Java embedded data engine from SQLite to SPL instance analysis

The data engines that can be embedded in Java applications seem to be rich, but in fact it is not easy to choose. Redis has poor computing power and is only suitable for simple query scenarios. The Spark architecture is complex and heavy, making deployment and maintenance very troublesome. Embedded databases such as H2\HSQLDB\Derby have simple structures, but their computing capabilities are insufficient and they do not even support basic window functions.

In contrast, SQLite has achieved a better balance in architecture and computing power, and is a widely used Java embedded data engine.

SQLite adapts to conventional basic application scenarios

SQLite has a simple structure. Although its core is developed in C language, it is well encapsulated and presented to the outside as a small Jar package, which can be easily integrated. in Java applications. SQLite provides a JDBC interface that can be called by Java:

Connection connection = DriverManager.getConnection("jdbc:sqlite::memory:");
Statement st = connection.createStatement();
st.execute("restore from d:/ex1");
ResultSet rs = st.executeQuery("SELECT * FROM orders");

SQLite provides standard SQL syntax, and there is no problem with conventional data processing and calculations. In particular, SQLite already supports window functions, which can easily implement many intra-group operations and has stronger computing power than other embedded databases.

SELECT x, y, row_number() OVER (ORDER BY y) AS row_number FROM t0 ORDER BY x;
SELECT a, b, group_concat(b, '.') OVER ( ORDER BY a ROWS BETWEEN 1 PRECEDING AND 1 FOLLOWING) AS group_concat FROM t1;

SQLite still has shortcomings when facing complex scenarios

SQLite has outstanding advantages, but it still has some shortcomings when it comes to complex application scenarios.

Java applications may process a variety of data sources, such as csv files, RDB, Excel, and Restful, but SQLite only handles simple cases, that is, it provides a directly available command line loader for text files such as csv. ：

.import --csv --skip 1 --schema temp /Users/scudata/somedata.csv tab1

For most other data sources, SQLite does not provide convenient interfaces. You can only hard-write code to load data, which requires calling the command line multiple times. The whole process is very cumbersome and timely.

Take loading RDB data source as an example. The general approach is to first use Java to execute the command line and convert the RDB library table to csv; then use JDBC to access SQLite and create the table structure; then use Java to execute the command line. Import the csv file into SQLite; finally index the new table to improve performance. This method is relatively rigid. If you want to flexibly define the table structure and table name, or determine the loaded data through calculation, the code will be more difficult to write.

Similarly, for other data sources, SQLite cannot be loaded directly, and it also needs to go through a tedious conversion process.

SQL is close to natural language, has a low learning threshold, and is easy to implement simple calculations, but it is not good at complex calculations, such as complex set calculations, ordered calculations, associated calculations, and multi-step calculations. SQLite uses SQL statements for calculations, and the advantages and disadvantages of SQL will be inherited. If you barely implement these complex calculations, the code will appear cumbersome and difficult to understand.

For example, the longest number of rising days for a certain stock, the SQL should be written like this:

select max(continuousDays)-1
from (select count(*) continuousDays
from (select sum(changeSign) over(order by tradeDate) unRiseDays
from (select tradeDate,
case when price>lag(price) over(order by tradeDate) then 0 else 1 end changeSign from AAPL) )
group by unRiseDays)

This is not just a problem with SQLite. In fact, due to incomplete aggregation, lack of serial numbers, and lack of Due to object references and other reasons, other SQL databases are not good at these operations.

Business logic consists of structured data calculation and process control. SQLite supports SQL and has structured data calculation capabilities. However, SQLite does not provide stored procedures and does not have independent process control capabilities, so it cannot implement general Business logic usually uses the judgment and loop statements of the Java main program. Since Java does not have professional structured data objects to carry SQLite data tables and records, the conversion process is cumbersome, the processing process is not smooth, and the development efficiency is not high.

As mentioned earlier, the SQLite core is a C program. Although it can be integrated into Java applications, it cannot be seamlessly integrated with Java. Exchanging data with the Java main program requires time-consuming conversion. Performance will be significantly insufficient when large amounts of data are involved or interactions are frequent. Also because the kernel is a C program, SQLite will destroy the consistency and robustness of the Java architecture to a certain extent.

