solr slave's performance issue after replicate the optimized index

[虞冰 <ic...@gmail.com>]

Hi all,

I have a performance issue~

I do a optimize on solr master every night.
but about a month ago, every time after the slaves get the new
optimized index, system cpu usage will raise from 0.3 - 0.5% to 7 -
10% (daily average), and servers's load average also become 2 times
more than normal.the load average remain high even I restart the
tomcat .

after many day's testing, I find that 4 ways to bring the slaves back
to normal load average.

1. reboot linux server
2. shutdown tomcat, manually rm the index data and do repilcate again
3. shutdown tomcat, copy indexdata as indexdata2, rm indexdata, mv
indexdata2 to indexdata, start tomcat
4. shutdown tomcat, use C to alloc 20G memory and free it, start server.

I can only guess it has some relationship with the memory or  the system cache.

Is this a solr bug or lucence bug or just system issue?


My System:

CentOS 5.6 X64   Tomcat 7.0 JRocket 6
Intel E5620 *2  24GB DDR3
Solr 3.1
Index size 7G (after optimize)  / 8G (before opitimize)


Many thanks~