从Cassandra加载单个(或几个)宽行到C#的性能最有效的方法是什么?我的宽行有10.000-100.000列。主键由多个值组成,但列键是单个字符串,列值是单个计数器(请参阅下面的架构)。
使用"跟踪"在cqlsh中我可以看到Cassandra可以在44米内选择一个包含17.000列的宽行,但是使用Datastax驱动程序将这些数据一直加载到C#需要700毫秒。有更快的方法吗?我需要在50-100ms内加载全行。 (有没有更原生的方式?最小化网络流量的方法?更快的驱动程序?驱动程序的另一种配置?或其他什么?)
我实际上并不需要所有17.000列。我只需要'support'> = 2的列,或者需要按'support'排序的前1000列。但由于“支持”是我的列值,我不知道在CQL中如何查询这样的内容。
这是我的表:
CREATE TABLE real_time.grouped_feature_support (
algorithm_id int,
group_by_feature_id int,
select_feature_id int,
group_by_feature_value text,
select_feature_value text,
support counter,
PRIMARY KEY ((algorithm_id, group_by_feature_id, select_feature_id, group_by_feature_value), select_feature_value)
这是我使用Datastax驱动程序访问数据的方法:
var table = session.GetTable<GroupedFeatureSupportDataEntry>();
var query = table.Where(x => x.CustomerAlgorithmId == customerAlgorithmId
&& x.GroupByFeatureId == groupedFeatureId
&& myGroupedFeatureValues.Contains(x.GroupByFeatureValue)
&& x.GroupByFeatureValue == groupedFeatureValue
&& x.SelectFeatureId == selectFeatureId)
.Select(x => new
{
x.GroupByFeatureValue,
x.SelectFeatureValue,
x.Support,
})
.Take(1000000);
var result = query.Execute();
答案 0 :(得分:0)
如果您在检索大型结果集时寻求最佳性能,则不应使用Linq-to-cql或其他任何映射组件。
您可以使用technique documented on the driver readme检索行,在您的情况下,它将类似于:
var query = "SELECT * from grouped_feature_support WHERE" +
" algorithm_id = ? AND group_by_feature_id = ? " +
" AND select_feature_id = ? AND group_by_feature_value = ?";
//Prepare the query once in your application lifetime
var ps = session.Prepare(query);
//Reuse the prepared statement by binding different parameters to it
var rs = session.Execute(ps.Bind(parameters));
foreach (var row in rs)
{
//The enumerator will yield all the rows from Cassandra
//Retrieving them in the back in blocks of 5000 (determined by the pagesize).
}
//You can also use a IEnumerable<T> Linq Extensions to filter
var filteredRows = rs.Where(r => r.GetValue<long>("support") > 2);