For Java applications, esProc SPL natively on the JVM is a better choice.

SPL fully supports various data sources

esProc SPL is an open source embedded data engine under the JVM. It has a simple architecture and can directly load data sources. It can be integrated and called by Java through the JDBC interface, and is convenient for subsequent calculations.

SPL has a simple architecture and does not require independent services. As long as the SPL Jar package is introduced, it can be deployed in the Java environment.

Load the data source directly, the code is short, the process is simple, and the timeliness is strong. For example, load Oracle:

对于SQLite擅长加载的csv文件，SPL也可以直接加载，使用内置函数而不是外部命令行，稳定且效率高，代码更简短：

=T("/Users/scudata/somedata.csv")

多种外部数据源。除了RDB和csv，SPL还直接支持txt\xls等文件，MongoDB、Hadoop、redis、ElasticSearch、Kafka、Cassandra等NoSQL，以及WebService XML、Restful Json等多层数据。比如，将HDSF里的文件加载到内存：

JDBC接口可以方便地集成。加载的数据量一般比较大，通常在应用的初始阶段运行一次，只须将上面的加载过程存为SPL脚本文件，在Java中以存储过程的形式引用脚本文件名：

Class.forName("com.esproc.jdbc.InternalDriver");
Connection conn =DriverManager.getConnection("jdbc:esproc:local://");
CallableStatement statement = conn.prepareCall("{call init()}");
statement.execute();

SPL的计算能力更强大

SPL提供了丰富的计算函数，可以轻松实现日常计算。SPL支持多种高级语法，大量的日期函数和字符串函数，很多用SQL难以表达的计算，用SPL都可以轻松实现，包括复杂的有序计算、集合计算、分步计算、关联计算，以及带流程控制的业务逻辑。

丰富的计算函数。SPL可以轻松实现各类日常计算：

标准SQL语法。SPL也提供了SQL-92标准的语法，比如分组汇总：

$select year(OrderDate) y,month(OrderDate) m, sum(Amount) s,count(1) c
from {Orders}
Where Amount>=? and Amount<? ;arg1,arg2

函数选项、层次参数等方便的语法。功能相似的函数可以共用一个函数名，只用函数选项区分差别，比SQL更加灵活方便。比如select函数的基本功能是过滤，如果只过滤出符合条件的第1条记录，可使用选项@1：

T.select@1(Amount>1000)

二分法排序，即对有序数据用二分法进行快速过滤，使用@b：

T.select@b(Amount>1000)

有序分组，即对分组字段有序的数据，将相邻且字段值相同的记录分为一组，使用@b：

T.groups@b(Client;sum(Amount))

函数选项还可以组合搭配，比如：

Orders.select@1b(Amount>1000)

结构化运算函数的参数有些很复杂，比如SQL就需要用各种关键字把一条语句的参数分隔成多个组，但这会动用很多关键字，也使语句结构不统一。SPL使用层次参数简化了复杂参数的表达，即通过分号、逗号、冒号自高而低将参数分为三层：

join(Orders:o,SellerId ; Employees:e,EId)

更丰富的日期和字符串函数。除了常见函数，比如日期增减、截取字符串，SPL还提供了更丰富的日期和字符串函数，在数量和功能上远远超过了SQL，同样运算时代码更短。比如：

季度增减：elapse@q(“2020-02-27”,-3) //返回2019-05-27

N个工作日之后的日期：workday(date(“2022-01-01”),25) //返回2022-02-04

字符串类函数，判断是否全为数字：isdigit(“12345”) //返回true

取子串前面的字符串：substr@l(“abCDcdef”,“cd”) //返回abCD

按竖线拆成字符串数组：“aa|bb|cc”.split(“|”) //返回[“aa”,“bb”,“cc”]

SPL还支持年份增减、求季度、按正则表达式拆分字符串、拆出SQL的where或select部分、拆出单词、按标记拆HTML等大量函数。

简化有序运算。涉及跨行的有序运算，通常都有一定的难度，比如比上期和同期比。SPL使用"字段[相对位置]"引用跨行的数据，可显著简化代码，还可以自动处理数组越界等特殊情况，比SQL窗口函数更加方便。比如，追加一个计算列rate，计算每条订单的金额增长率：

=T.derive(AMOUNT/AMOUNT[-1]-1: rate)

