[
https://issues.apache.org/jira/browse/BLUR-5?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13474735#comment-13474735
]
Aaron McCurry commented on BLUR-5:
----------------------------------
I think that the simple implementation is pretty straight forward, as you write
the blocks to hdfs write them into the cache. I think that interesting piece
is going to be controlling the write-through cache during merges, so that large
merges does not unload everything from the cache. Also limiting cold spots in
index (the parts of the index that are rarely used) from forcing out needed
cache data, I think the best solution for controlling that behavior is moving
to a LIRS caching schema instead of a the simple LRU.
Also in the Lucene 4.0 API the IOContext in the Directory provides some hints
as to what kind of action is being performed e.g. merge, search, etc. I have a
working version of BlockCache and a simple hdfs directory using Lucene 4.0 in
the blur-new-api-prototype project on the new-api-prototype branch. I would
start there.
> Write through caching for the BlockCache
> ----------------------------------------
>
> Key: BLUR-5
> URL: https://issues.apache.org/jira/browse/BLUR-5
> Project: Apache Blur
> Issue Type: Improvement
> Reporter: Aaron McCurry
>
> This will allow for better NRT update performance because the writer will not
> have to read the NRT segments from HDFS.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
