我有一个看起来像下面的 ETL,除了我们从 mysql 中提取 8 个字段并且有 8 个顶点类和 7 个边缘类。
{
"config": {
"parallel": true
},
"extractor": {
"jdbc": {
"driver": "com.mysql.jdbc.Driver",
"url": "jdbc:mysql://<some IP>/car_database",
"userName": "admin",
"userPassword": "admin",
"query": "select uid, type, color, age from car_info;",
"fetchSize": -2147483648
}
},
"transformers": [
{
"vertex": {
"class": "Car"
}
},
{
"field": {
"fieldName": "value",
"expression": "uid"
}
},
{
"edge": {
"class": "HasType",
"lookup": "Type.value",
"joinFieldName": "type",
"unresolvedLinkAction": "CREATE"
}
},
{
"edge": {
"class": "HasColor",
"lookup": "Color.value",
"joinFieldName": "color",
"unresolvedLinkAction": "CREATE"
}
},
{
"edge": {
"class": "HasAge",
"lookup": "Age.value",
"joinFieldName": "age",
"unresolvedLinkAction": "CREATE"
}
},
{
"field": {
"fieldNames" : [
"uid",
"type",
"color",
"age"
],
"operation": "remove"
}
}
],
"loader": {
"orientdb": {
"dbURL": "remote:<some IP >/CarGraph",
"dbType": "graph",
"txUseLog": false,
"wal": false,
"tx": false,
"batchCommit": 1000,
"classes": [
{
"name": "Car",
"extends": "V"
},
{
"name": "Type",
"extends": "V"
},
{
"name": "HasType",
"extends": "E"
},
{
"name": "Color",
"extends": "V"
},
{
"name": "HasColor",
"extends": "E"
},
{
"name": "Age",
"extends": "V"
},
{
"name": "HasAge",
"extends": "E"
}
],
"indexes": [
{
"class": "Car",
"fields": [
"value:string"
],
"type": "UNIQUE"
},
{
"class": "Type",
"fields": [
"value:string"
],
"type": "UNIQUE"
},
{
"class": "Color",
"fields": [
"value:string"
],
"type": "UNIQUE"
},
{
"class": "Age",
"fields": [
"value:string"
],
"type": "UNIQUE"
}
]
}
}
}
这是我的默认分布式数据库配置文件:
{
"autoDeploy": true,
"hotAlignment": false,
"executionMode": "undefined",
"readQuorum": 1,
"writeQuorum": 1,
"failureAvailableNodesLessQuorum": false,
"readYourWrites": true,
"servers": {
"*": "master"
},
"clusters": {
"internal": {
},
"index": {
},
"*": {
"servers": ["<NEW_NODE>"]
}
}
}
我正在尝试实现一个 orient 集群,目前在不同的服务器上有 2 个节点。当我在其中一个图形框上运行导入时(我们当前的 QA 环境中只有 200 行,prod 将超过 3500 万),我看到每秒 1-3 个节点的加载时间,即使仲裁设置为 1。我已经尝试了 5-6 种不同的组合,切换并行、仲裁值、不同的服务器设置,但只要加载器指向远程数据库,它就真的很慢。
相同的 etl 在我的本地机器上以 1500-2000/s 的速度运行到 plocal 图中。
我发现使用指向 plocal 图的加载器运行导入的最快方法。然后,一旦它完成重新启动分布式服务器并允许 2 个节点同步。
是否有我遗漏的东西,或者某些可能导致导入速度如此缓慢的服务器配置?东方版本 2.1.6。