综合运用位置表达式和有序函数，很多SQL难以实现的有序运算，都可以用SPL轻松解决。比如，根据考勤表，找出连续 4 周每天均出勤达 7 小时的学生：

简化集合运算，SPL的集合化更加彻底，配合灵活的语法和强大的集合函数，可大幅简化复杂的集合计算。比如，在各部门找出比本部门平均年龄小的员工：

计算某支股票最长的连续上涨天数：

简化关联计算。SPL支持对象引用的形式表达关联，可以通过点号直观地访问关联表，避免使用JOIN导致的混乱繁琐，尤其适合复杂的多层关联和自关联。比如，根据员工表计算女经理的男员工：

=employees.select(gender:"male",dept.manager.gender:"female")

方便的分步计算，SPL集合化更加彻底，可以用变量方便地表达集合，适合多步骤计算，SQL要用嵌套表达的运算，用SPL可以更轻松实现。比如，找出销售额累计占到一半的前n个大客户，并按销售额从大到小排序：

流程控制语法。SPL提供了流程控制语句，配合内置的结构化数据对象，可以方便地实现各类业务逻辑。

分支判断语句：

循环语句：

与Java的循环类似，SPL还可用break关键字跳出（中断）当前循环体，或用next关键字跳过（忽略）本轮循环，不展开说了。

计算性能更好。在内存计算方面，除了常规的主键和索引外，SPL还提供了很多高性能的数据结构和算法支持，比大多数使用SQL的内存数据库性能好得多，且占用内存更少，比如预关联技术、并行计算、指针式复用。

优化体系结构

SPL支持JDBC接口，代码可外置于Java，耦合性更低，也可内置于Java，调用更简单。SPL支持解释执行和热切换，代码方便移植和管理运营，支持内外存混合计算。

外置代码耦合性低。SPL代码可外置于Java，通过文件名被调用，既不依赖数据库，也不依赖Java，业务逻辑和前端代码天然解耦。

对于较短的计算，也可以像SQLite那样合并成一句，写在Java代码中：

Class.forName("com.esproc.jdbc.InternalDriver");
Connection conn =DriverManager.getConnection("jdbc:esproc:local://");
Statement statement = conn.createStatement();
String arg1="1000";
String arg2="2000"
ResultSet result = statement.executeQuery(=Orders.select(Amount>="+arg1+" && Amount<"+arg2+"). groups(year(OrderDate):y,month(OrderDate):m; sum(Amount):s,count(1):c)");

解释执行和热切换。业务逻辑数量多，复杂度高，变化是常态。良好的系统构架，应该有能力应对变化的业务逻辑。SPL是基于Java的解释型语言，无须编译就能执行，脚本修改后立即生效，支持不停机的热切换，适合应对变化的业务逻辑。

方便代码移植。SPL通过数据源名从数据库取数，如果需要移植，只要改动配置文件中的数据源配置信息，而不必修改SPL代码。SPL支持动态数据源，可通过参数或宏切换不同的数据库，从而进行更方便的移植。为了进一步增强可移植性，SPL还提供了与具体数据库无关的标准SQL语法，使用sqltranslate函数可将标准SQL转为主流方言SQL，仍然通过query函数执行。

方便管理运营。由于支持库外计算，代码可被第三方工具管理，方便团队协作；SPL脚本可以按文件目录进行存放，方便灵活，管理成本低；SPL对数据库的权限要求类似Java，不影响数据安全。

内外存混合计算。有些数据太大，无法放入内存，但又要与内存表共同计算，这种情况可利用SPL实现内外存混合计算。比如，主表orders已加载到内存，大明细表orderdetail是文本文件，下面进行主表和明细表的关联计算：

SQLite使用简单方便，但数据源加载繁琐，计算能力不足。SPL架构也非常简单，并直接支持更多数据源。SPL计算能力强大，提供了丰富的计算函数，可以轻松实现SQL不擅长的复杂计算。SPL还提供多种优化体系结构的手段，代码既可外置也可内置于Java，支持解释执行和热切换，方便移植和管理运营，并支持内外存混合计算。

The above is the detailed content of Java embedded data engine from SQLite to SPL instance analysis. For more information, please follow other related articles on the PHP Chinese website!