Setup
1 Node cluster on my local laptop: 8core, Xms=8G, Xmx=8G
专注于为中小企业提供做网站、成都网站建设服务,电脑端+手机端+微信端的三站合一,更高效的管理,为中小企业临河免费做网站提供优质的服务。我们立足成都，凝聚了一批互联网行业人才，有力地推动了上1000家企业的稳健成长，帮助中小企业通过网站建设实现规模扩充和转变。
Indexing performance (Single index):
10 million payments, each one about 5KB, with batch size = 10000. Each batch takes roughly 2.5 s → 4 s, total time to index 10 million payment is around 50 min
Indexing performance (Multiple indices):
20 separate indices store totally 10 million payments. Indexing execution is slightly faster than single index case. Each batch takes roughly 1.7 s → 3.8 s, total time to index 10 million payment is around 38 min
Parameters required for bulk load operation
Elasticsearch config: http.max_content_length: 500mb
Client time out adjustment:
RestClient.builder(HttpHost("localhost", 9200))
.setRequestConfigCallback {
it.apply {
this.setConnectTimeout(5000)
this.setSocketTimeout(60000)
}
}.setMaxRetryTimeoutMillis(60000))Initially batch size is set to 100000, elastic search server becomes unstable with high GC frequency, occupying a large percent of CPU time. So larger batch size does not always imply higher performance
Query aggregation performance:
Test query: real aggregation query used by rule engine
{
"aggregations": {
"date_range": {
"range": {
"field": "createdAt",
"ranges": [
{
"key": "LAST_7_DAYS",
"from": 1544400968485,
"to": 1545005768486
}
],
"keyed": false
},
"aggregations": {
"filter_aggregator": {
"filters": {
"filters": {
"602c7d66-e990-4dfb-b6e2-72b62ff159d5": {
"terms": {
"beneficiaryId.keyword": [
"602c7d66-e990-4dfb-b6e2-72b62ff159d5"
],
"boost": 1
}
},
"67cab0c8-2510-443d-8f00-bce19c04815e": {
"terms": {
"bankAccountUserId.keyword": [
"67cab0c8-2510-443d-8f00-bce19c04815e"
],
"boost": 1
}
},
"8da52e51-eabf-4f6c-b9f0-e222933c1cb7": {
"terms": {
"payerId.keyword": [
"8da52e51-eabf-4f6c-b9f0-e222933c1cb7"
],
"boost": 1
}
},
"8da52e51-eabf-4f6c-b9f0-e222933c1cb7_602c7d66-e990-4dfb-b6e2-72b62ff159d5": {
"bool": {
"filter": [
{
"terms": {
"payerId.keyword": [
"8da52e51-eabf-4f6c-b9f0-e222933c1cb7"
],
"boost": 1
}
},
{
"terms": {
"beneficiaryId.keyword": [
"602c7d66-e990-4dfb-b6e2-72b62ff159d5"
],
"boost": 1
}
}
],
"adjust_pure_negative": true,
"boost": 1
}
},
"9a1b4bad-ccf5-4c67-8718-02696cb351e4": {
"terms": {
"clientId.keyword": [
"9a1b4bad-ccf5-4c67-8718-02696cb351e4"
],
"boost": 1
}
},
"9a1b4bad-ccf5-4c67-8718-02696cb351e4_602c7d66-e990-4dfb-b6e2-72b62ff159d5": {
"bool": {
"filter": [
{
"terms": {
"clientId.keyword": [
"9a1b4bad-ccf5-4c67-8718-02696cb351e4"
],
"boost": 1
}
},
{
"terms": {
"beneficiaryId.keyword": [
"602c7d66-e990-4dfb-b6e2-72b62ff159d5"
],
"boost": 1
}
}
],
"adjust_pure_negative": true,
"boost": 1
}
},
"9a1b4bad-ccf5-4c67-8718-02696cb351e4_8da52e51-eabf-4f6c-b9f0-e222933c1cb7": {
"bool": {
"filter": [
{
"terms": {
"clientId.keyword": [
"9a1b4bad-ccf5-4c67-8718-02696cb351e4"
],
"boost": 1
}
},
{
"terms": {
"payerId.keyword": [
"8da52e51-eabf-4f6c-b9f0-e222933c1cb7"
],
"boost": 1
}
}
],
"adjust_pure_negative": true,
"boost": 1
}
},
"9a1b4bad-ccf5-4c67-8718-02696cb351e4_8da52e51-eabf-4f6c-b9f0-e222933c1cb7_602c7d66-e990-4dfb-b6e2-72b62ff159d5": {
"bool": {
"filter": [
{
"terms": {
"clientId.keyword": [
"9a1b4bad-ccf5-4c67-8718-02696cb351e4"
],
"boost": 1
}
},
{
"terms": {
"payerId.keyword": [
"8da52e51-eabf-4f6c-b9f0-e222933c1cb7"
],
"boost": 1
}
},
{
"terms": {
"beneficiaryId.keyword": [
"602c7d66-e990-4dfb-b6e2-72b62ff159d5"
],
"boost": 1
}
}
],
"adjust_pure_negative": true,
"boost": 1
}
}
},
"other_bucket": false,
"other_bucket_key": "_other_"
},
"aggregations": {
"beneficiary_amount": {
"stats": {
"field": "beneficiaryAmountUsd"
}
},
"payer_amount": {
"stats": {
"field": "payerAmountUsd"
}
},
"distinct_count_beneficiary": {
"cardinality": {
"field": "beneficiaryId.keyword"
}
},
"distinct_count_payer": {
"cardinality": {
"field": "payerId.keyword"
}
},
"distinct_count_client": {
"cardinality": {
"field": "clientId.keyword"
}
},
"distinct_count_bank_acc": {
"cardinality": {
"field": "bankAccountUserId.keyword"
}
},
"distinct_count_bene_country": {
"cardinality": {
"field": "beneficiaryCountry.keyword"
