Hi, Erick Thanks for you quick replay,
we are using Solr 4.9.0 and use 4 Aliyun cloud instance with 4 core cpu 32G mem and 1G SSD shard distribute as: we have 4 shard Node shard1_0 shard1_1 shard2_0 shard2_1 prmsop01 10.173.225.147 E E prmsop02 10.173.226.78 E E prmsop03 10.173.225.163 E E prmsop04 10.173.224.33 E E and each shard index size is 24G. currently we insert 500 document per second. Thanks. > On Mar 4, 2015, at 12:21, Erick Erickson <erickerick...@gmail.com> wrote: > > It's always important to tell us _what_ version of Solr you are > running. There have > been many improvements in this whole area, perhaps it's already fixed? > > Best, > Erick > > On Tue, Mar 3, 2015 at 6:20 PM, 龚俊衡 <junheng.g...@icloud.com> wrote: >> Hi, >> >> I found when a replica recovering, one of cpu core (usually cpu0) will load >> 100%, and then leader update will fail cause this replica can not response >> leader’s /update command >> >> this will cause leader send other recovery to this replica then this replica >> in a recover loop. >> >> my question is it’s possible to avoid command process thread and recovery >> thread running on different cpu core?