Elasticsearch增加堆大小

Question

We running Elasticsearch inside a docker container on Amazons ECS. We noticed that the heap slightly increase over time. The first time we noticed it was when it raised above 70% and started to throw away requests (indices.breaker.total.limit).

The thing is that I never seen a decreasing heap, feels fishy!

So far we have increased the instance size, now running a instance with 30G memory. The heap is set to aprox half the memory, ES_HEAP_SIZE=14g (Xmx=Xms=14g).

Someone else that have similar experience? Is it a bug in Elasticsearch? Or only incorrect configurated?

Elasticsearch ver: 1.5.1

> curl localhost:9200/_cat/fielddata?v

id                     host         ip         node            total position deal_status heading.sortorder org_relationlastmodified deal_value deal_probability 1_ZipCodeVisit_0   tag employer_tag 1_CityVisit_0 dateofregistration temperature uniqueId _type quick_ratio org_relation employer_relationlastmodified turnover turnover_per_employee deal_source employer_relation deal_statusdate 1_custom_1466 average_salary_per_employee deal_orderdate 0_NumberOfEmployeesRange_20 1_LegalForm_0 1_custom_1816 0_WorksiteType_100 0_LineOfBusiness_2 equity_ratio profitmargin 0_LineOfBusiness_1 0_CountyVisit_40 0_NumberOfEmployeesRange_22 0_MunicipalityVisit_61 0_LegalForm_110 dividends 1_custom_1744 0_MunicipalityVisit_60 responsiblecoworker result_before_tax
XMTlkdnsToKvMHqgApMBAg 5dc819096905 172.17.0.2 Hitman        729.8mb    8.1mb       1.1mb           261.5mb                    1.7mb    305.3kb          849.1kb           20.9mb 6.4mb        1.3mb        19.3mb             12.3mb          0b  283.7mb 9.6mb       5.1mb      810.5kb                       632.2kb   11.6mb                 4.1mb     150.8kb           566.4kb         568.6kb        34.1kb                       4.2mb        973.5kb                       5.7mb         4.6mb        37.4kb              4.9mb              8.1mb        4.7mb        4.2mb              9.2mb            3.3mb                       4.2mb                802.9kb           3.9mb     4.3mb        37.7kb                  7.5mb               2.4mb               5mb
dHAoWkHMQKSnwAB0KrJRJw 8ffc068518f9 172.17.0.2 Moira Brandon 718.9mb    8.2mb       1.1mb           261.5mb                    1.3mb      124kb          793.3kb           19.6mb 6.4mb          1mb        19.1mb             10.2mb          0b  283.8mb 9.6mb       5.2mb      714.7kb                       791.3kb    8.8mb                 3.4mb          0b           422.6kb          83.9kb        16.8kb                       4.6mb        989.4kb                       5.6mb         4.5mb            0b              5.2mb              7.9mb        4.1mb        4.3mb                9mb            3.2mb                       4.3mb                     0b           3.8mb     4.3mb            0b                  7.1mb               2.5mb             4.4mb

[Update 2016-10-24] We have updated to version 2.4.0 but we still experience the same problem. If I force a GC, the heap is released to about 4%, that's the same value as a fresh instance.

Example for an instance with 73% heap, the jvm mem shows that old one is about 10G, not sure if that's normal

jvm mem heap percent 73%
"young":    "used_in_bytes" : 199026920
"survivor": "used_in_bytes" : 2422528
"old":      "used_in_bytes" : 10754631392

What triggers a GC? Should we let the heap increase above 70%?

Answer 1

This maybe related to this kind-of known behavior in pre 2.X versions, that affect mainly Kibana, but I guess also elasticsearch.

See this github issue : https://github.com/elastic/kibana/issues/5170

It may be the same in your case, which basically boils down to this node issue : https://github.com/nodejs/node/issues/2683

It may also be a configuration in ES that is not ideal. Look for this usual suspect in elasticsearch config:

bootstrap.mlockall: true

Do you have a lot of shards / replicas ? Do you use Kibana as well ?