mirror of
https://github.com/csunny/DB-GPT.git
synced 2025-08-10 12:42:34 +00:00
connect database
This commit is contained in:
csunny
parent
8eb37122c4
commit
a71c8b6d56
200
pilot/app.py
200
pilot/app.py
@ -33,198 +33,22 @@ def knowledged_qa_demo(text_list):
|
|||||||
|
|
||||||
|
|
||||||
def get_answer(q):
|
def get_answer(q):
|
||||||
base_knowledge = """
|
base_knowledge = """ """
|
||||||
执行计划是对一条 SQL 查询语句在数据库中执行过程的描述。用户可以通过 EXPLAIN 命令查看优化器针对指定 SQL 生成的逻辑执行计划。
|
|
||||||
|
|
||||||
如果要分析某条 SQL 的性能问题，通常需要先查看 SQL 的执行计划，排查每一步 SQL 执行是否存在问题。所以读懂执行计划是 SQL 优化的先决条件，而了解执行计划的算子是理解 EXPLAIN 命令的关键。
|
|
||||||
|
|
||||||
OceanBase 数据库的执行计划命令有三种模式：EXPLAIN BASIC、EXPLAIN 和 EXPLAIN EXTENDED。这三种模式对执行计划展现不同粒度的细节信息:
|
|
||||||
|
|
||||||
EXPLAIN BASIC 命令用于最基本的计划展示。
|
|
||||||
|
|
||||||
EXPLAIN EXTENDED 命令用于最详细的计划展示（通常在排查问题时使用这种展示模式）。
|
|
||||||
|
|
||||||
EXPLAIN 命令所展示的信息可以帮助普通用户了解整个计划的执行方式。
|
|
||||||
|
|
||||||
EXPLAIN 命令格式如下：
|
|
||||||
EXPLAIN [BASIC | EXTENDED | PARTITIONS | FORMAT = format_name] [PRETTY | PRETTY_COLOR] explainable_stmt
|
|
||||||
format_name:
|
|
||||||
{ TRADITIONAL | JSON }
|
|
||||||
explainable_stmt:
|
|
||||||
{ SELECT statement
|
|
||||||
| DELETE statement
|
|
||||||
| INSERT statement
|
|
||||||
| REPLACE statement
|
|
||||||
| UPDATE statement }
|
|
||||||
|
|
||||||
|
|
||||||
EXPLAIN 命令适用于 SELECT、DELETE、INSERT、REPLACE 和 UPDATE 语句，显示优化器所提供的有关语句执行计划的信息，包括如何处理该语句，如何联接表以及以何种顺序联接表等信息。
|
|
||||||
|
|
||||||
一般来说，可以使用 EXPLAIN EXTENDED 命令，将表扫描的范围段展示出来。使用 EXPLAIN OUTLINE 命令可以显示 Outline 信息。
|
|
||||||
|
|
||||||
FORMAT 选项可用于选择输出格式。TRADITIONAL 表示以表格格式显示输出，这也是默认设置。JSON 表示以 JSON 格式显示信息。
|
|
||||||
|
|
||||||
使用 EXPLAIN PARTITITIONS 也可用于检查涉及分区表的查询。如果检查针对非分区表的查询，则不会产生错误，但 PARTIONS 列的值始终为 NULL。
|
|
||||||
|
|
||||||
对于复杂的执行计划，可以使用 PRETTY 或者 PRETTY_COLOR 选项将计划树中的父节点和子节点使用树线或彩色树线连接起来，使得执行计划展示更方便阅读。示例如下：
|
|
||||||
obclient> CREATE TABLE p1table(c1 INT ,c2 INT) PARTITION BY HASH(c1) PARTITIONS 2;
|
|
||||||
Query OK, 0 rows affected
|
|
||||||
|
|
||||||
obclient> CREATE TABLE p2table(c1 INT ,c2 INT) PARTITION BY HASH(c1) PARTITIONS 4;
|
|
||||||
Query OK, 0 rows affected
|
|
||||||
|
|
||||||
obclient> EXPLAIN EXTENDED PRETTY_COLOR SELECT * FROM p1table p1 JOIN p2table p2 ON p1.c1=p2.c2\G
|
|
||||||
*************************** 1. row ***************************
|
|
||||||
Query Plan: ==========================================================
|
|
||||||
|ID|OPERATOR |NAME |EST. ROWS|COST|
|
|
||||||
----------------------------------------------------------
|
|
||||||
|0 |PX COORDINATOR | |1 |278 |
|
|
||||||
|1 | EXCHANGE OUT DISTR |:EX10001|1 |277 |
|
|
||||||
|2 | HASH JOIN | |1 |276 |
|
|
||||||
|3 | ├PX PARTITION ITERATOR | |1 |92 |
|
|
||||||
|4 | │ TABLE SCAN |P1 |1 |92 |
|
|
||||||
|5 | └EXCHANGE IN DISTR | |1 |184 |
|
|
||||||
|6 | EXCHANGE OUT DISTR (PKEY)|:EX10000|1 |184 |
|
|
||||||
|7 | PX PARTITION ITERATOR | |1 |183 |
|
|
||||||
|8 | TABLE SCAN |P2 |1 |183 |
|
|
||||||
==========================================================
|
|
||||||
|
|
||||||
Outputs & filters:
|
|
||||||
-------------------------------------
|
|
||||||
0 - output([INTERNAL_FUNCTION(P1.C1, P1.C2, P2.C1, P2.C2)]), filter(nil)
|
|
||||||
1 - output([INTERNAL_FUNCTION(P1.C1, P1.C2, P2.C1, P2.C2)]), filter(nil), dop=1
|
|
||||||
2 - output([P1.C1], [P2.C2], [P1.C2], [P2.C1]), filter(nil),
|
|
||||||
equal_conds([P1.C1 = P2.C2]), other_conds(nil)
|
|
||||||
3 - output([P1.C1], [P1.C2]), filter(nil)
|
|
||||||
4 - output([P1.C1], [P1.C2]), filter(nil),
|
|
||||||
access([P1.C1], [P1.C2]), partitions(p[0-1])
|
|
||||||
5 - output([P2.C2], [P2.C1]), filter(nil)
|
|
||||||
6 - (#keys=1, [P2.C2]), output([P2.C2], [P2.C1]), filter(nil), dop=1
|
|
||||||
7 - output([P2.C1], [P2.C2]), filter(nil)
|
|
||||||
8 - output([P2.C1], [P2.C2]), filter(nil),
|
|
||||||
access([P2.C1], [P2.C2]), partitions(p[0-3])
|
|
||||||
|
|
||||||
1 row in set
|
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
## 执行计划形状与算子信息
|
|
||||||
|
|
||||||
在数据库系统中，执行计划在内部通常是以树的形式来表示的，但是不同的数据库会选择不同的方式展示给用户。
|
|
||||||
|
|
||||||
如下示例分别为 PostgreSQL 数据库、Oracle 数据库和 OceanBase 数据库对于 TPCDS Q3 的计划展示。
|
|
||||||
|
|
||||||
```sql
|
|
||||||
obclient> SELECT /*TPC-DS Q3*/ *
|
|
||||||
FROM (SELECT dt.d_year,
|
|
||||||
item.i_brand_id brand_id,
|
|
||||||
item.i_brand brand,
|
|
||||||
Sum(ss_net_profit) sum_agg
|
|
||||||
FROM date_dim dt,
|
|
||||||
store_sales,
|
|
||||||
item
|
|
||||||
WHERE dt.d_date_sk = store_sales.ss_sold_date_sk
|
|
||||||
AND store_sales.ss_item_sk = item.i_item_sk
|
|
||||||
AND item.i_manufact_id = 914
|
|
||||||
AND dt.d_moy = 11
|
|
||||||
GROUP BY dt.d_year,
|
|
||||||
item.i_brand,
|
|
||||||
item.i_brand_id
|
|
||||||
ORDER BY dt.d_year,
|
|
||||||
sum_agg DESC,
|
|
||||||
brand_id)
|
|
||||||
WHERE ROWNUM <= 100;
|
|
||||||
|
|
||||||
PostgreSQL 数据库执行计划展示如下：
|
|
||||||
Limit (cost=13986.86..13987.20 rows=27 width=91)
|
|
||||||
Sort (cost=13986.86..13986.93 rows=27 width=65)
|
|
||||||
Sort Key: dt.d_year, (sum(store_sales.ss_net_profit)), item.i_brand_id
|
|
||||||
HashAggregate (cost=13985.95..13986.22 rows=27 width=65)
|
|
||||||
Merge Join (cost=13884.21..13983.91 rows=204 width=65)
|
|
||||||
Merge Cond: (dt.d_date_sk = store_sales.ss_sold_date_sk)
|
|
||||||
Index Scan using date_dim_pkey on date_dim dt (cost=0.00..3494.62 rows=6080 width=8)
|
|
||||||
Filter: (d_moy = 11)
|
|
||||||
Sort (cost=12170.87..12177.27 rows=2560 width=65)
|
|
||||||
Sort Key: store_sales.ss_sold_date_sk
|
|
||||||
Nested Loop (cost=6.02..12025.94 rows=2560 width=65)
|
|
||||||
Seq Scan on item (cost=0.00..1455.00 rows=16 width=59)
|
|
||||||
Filter: (i_manufact_id = 914)
|
|
||||||
Bitmap Heap Scan on store_sales (cost=6.02..658.94 rows=174 width=14)
|
|
||||||
Recheck Cond: (ss_item_sk = item.i_item_sk)
|
|
||||||
Bitmap Index Scan on store_sales_pkey (cost=0.00..5.97 rows=174 width=0)
|
|
||||||
Index Cond: (ss_item_sk = item.i_item_sk)
|
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
Oracle 数据库执行计划展示如下：
|
|
||||||
Plan hash value: 2331821367
|
|
||||||
--------------------------------------------------------------------------------------------------
|
|
||||||
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
|
|
||||||
--------------------------------------------------------------------------------------------------
|
|
||||||
| 0 | SELECT STATEMENT | | 100 | 9100 | 3688 (1)| 00:00:01 |
|
|
||||||
|* 1 | COUNT STOPKEY | | | | | |
|
|
||||||
| 2 | VIEW | | 2736 | 243K| 3688 (1)| 00:00:01 |
|
|
||||||
|* 3 | SORT ORDER BY STOPKEY | | 2736 | 256K| 3688 (1)| 00:00:01 |
|
|
||||||
| 4 | HASH GROUP BY | | 2736 | 256K| 3688 (1)| 00:00:01 |
|
|
||||||
|* 5 | HASH JOIN | | 2736 | 256K| 3686 (1)| 00:00:01 |
|
|
||||||
|* 6 | TABLE ACCESS FULL | DATE_DIM | 6087 | 79131 | 376 (1)| 00:00:01 |
|
|
||||||
| 7 | NESTED LOOPS | | 2865 | 232K| 3310 (1)| 00:00:01 |
|
|
||||||
| 8 | NESTED LOOPS | | 2865 | 232K| 3310 (1)| 00:00:01 |
|
|
||||||
|* 9 | TABLE ACCESS FULL | ITEM | 18 | 1188 | 375 (0)| 00:00:01 |
|
|
||||||
|* 10 | INDEX RANGE SCAN | SYS_C0010069 | 159 | | 2 (0)| 00:00:01 |
|
|
||||||
| 11 | TABLE ACCESS BY INDEX ROWID| STORE_SALES | 159 | 2703 | 163 (0)| 00:00:01 |
|
|
||||||
--------------------------------------------------------------------------------------------------
|
|
||||||
|
|
||||||
OceanBase 数据库执行计划展示如下：
|
|
||||||
|ID|OPERATOR |NAME |EST. ROWS|COST |
|
|
||||||
-------------------------------------------------------
|
|
||||||
|0 |LIMIT | |100 |81141|
|
|
||||||
|1 | TOP-N SORT | |100 |81127|
|
|
||||||
|2 | HASH GROUP BY | |2924 |68551|
|
|
||||||
|3 | HASH JOIN | |2924 |65004|
|
|
||||||
|4 | SUBPLAN SCAN |VIEW1 |2953 |19070|
|
|
||||||
|5 | HASH GROUP BY | |2953 |18662|
|
|
||||||
|6 | NESTED-LOOP JOIN| |2953 |15080|
|
|
||||||
|7 | TABLE SCAN |ITEM |19 |11841|
|
|
||||||
|8 | TABLE SCAN |STORE_SALES|161 |73 |
|
|
||||||
|9 | TABLE SCAN |DT |6088 |29401|
|
|
||||||
=======================================================
|
|
||||||
|
|
||||||
由示例可见，OceanBase 数据库的计划展示与 Oracle 数据库类似。
|
|
||||||
|
|
||||||
OceanBase 数据库执行计划中的各列的含义如下：
|
|
||||||
列名 含义
|
|
||||||
ID 执行树按照前序遍历的方式得到的编号（从 0 开始）。
|
|
||||||
OPERATOR 操作算子的名称。
|
|
||||||
NAME 对应表操作的表名（索引名）。
|
|
||||||
EST. ROWS 估算该操作算子的输出行数。
|
|
||||||
COST 该操作算子的执行代价（微秒）。
|
|
||||||
|
|
||||||
|
|
||||||
OceanBase 数据库 EXPLAIN 命令输出的第一部分是执行计划的树形结构展示。其中每一个操作在树中的层次通过其在 operator 中的缩进予以展示，层次最深的优先执行，层次相同的以特定算子的执行顺序为标准来执行。
|
|
||||||
|
|
||||||
问题: update a not exists (b…)
|
|
||||||
我一开始以为 B是驱动表，B的数据挺多的 后来看到NLAJ，是说左边的表关联右边的表
|
|
||||||
所以这个的驱动表是不是实际是A，用A的匹配B的，这个理解有问题吗
|
|
||||||
|
|
||||||
回答: 没错 A 驱动 B的
|
|
||||||
|
|
||||||
问题: 光知道最下最右的是驱动表了 所以一开始搞得有点懵 :sweat_smile:
|
|
||||||
|
|
||||||
回答: nlj应该原理应该都是左表(驱动表)的记录探测右表(被驱动表)， 选哪张成为左表或右表就基于一些其他考量了，比如数据量， 而anti join/semi join只是对 not exist/exist的一种优化，相关的原理和资料网上可以查阅一下
|
|
||||||
|
|
||||||
问题: 也就是nlj 就是按照之前理解的谁先执行 谁就是驱动表 也就是执行计划中的最右的表
|
|
||||||
而anti join/semi join，谁在not exist左面，谁就是驱动表。这么理解对吧
|
|
||||||
|
|
||||||
回答: nlj也是左表的表是驱动表，这个要了解下计划执行方面的基本原理，取左表的一行数据，再遍历右表，一旦满足连接条件，就可以返回数据
|
|
||||||
anti/semi只是因为not exists/exist的语义只是返回左表数据，改成anti join是一种计划优化，连接的方式比子查询更优
|
|
||||||
"""
|
|
||||||
text_list = [base_knowledge]
|
text_list = [base_knowledge]
|
||||||
index = knowledged_qa_demo(text_list)
|
index = knowledged_qa_demo(text_list)
|
||||||
response = index.query(q)
|
response = index.query(q)
|
||||||
return response.response
|
return response.response
|
||||||
|
|
||||||
|
def get_similar(q):
|
||||||
|
from pilot.vector_store.extract_tovec import knownledge_tovec
|
||||||
|
docsearch = knownledge_tovec("./datasets/plan.md")
|
||||||
|
docs = docsearch.similarity_search_with_score(q, k=1)
|
||||||
|
|
||||||
|
for doc in docs:
|
||||||
|
dc, s = doc
|
||||||
|
print(dc.page_content)
|
||||||
|
yield dc.page_content
|
||||||
|
|
||||||
if __name__ == "__main__":
|
if __name__ == "__main__":
|
||||||
# agent_demo()
|
# agent_demo()
|
||||||
|
|
||||||
@ -235,7 +59,7 @@ if __name__ == "__main__":
|
|||||||
text_output = gr.TextArea()
|
text_output = gr.TextArea()
|
||||||
text_button = gr.Button()
|
text_button = gr.Button()
|
||||||
|
|
||||||
text_button.click(get_answer, inputs=text_input, outputs=text_output)
|
text_button.click(get_similar, inputs=text_input, outputs=text_output)
|
||||||
|
|
||||||
demo.queue(concurrency_count=3).launch(server_name="0.0.0.0")
|
demo.queue(concurrency_count=3).launch(server_name="0.0.0.0")
|
||||||
|
|
||||||
|
|||||||
@ -1,2 +1,36 @@
|
|||||||
#!/usr/bin/env python3
|
#!/usr/bin/env python3
|
||||||
# -*- coding: utf-8 -*-
|
# -*- coding: utf-8 -*-
|
||||||
|
|
||||||
|
import pymysql
|
||||||
|
|
||||||
|
class MySQLOperator:
|
||||||
|
"""Connect MySQL Database fetch MetaData For LLM Prompt """
|
||||||
|
def __init__(self, user, password, host="localhost", port=3306) -> None:
|
||||||
|
|
||||||
|
self.conn = pymysql.connect(
|
||||||
|
host=host,
|
||||||
|
user=user,
|
||||||
|
passwd=password,
|
||||||
|
charset="utf8mb4",
|
||||||
|
cursorclass=pymysql.cursors.DictCursor
|
||||||
|
)
|
||||||
|
|
||||||
|
def get_schema(self, schema_name):
|
||||||
|
|
||||||
|
with self.conn.cursor() as cursor:
|
||||||
|
_sql = f"""
|
||||||
|
select concat(table_name, "(" , group_concat(column_name), ")") as schema_info from information_schema.COLUMNS where table_schema="{schema_name}" group by TABLE_NAME;
|
||||||
|
"""
|
||||||
|
cursor.execute(_sql)
|
||||||
|
results = cursor.fetchall()
|
||||||
|
return results
|
||||||
|
|
||||||
|
|
||||||
|
if __name__ == "__main__":
|
||||||
|
mo = MySQLOperator(
|
||||||
|
"root",
|
||||||
|
"aa123456",
|
||||||
|
)
|
||||||
|
|
||||||
|
schema = mo.get_schema("dbgpt")
|
||||||
|
print(schema)
|
||||||
185
pilot/datasets/plan.md
Normal file
185
pilot/datasets/plan.md
Normal file
@ -0,0 +1,185 @@
|
|||||||
|
执行计划是对一条 SQL 查询语句在数据库中执行过程的描述。用户可以通过 EXPLAIN 命令查看优化器针对指定 SQL 生成的逻辑执行计划。
|
||||||
|
|
||||||
|
如果要分析某条 SQL 的性能问题，通常需要先查看 SQL 的执行计划，排查每一步 SQL 执行是否存在问题。所以读懂执行计划是 SQL 优化的先决条件，而了解执行计划的算子是理解 EXPLAIN 命令的关键。
|
||||||
|
|
||||||
|
OceanBase 数据库的执行计划命令有三种模式：EXPLAIN BASIC、EXPLAIN 和 EXPLAIN EXTENDED。这三种模式对执行计划展现不同粒度的细节信息:
|
||||||
|
|
||||||
|
EXPLAIN BASIC 命令用于最基本的计划展示。
|
||||||
|
|
||||||
|
EXPLAIN EXTENDED 命令用于最详细的计划展示（通常在排查问题时使用这种展示模式）。
|
||||||
|
|
||||||
|
EXPLAIN 命令所展示的信息可以帮助普通用户了解整个计划的执行方式。
|
||||||
|
|
||||||
|
EXPLAIN 命令格式如下：
|
||||||
|
EXPLAIN [BASIC | EXTENDED | PARTITIONS | FORMAT = format_name] [PRETTY | PRETTY_COLOR] explainable_stmt
|
||||||
|
format_name:
|
||||||
|
{ TRADITIONAL | JSON }
|
||||||
|
explainable_stmt:
|
||||||
|
{ SELECT st
|
||||||
|
| DELETE statement
|
||||||
|
| INSERT statement
|
||||||
|
| REPLACE statement
|
||||||
|
| UPDATE statement }
|
||||||
|
|
||||||
|
|
||||||
|
EXPLAIN 命令适用于 SELECT、DELETE、INSERT、REPLACE 和 UPDATE 语句，显示优化器所提供的有关语句执行计划的信息，包括如何处理该语句，如何联接表以及以何种顺序联接表等信息。
|
||||||
|
|
||||||
|
一般来说，可以使用 EXPLAIN EXTENDED 命令，将表扫描的范围段展示出来。使用 EXPLAIN OUTLINE 命令可以显示 Outline 信息。
|
||||||
|
|
||||||
|
FORMAT 选项可用于选择输出格式。TRADITIONAL 表示以表格格式显示输出，这也是默认设置。JSON 表示以 JSON 格式显示信息。
|
||||||
|
|
||||||
|
使用 EXPLAIN PARTITITIONS 也可用于检查涉及分区表的查询。如果检查针对非分区表的查询，则不会产生错误，但 PARTIONS 列的值始终为 NULL。
|
||||||
|
|
||||||
|
对于复杂的执行计划，可以使用 PRETTY 或者 PRETTY_COLOR 选项将计划树中的父节点和子节点使用树线或彩色树线连接起来，使得执行计划展示更方便阅读。示例如下：
|
||||||
|
obclient> CREATE TABLE p1table(c1 INT ,c2 INT) PARTITION BY HASH(c1) PARTITIONS 2;
|
||||||
|
Query OK, 0 rows affected
|
||||||
|
|
||||||
|
obclient> CREATE TABLE p2table(c1 INT ,c2 INT) PARTITION BY HASH(c1) PARTITIONS 4;
|
||||||
|
Query OK, 0 rows affected
|
||||||
|
|
||||||
|
obclient> EXPLAIN EXTENDED PRETTY_COLOR SELECT * FROM p1table p1 JOIN p2table p2 ON p1.c1=p2.c2\G
|
||||||
|
*************************** 1. row ***************************
|
||||||
|
Query Plan: ==========================================================
|
||||||
|
|ID|OPERATOR |NAME |EST. ROWS|COST|
|
||||||
|
----------------------------------------------------------
|
||||||
|
|0 |PX COORDINATOR | |1 |278 |
|
||||||
|
|1 | EXCHANGE OUT DISTR |:EX10001|1 |277 |
|
||||||
|
|2 | HASH JOIN | |1 |276 |
|
||||||
|
|3 | ├PX PARTITION ITERATOR | |1 |92 |
|
||||||
|
|4 | │ TABLE SCAN |P1 |1 |92 |
|
||||||
|
|5 | └EXCHANGE IN DISTR | |1 |184 |
|
||||||
|
|6 | EXCHANGE OUT DISTR (PKEY)|:EX10000|1 |184 |
|
||||||
|
|7 | PX PARTITION ITERATOR | |1 |183 |
|
||||||
|
|8 | TABLE SCAN |P2 |1 |183 |
|
||||||
|
==========================================================
|
||||||
|
|
||||||
|
Outputs & filters:
|
||||||
|
-------------------------------------
|
||||||
|
0 - output([INTERNAL_FUNCTION(P1.C1, P1.C2, P2.C1, P2.C2)]), filter(nil)
|
||||||
|
1 - output([INTERNAL_FUNCTION(P1.C1, P1.C2, P2.C1, P2.C2)]), filter(nil), dop=1
|
||||||
|
2 - output([P1.C1], [P2.C2], [P1.C2], [P2.C1]), filter(nil),
|
||||||
|
equal_conds([P1.C1 = P2.C2]), other_conds(nil)
|
||||||
|
3 - output([P1.C1], [P1.C2]), filter(nil)
|
||||||
|
4 - output([P1.C1], [P1.C2]), filter(nil),
|
||||||
|
access([P1.C1], [P1.C2]), partitions(p[0-1])
|
||||||
|
5 - output([P2.C2], [P2.C1]), filter(nil)
|
||||||
|
6 - (#keys=1, [P2.C2]), output([P2.C2], [P2.C1]), filter(nil), dop=1
|
||||||
|
7 - output([P2.C1], [P2.C2]), filter(nil)
|
||||||
|
8 - output([P2.C1], [P2.C2]), filter(nil),
|
||||||
|
access([P2.C1], [P2.C2]), partitions(p[0-3])
|
||||||
|
|
||||||
|
1 row in set
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
|
## 执行计划形状与算子信息
|
||||||
|
|
||||||
|
在数据库系统中，执行计划在内部通常是以树的形式来表示的，但是不同的数据库会选择不同的方式展示给用户。
|
||||||
|
|
||||||
|
如下示例分别为 PostgreSQL 数据库、Oracle 数据库和 OceanBase 数据库对于 TPCDS Q3 的计划展示。
|
||||||
|
|
||||||
|
```sql
|
||||||
|
obclient> SELECT /*TPC-DS Q3*/ *
|
||||||
|
FROM (SELECT dt.d_year,
|
||||||
|
item.i_brand_id brand_id,
|
||||||
|
item.i_brand brand,
|
||||||
|
Sum(ss_net_profit) sum_agg
|
||||||
|
FROM date_dim dt,
|
||||||
|
store_sales,
|
||||||
|
item
|
||||||
|
WHERE dt.d_date_sk = store_sales.ss_sold_date_sk
|
||||||
|
AND store_sales.ss_item_sk = item.i_item_sk
|
||||||
|
AND item.i_manufact_id = 914
|
||||||
|
AND dt.d_moy = 11
|
||||||
|
GROUP BY dt.d_year,
|
||||||
|
item.i_brand,
|
||||||
|
item.i_brand_id
|
||||||
|
ORDER BY dt.d_year,
|
||||||
|
sum_agg DESC,
|
||||||
|
brand_id)
|
||||||
|
WHERE ROWNUM <= 100;
|
||||||
|
|
||||||
|
PostgreSQL 数据库执行计划展示如下：
|
||||||
|
Limit (cost=13986.86..13987.20 rows=27 width=91)
|
||||||
|
Sort (cost=13986.86..13986.93 rows=27 width=65)
|
||||||
|
Sort Key: dt.d_year, (sum(store_sales.ss_net_profit)), item.i_brand_id
|
||||||
|
HashAggregate (cost=13985.95..13986.22 rows=27 width=65)
|
||||||
|
Merge Join (cost=13884.21..13983.91 rows=204 width=65)
|
||||||
|
Merge Cond: (dt.d_date_sk = store_sales.ss_sold_date_sk)
|
||||||
|
Index Scan using date_dim_pkey on date_dim dt (cost=0.00..3494.62 rows=6080 width=8)
|
||||||
|
Filter: (d_moy = 11)
|
||||||
|
Sort (cost=12170.87..12177.27 rows=2560 width=65)
|
||||||
|
Sort Key: store_sales.ss_sold_date_sk
|
||||||
|
Nested Loop (cost=6.02..12025.94 rows=2560 width=65)
|
||||||
|
Seq Scan on item (cost=0.00..1455.00 rows=16 width=59)
|
||||||
|
Filter: (i_manufact_id = 914)
|
||||||
|
Bitmap Heap Scan on store_sales (cost=6.02..658.94 rows=174 width=14)
|
||||||
|
Recheck Cond: (ss_item_sk = item.i_item_sk)
|
||||||
|
Bitmap Index Scan on store_sales_pkey (cost=0.00..5.97 rows=174 width=0)
|
||||||
|
Index Cond: (ss_item_sk = item.i_item_sk)
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
|
Oracle 数据库执行计划展示如下：
|
||||||
|
Plan hash value: 2331821367
|
||||||
|
--------------------------------------------------------------------------------------------------
|
||||||
|
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
|
||||||
|
--------------------------------------------------------------------------------------------------
|
||||||
|
| 0 | SELECT STATEMENT | | 100 | 9100 | 3688 (1)| 00:00:01 |
|
||||||
|
|* 1 | COUNT STOPKEY | | | | | |
|
||||||
|
| 2 | VIEW | | 2736 | 243K| 3688 (1)| 00:00:01 |
|
||||||
|
|* 3 | SORT ORDER BY STOPKEY | | 2736 | 256K| 3688 (1)| 00:00:01 |
|
||||||
|
| 4 | HASH GROUP BY | | 2736 | 256K| 3688 (1)| 00:00:01 |
|
||||||
|
|* 5 | HASH JOIN | | 2736 | 256K| 3686 (1)| 00:00:01 |
|
||||||
|
|* 6 | TABLE ACCESS FULL | DATE_DIM | 6087 | 79131 | 376 (1)| 00:00:01 |
|
||||||
|
| 7 | NESTED LOOPS | | 2865 | 232K| 3310 (1)| 00:00:01 |
|
||||||
|
| 8 | NESTED LOOPS | | 2865 | 232K| 3310 (1)| 00:00:01 |
|
||||||
|
|* 9 | TABLE ACCESS FULL | ITEM | 18 | 1188 | 375 (0)| 00:00:01 |
|
||||||
|
|* 10 | INDEX RANGE SCAN | SYS_C0010069 | 159 | | 2 (0)| 00:00:01 |
|
||||||
|
| 11 | TABLE ACCESS BY INDEX ROWID| STORE_SALES | 159 | 2703 | 163 (0)| 00:00:01 |
|
||||||
|
--------------------------------------------------------------------------------------------------
|
||||||
|
|
||||||
|
OceanBase 数据库执行计划展示如下：
|
||||||
|
|ID|OPERATOR |NAME |EST. ROWS|COST |
|
||||||
|
-------------------------------------------------------
|
||||||
|
|0 |LIMIT | |100 |81141|
|
||||||
|
|1 | TOP-N SORT | |100 |81127|
|
||||||
|
|2 | HASH GROUP BY | |2924 |68551|
|
||||||
|
|3 | HASH JOIN | |2924 |65004|
|
||||||
|
|4 | SUBPLAN SCAN |VIEW1 |2953 |19070|
|
||||||
|
|5 | HASH GROUP BY | |2953 |18662|
|
||||||
|
|6 | NESTED-LOOP JOIN| |2953 |15080|
|
||||||
|
|7 | TABLE SCAN |ITEM |19 |11841|
|
||||||
|
|8 | TABLE SCAN |STORE_SALES|161 |73 |
|
||||||
|
|9 | TABLE SCAN |DT |6088 |29401|
|
||||||
|
=======================================================
|
||||||
|
|
||||||
|
由示例可见，OceanBase 数据库的计划展示与 Oracle 数据库类似。
|
||||||
|
|
||||||
|
OceanBase 数据库执行计划中的各列的含义如下：
|
||||||
|
列名 含义
|
||||||
|
ID 执行树按照前序遍历的方式得到的编号（从 0 开始）。
|
||||||
|
OPERATOR 操作算子的名称。
|
||||||
|
NAME 对应表操作的表名（索引名）。
|
||||||
|
EST. ROWS 估算该操作算子的输出行数。
|
||||||
|
COST 该操作算子的执行代价（微秒）。
|
||||||
|
|
||||||
|
|
||||||
|
OceanBase 数据库 EXPLAIN 命令输出的第一部分是执行计划的树形结构展示。其中每一个操作在树中的层次通过其在 operator 中的缩进予以展示，层次最深的优先执行，层次相同的以特定算子的执行顺序为标准来执行。
|
||||||
|
|
||||||
|
问题: update a not exists (b…)
|
||||||
|
我一开始以为 B是驱动表，B的数据挺多的 后来看到NLAJ，是说左边的表关联右边的表
|
||||||
|
所以这个的驱动表是不是实际是A，用A的匹配B的，这个理解有问题吗
|
||||||
|
|
||||||
|
回答: 没错 A 驱动 B的
|
||||||
|
|
||||||
|
问题: 光知道最下最右的是驱动表了 所以一开始搞得有点懵 :sweat_smile:
|
||||||
|
|
||||||
|
回答: nlj应该原理应该都是左表(驱动表)的记录探测右表(被驱动表)， 选哪张成为左表或右表就基于一些其他考量了，比如数据量， 而anti join/semi join只是对 not exist/exist的一种优化，相关的原理和资料网上可以查阅一下
|
||||||
|
|
||||||
|
问题: 也就是nlj 就是按照之前理解的谁先执行 谁就是驱动表 也就是执行计划中的最右的表
|
||||||
|
而anti join/semi join，谁在not exist左面，谁就是驱动表。这么理解对吧
|
||||||
|
|
||||||
|
回答: nlj也是左表的表是驱动表，这个要了解下计划执行方面的基本原理，取左表的一行数据，再遍历右表，一旦满足连接条件，就可以返回数据
|
||||||
|
anti/semi只是因为not exists/exist的语义只是返回左表数据，改成anti join是一种计划优化，连接的方式比子查询更优
|
||||||
@ -227,7 +227,7 @@ def build_single_model_ui():
|
|||||||
"""
|
"""
|
||||||
|
|
||||||
state = gr.State()
|
state = gr.State()
|
||||||
notice = gr.Markdown(notice_markdown, elem_id="notice_markdown")
|
gr.Markdown(notice_markdown, elem_id="notice_markdown")
|
||||||
|
|
||||||
with gr.Accordion("参数", open=False, visible=False) as parameter_row:
|
with gr.Accordion("参数", open=False, visible=False) as parameter_row:
|
||||||
temperature = gr.Slider(
|
temperature = gr.Slider(
|
||||||
@ -248,22 +248,29 @@ def build_single_model_ui():
|
|||||||
label="最大输出Token数",
|
label="最大输出Token数",
|
||||||
)
|
)
|
||||||
|
|
||||||
chatbot = grChatbot(elem_id="chatbot", visible=False).style(height=550)
|
with gr.Tabs():
|
||||||
with gr.Row():
|
with gr.TabItem("知识问答", elem_id="QA"):
|
||||||
with gr.Column(scale=20):
|
pass
|
||||||
textbox = gr.Textbox(
|
|
||||||
show_label=False,
|
|
||||||
placeholder="Enter text and press ENTER",
|
|
||||||
visible=False,
|
|
||||||
).style(container=False)
|
|
||||||
|
|
||||||
with gr.Column(scale=2, min_width=50):
|
with gr.TabItem("SQL生成与诊断", elem_id="SQL"):
|
||||||
send_btn = gr.Button(value="" "发送", visible=False)
|
# TODO A selector to choose database
|
||||||
|
pass
|
||||||
|
|
||||||
|
with gr.Blocks():
|
||||||
|
chatbot = grChatbot(elem_id="chatbot", visible=False).style(height=550)
|
||||||
|
with gr.Row():
|
||||||
|
with gr.Column(scale=20):
|
||||||
|
textbox = gr.Textbox(
|
||||||
|
show_label=False,
|
||||||
|
placeholder="Enter text and press ENTER",
|
||||||
|
visible=False,
|
||||||
|
).style(container=False)
|
||||||
|
with gr.Column(scale=2, min_width=50):
|
||||||
|
send_btn = gr.Button(value="发送", visible=False)
|
||||||
|
|
||||||
|
|
||||||
with gr.Row(visible=False) as button_row:
|
with gr.Row(visible=False) as button_row:
|
||||||
regenerate_btn = gr.Button(value="🔄" "重新生成", interactive=False)
|
regenerate_btn = gr.Button(value="重新生成", interactive=False)
|
||||||
clear_btn = gr.Button(value="🗑️" "清理", interactive=False)
|
clear_btn = gr.Button(value="清理", interactive=False)
|
||||||
|
|
||||||
gr.Markdown(learn_more_markdown)
|
gr.Markdown(learn_more_markdown)
|
||||||
|
|
||||||
|
|||||||
23
pilot/vector_store/extract_tovec.py
Normal file
23
pilot/vector_store/extract_tovec.py
Normal file
@ -0,0 +1,23 @@
|
|||||||
|
#!/usr/bin/env python3
|
||||||
|
# -*- coding:utf-8 -*-
|
||||||
|
|
||||||
|
|
||||||
|
from langchain.text_splitter import CharacterTextSplitter
|
||||||
|
from langchain.vectorstores import Chroma
|
||||||
|
from pilot.model.vicuna_llm import VicunaEmbeddingLLM
|
||||||
|
|
||||||
|
embeddings = VicunaEmbeddingLLM()
|
||||||
|
|
||||||
|
def knownledge_tovec(filename):
|
||||||
|
with open(filename, "r") as f:
|
||||||
|
knownledge = f.read()
|
||||||
|
|
||||||
|
text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
|
||||||
|
texts = text_splitter.split_text(knownledge)
|
||||||
|
docsearch = Chroma.from_texts(
|
||||||
|
texts, embeddings, metadatas=[{"source": str(i)} for i in range(len(texts))]
|
||||||
|
)
|
||||||
|
return docsearch
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
Loading…
Reference in New Issue
Block a user