My cluster have unexpected high loading even there is no query / indexing
recently. I suspect it is the segment merging.
How can I tell if it is becuase of merging? I tried to look into the
HotThread... but I can't really understand the meaning.
I tried to turn off merging completely. That means set
"indices.store.throttle.type" to none. It seems there is some improvement,
but STILL swa some unexpected high loading even when idle.
My cluster have unexpected high loading even there is no query / indexing
recently. I suspect it is the segment merging.
How can I tell if it is becuase of merging? I tried to look into the
HotThread... but I can't really understand the meaning.
I tried to turn off merging completely. That means set
"indices.store.throttle.type" to none. It seems there is some improvement,
but STILL swa some unexpected high loading even when idle.
I had 9 nodes, all Intel i7, average loading stay under 4 when ES is
'normal'.
I had attached 3 copies of HotThread results, they were captured within 20
mins.
The "Lucene Merge Thread" appears on 2 to 3 nodes, using 50-60% CPU, but as
i mentioned above, the system load-average are over 30!
On Sunday, March 29, 2015 at 3:49:05 PM UTC+8, Mark Walkom wrote:
How high?
ES will do things int he background (mostly around lucene merging), but is
it causing a problem?
On 29 March 2015 at 13:28, Felix Ng <feli...@gmail.com <javascript:>>
wrote:
My cluster have unexpected high loading even there is no query / indexing
recently. I suspect it is the segment merging.
How can I tell if it is becuase of merging? I tried to look into the
HotThread... but I can't really understand the meaning.
I tried to turn off merging completely. That means set
"indices.store.throttle.type" to none. It seems there is some improvement,
but STILL swa some unexpected high loading even when idle.
If you are not sure about merging, you should look for other reasons for
high load. Identify the processes with high activity. Check if your storage
I/O system can keep up.
You can not turn off merging. With indices.store.throttle.type: none you
diasble throttling.
Optimizing by manual operation is always second best and not required
every night.
My cluster have unexpected high loading even there is no query / indexing
recently. I suspect it is the segment merging.
How can I tell if it is becuase of merging? I tried to look into the
HotThread... but I can't really understand the meaning.
I tried to turn off merging completely. That means set
"indices.store.throttle.type" to none. It seems there is some improvement,
but STILL swa some unexpected high loading even when idle.
Apache, Apache Lucene, Apache Hadoop, Hadoop, HDFS and the yellow elephant
logo are trademarks of the
Apache Software Foundation
in the United States and/or other countries.