[
https://issues.apache.org/jira/browse/CASSANDRA-3141?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13098993#comment-13098993
]
Sylvain Lebresne commented on CASSANDRA-3141:
---------------------------------------------
If we want to be precise, this doesn't work correctly. In the sense that if you
add a column and there is already an existing column with the same name, this
won't compute the serialized size correctly.
Now we could say that this doesn't matter much in the sense that
# If you use SSTSUW in cases where you update the same column a lot, you're
probably doing it wrong.
# Even when that happens, the consequence is that you will 'flush to disk'
more often than you would otherwise. Which ain't necessarily a big deal.
# It is an estimation anyway
That being said, I wonder if this call to serializedSize() is really that
costly. Maybe it adds a little bit of cost when you 'reopen' a row multiple
times, but you are not supposed to do that too much really (if ever).
> SSTableSimpleUnsortedWriter call to ColumnFamily.serializedSize iterate
> through the whole columns
> -------------------------------------------------------------------------------------------------
>
> Key: CASSANDRA-3141
> URL: https://issues.apache.org/jira/browse/CASSANDRA-3141
> Project: Cassandra
> Issue Type: Improvement
> Components: Core
> Affects Versions: 0.8.3
> Reporter: Benoit Perroud
> Priority: Minor
> Fix For: 0.8.6
>
> Attachments: CachedSizeCF.patch
>
>
> Every time newRow is called, serializedSize iterate through all the columns
> to compute the size.
> Once 1'000'000 columns exist in the CF, it becomes painfull to do at every
> iteration the same computation. Caching the size and incrementing when a
> Column is added could be an option.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
