updated
list.
In the past, we didn't check the data before updating, we just wrote whether it
was changed or not, so the entire list would be updated, potentially with the
same data. We recently updated the code to check whether there were any
changes. And, it no longer updates: If anything differs it deletes the entire
partition and writes a new one.
> Updating a CQL List many times creates many tombstones
> -------------------------------------------------------
>
> Key: CASSANDRA-10547
> URL: https://issues.apache.org/jira/browse/CASSANDRA-10547
> Project: Cassandra
> Issue Type: Bug
> Environment: Cassandra 2.1.9, Java driver 2.1.5
> Reporter: James Bishop
> Attachments: tombstone.snippet
>
>
> We encountered a TombstoneOverwhelmingException in cassandra system.log which
> caused some of our CQL queries to fail.
> We are able to reproduce this issue by updating a CQL List column many times.
> The number of tombstones created seems to be related to (number of list items
> * number of list updates). We update the entire list on each update using the
> java driver. (see attached code for details)
> Running nodetool compact does not help, but nodetool flush does. It appears
> that the tombstones are being accumulated in memory.
> For example if we update a list of 100 items 1000 times, this creates more
> than 100,000 tombstones and exceeds the default tombstone_failure_threshold.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
