[
https://issues.apache.org/jira/browse/PHOENIX-3884?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16025835#comment-16025835
]
Lars Hofhansl commented on PHOENIX-3884:
----------------------------------------
KeyValueUtil.length(cell) seems to slightly underestimate the size, though. I
think it misses the memstoreTS stored in the HFiles for each KV. That TS is at
least stored initially, also that TS is store compressed, so hard to measure,
but it would explain the discrepancy in this size and the actual byte sizes on
disk (flushed, fully compacted, no FAST_DIFF, etc)
Still.. This is good enough.
> Correct MutationState size estimation
> -------------------------------------
>
> Key: PHOENIX-3884
> URL: https://issues.apache.org/jira/browse/PHOENIX-3884
> Project: Phoenix
> Issue Type: Bug
> Affects Versions: 4.10.0
> Reporter: Lars Hofhansl
> Assignee: Lars Hofhansl
> Attachments: 3884.txt
>
>
> Currently the Mutation is estimated by called Mutation.heapSize(), which adds
> all the overhead needed to store the Mutation on the Java heap and has little
> to do with the actual size on the wire or the size of disk.
> With a sample row with a 20 byte key and 10 columns with a qualifier length
> and value length of this reports 1800 bytes, where the size is closer to
> 600-700 bytes.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
