Thank you for explaining the index.translog.durability setting, it's good to know. However, it made no difference. While I am thinking of a way to give you something reproducible, I thought of checking and comparing the NodeStats between the two versions. On clean installs, I have run the same data through version 1.6.0 and 2.1.0. I am attaching the NodeStats for both versions.
There are some interesting differences.
-
Indexing - 1.6.0
"indexing" : {
"index_total" : 86797,
"index_time_in_millis" : 7894,
"index_current" : 0,
"delete_total" : 0,
"delete_time_in_millis" : 0,
"delete_current" : 0,
"noop_update_total" : 0,
"is_throttled" : false,
"throttle_time_in_millis" : 0
},
Indexing - 2.1.0
"indexing" : {
"index_total" : 86797,
"index_time_in_millis" : 10204,
"index_current" : 0,
"index_failed" : 13,
"delete_total" : 0,
"delete_time_in_millis" : 0,
"delete_current" : 0,
"noop_update_total" : 0,
"is_throttled" : false,
"throttle_time_in_millis" : 0
},
I notice the new field "index_failed" which has a non-zero value. How should I interpret that?
-
Merges - 1.6.0
"merges" : {
"current" : 0,
"current_docs" : 0,
"current_size_in_bytes" : 0,
"total" : 12,
"total_time_in_millis" : 1740,
"total_docs" : 216424,
"total_size_in_bytes" : 15248619
},
Merges - 2.1.0
"merges" : {
"current" : 0,
"current_docs" : 0,
"current_size_in_bytes" : 0,
"total" : 13,
"total_time_in_millis" : 4878,
"total_docs" : 278771,
"total_size_in_bytes" : 29956261,
"total_stopped_time_in_millis" : 0,
"total_throttled_time_in_millis" : 0,
"total_auto_throttle_in_bytes" : 62914560
},
This is the most surprising. Although it's the same data, the number of documents has gone up and the size is roughly double. (I repeat that in both cases there config is 1 shard/index, 0 replicas).
Are these giving a hint about what could be happening differently and explaining the slowdown?
The full files can be downloaded at:
https://s3.amazonaws.com/prelert-dimitris/node_stats/node_stats_210.json
https://s3.amazonaws.com/prelert-dimitris/node_stats/node_stats_160.json