Skip to content

Conversation

@924060929
Copy link
Contributor

Proposed changes

cherry pick from #33262, #33722, #33799

Before this pr, the query must pass through parser, analyzer, rewriter, optimizer and translator, then we can check whether this query can use sql cache, if the query is too long, or the number of join tables too big, the plan time usually >= 500ms.

This pr reduce this time by skip the fashion plan path, because we can reuse the previous physical plan and query result if no any changed. In some cases we should not parse sql from sql cache, e.g. table structure changed, data changed, user policies changed, privileges changed, contains non-deterministic functions, and user variables changed.

In my test case: query a view which has lots of join and union, and the tables has empty partition, the query latency is about 3ms. if not parse sql from sql cache, the plan time is about 550ms

## Features
1. use Config.sql_cache_manage_num to control how many sql cache be reused in on fe
2. if explain plan appear some plans contains `LogicalSqlCache` or `PhysicalSqlCache`, it means the query can use sql cache, like this:
```sql
mysql> set enable_sql_cache=true;
Query OK, 0 rows affected (0.00 sec)

mysql> explain physical plan select * from test.t;
+----------------------------------------------------------------------------------+
| Explain String(Nereids Planner)                                                  |
+----------------------------------------------------------------------------------+
| cost = 3.135                                                                     |
| PhysicalResultSink[53] ( outputExprs=[c1#0, c2#1] )                              |
| +--PhysicalDistribute[50]@0 ( stats=3, distributionSpec=DistributionSpecGather ) |
|    +--PhysicalOlapScan[t]@0 ( stats=3 )                                          |
+----------------------------------------------------------------------------------+
4 rows in set (0.02 sec)

mysql> select * from test.t;
+------+------+
| c1   | c2   |
+------+------+
|    1 |    2 |
|   -2 |   -2 |
| NULL |   30 |
+------+------+
3 rows in set (0.05 sec)

mysql> explain physical plan select * from test.t;
+-------------------------------------------------------------------------------------------+
| Explain String(Nereids Planner)                                                           |
+-------------------------------------------------------------------------------------------+
| cost = 0.0                                                                                |
| PhysicalSqlCache[2] ( queryId=78511f515cda466b-95385d892d6c68d0, backend=127.0.0.1:9050 ) |
| +--PhysicalResultSink[52] ( outputExprs=[c1#0, c2#1] )                                    |
|    +--PhysicalDistribute[49]@0 ( stats=3, distributionSpec=DistributionSpecGather )       |
|       +--PhysicalOlapScan[t]@0 ( stats=3 )                                                |
+-------------------------------------------------------------------------------------------+
5 rows in set (0.01 sec)
```

(cherry picked from commit 03bd2a3)
…ontends (apache#33722)

fix some sql cache consistence bug between multiple frontends which introduced by [enhancement](Nereids) Enable parse sql from sql cache apache#33262, fix by use row policy as the part of sql cache key.
support dynamic update the num of fe manage sql cache key

(cherry picked from commit 90abd76)
1. dry run query should not use sql cache
2. fix test sql cache in cloud mode
3. enable cache OneRowRelation and EmptyRelation in frontend to skip parse sql

(cherry picked from commit dc80ecf)
@doris-robot
Copy link

Thank you for your contribution to Apache Doris.
Don't know what should be done next? See How to process your PR

Since 2024-03-18, the Document has been moved to doris-website.
See Doris Document.

@yiguolei yiguolei merged commit b8168f8 into apache:branch-2.1 Apr 18, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants