elasticsearch 写入优化

轩辕阁

　　全量dump数据时，为优化性能，可做如下优化。

• 分片设置，不分片
  

http://localhost:9200/test_index/_settings/
{
"index": {
"number_of_replicas": 0
}
}

• 刷新设置，不刷新
  

http://localhost:9200/test_index/_settings/
{
"index": {
"refresh_interval": "-1"
}
}

• translog 大小设置，调大 默认512M
  

http://localhost:9200/test_index/_settings/
{
"index.translog.flush_threshold_size": "1024mb"
}

• 如果是SSD硬盘，修改段合并速率
  

http://localhost:9200/_cluster/settings/
{
"persistent": {
"indices.store.throttle.max_bytes_per_sec": "200mb"
}
}
http://localhost:9200/_cluster/settings/
{
"transient": {
"indices.store.throttle.type": "none"
}
}
　　全量dump后，再恢复一下

• 分片设置
  

http://localhost:9200/test_index/_settings/
{
"index": {
"number_of_replicas": 2
}
}

• 刷新设置，不刷新
  

http://localhost:9200/test_index/_settings/
{
"index": {
"refresh_interval": "1s"
}
}

• translog 大小设置
  

http://localhost:9200/test_index/_settings/
{
"index.translog.flush_threshold_size": "512mb"
}
　　当然应该使用bulk模式， elasticsearch单次提交数据尽量在15M以内。