}
},
"distinct_count_payer_country": {
"cardinality": {
"field": "payerCountry.keyword"
}
},
"distinct_count_bene_currency": {
"cardinality": {
"field": "beneficiaryCurrency.keyword"
}
},
"distinct_count_payer_currency": {
"cardinality": {
"field": "payerCurrency.keyword"
}
},
"structured_payment_amount_personal": {
"range": {
"field": "payerAmountUsd",
"ranges": [
{
"from": 9000,
"to": 9999.999
}
],
"keyed": false
}
},
"structured_payment_amount_company": {
"range": {
"field": "payerAmountUsd",
"ranges": [
{
"from": 112500000,
"to": 124999999.999
}
],
"keyed": false
}
}
}
}
}
}
}
}
Test result : (Single Index)
| Scenario | Number of run | Execution times | Min | Max | Average |
|---|---|---|---|---|---|
| Single thread Search result hit Result size unset | 10 | 6s, 5.9s, 5.9s, 6s, 5.9s, 6.2s, 6.2s, 6.2s, 6.6s, 8.2s | 5.9s | 8.2s | 6.93s |
| Single thread Search result hit Result size = 0 | 10 | 7200ms, 12ms, 19ms, 20ms, 23ms, 22ms, 28ms, 21ms, 19ms, 29ms | 12ms | 7200ms | 19.3ms (First execution takes 7.2s, deviates too much from the rest, so exclude it) |
| Single thread Search result none hit Result size unset | 10 | 510ms, 529ms, 549ms, 512ms, 489ms, 520ms, 506ms, 500ms, 493ms, 499ms | 489ms | 549ms | 510.7ms |
| Single thread Search result none hit Result size = 0 | 10 | 389ms, 1ms, 2ms, 3ms, 3ms, 1ms, 1ms, 1ms, 2ms, 3ms | 389ms | 1ms | 1.89ms (First execution takes 389ms, deviates too much from the rest, so exclude it) |
| 20 threads Search result hit Result size unset | 10 | Exception, listener timeout after waiting for [60000] ms | |||
| 20 threads Search result hit Result size = 0 | 10 | 18ms, 29ms, 61ms, 53ms, 93ms, 20ms, 39ms, 82ms, 17ms, 24ms | 17ms | 82ms | 43.6ms |
| 20 threads Search result none hit Result size unset | 10 | 4.4s, 5.8s, 4.9s, 5.6s 4.6s, 5.4s, 5.3s 4.7s, 5.8s, 4.7s | 5.8s | 4.4s | 5.1s |
| 20 threads Search result none hit Result size = 0 | 10 | 1.4s, 25ms, 6ms, 22ms, 276ms, 250ms, 19ms, 30ms, 11ms, 18ms | 1.4s | 11ms | 73ms (First execution takes 1.4s, deviates too much from the rest, so exclude it) |
Test result: (Multiple Index)
| Scenario | Number of run | Execution times | Min | Max | Average |
|---|---|---|---|---|---|
| Single thread Search result hit Result size unset | 10 | 12.6s, 12.7s 12.6s 12.7s, 13.1s, 13.1s, 13.2s, 13,1s, 13.1s, 13.1s | 12.6s | 13.2s | 12.9s |
| Single thread Search result hit Result size = 0 | 10 | 212ms, 147ms, 170ms, 272ms, 223ms, 222ms, 207ms, 255ms, 219ms, 219ms | 147ms | 272ms | 214.6ms |
| Single thread Search result none hit Result size unset | 10 | 1.2s, 1.1s, 1.1s, 1.1s, 1.1s, 1.2s, 1.1s, 1.1s, 1.2s, 1.1s | 1.1s | 1.2s | 1.13s |
| Single thread Search result none hit Result size = 0 | 10 | 909ms, 23ms, 23ms, 35ms, 14ms, 26ms, 19ms, 12ms, 15ms, 22m | 12ms | 909ms | 18.9ms (First execution takes 909ms, deviates too much from the rest, so exclude it) |
| 20 threads Search result hit Result size unset | 10 | Exception, listener timeout after waiting for [60000] ms | |||
| 20 threads Search result hit Result size = 0 | 10 | 18ms, 29ms, 61ms, 53ms, 93ms, 20ms, 39ms, 82ms, 17ms, 24ms | 17ms | 93ms | 43.6ms |
| 20 threads Search result none hit Result size unset | 10 | 7.8s, 7.6s, 7.1s, 6.9s, 7.4s, 8s, 7.3, 7.2s, 7.3s, 7.4s | 6.9s | 8s | 7.4s |
| 20 threads Search result none hit Result size = 0 | 10 | 202ms, 91ms, 100ms, 146ms, 182ms, 131ms, 158ms, 48ms, 171ms, 152ms | 48ms | 202ms | 138.1ms |
Conclusion:
Aggregation performance hinges on the number of documents that matches the aggregation
Result size parameter has significant impact on aggregation performance. not only because it skipped returning hit documents, but also because it enables caching for aggregation result, otherwise, you have to force result caching by explicitly setting request_cache=true
https://www.elastic.co/guide/en/elasticsearch/reference/6.6/shard-request-cache.html
Executing query concurrently can also have negative impact on performance
Increasing number of indices have positive impact on index speed but have large negative impact on aggregation if the aggregation is performed across indices
