Details
-
Bug
-
Status: Needs Feedback (View Workflow)
-
Major
-
Resolution: Unresolved
-
10.11.8, 12.2.1
Description
Hello MariaDB Team,
I would like to report what appears to be a significant performance regression affecting SELECT queries executed on complex VIEWs in MariaDB 10.11.x.
Affected versions
MariaDB 10.11.2 – fast and stable, expected performance
MariaDB 10.11.8 and newer (tested up to 12.1) – severe performance degradation
Our production environment is currently running on MariaDB 10.11.2, where the workload performs well and has been stable for a long time.
Problem summary
Under the same conditions (identical schema, data set, configuration, and steady-state workload):
MariaDB 10.11.2 executes the query in approximately 3 seconds
MariaDB 10.11.8+ executes the same query in approximately 35 seconds
This represents a performance drop of more than 10×.
Workload characteristics
Storage engine: InnoDB
Queries: SELECTs executed on complex VIEWs
VIEWs contain multiple JOINs
Frequent UPDATEs on underlying base tables
Dataset fully fits in RAM
Buffer pool hit rate close to 100%
No lock waits or long-running transactions
Disk I/O during query execution is negligible
Key observations
The issue primarily affects SELECT queries executed on VIEWs
Equivalent SELECT queries run directly against the underlying base tables are significantly faster
EXPLAIN plans look logically equivalent between versions
Increasing innodb_buffer_pool_size (to several GB) does not improve performance on 10.11.8+
Adaptive Hash Index ON/OFF does not materially change the behavior
InnoDB status shows no contention and no pending I/O
The regression is consistently reproducible
Regression scope
The problem is present in all tested versions starting from MariaDB 10.11.8 and newer (last tested version: 12.1).
I have not yet identified the exact release where the regression was introduced, as I have not performed a full version-by-version bisect.
Diagnostics provided
I am attaching EXPLAIN output for the affected query from:
MariaDB 10.11.2 (fast execution)
MariaDB 10.11.8+ (slow execution)
Please note that these diagnostics should not be interpreted as coming from intentionally different datasets. At the time of collecting this information, I did not repeatedly upgrade and downgrade the database solely for demonstration purposes, as this would have required a significant amount of time. If a strict side-by-side comparison on fully identical data is required, I can prepare that at a later stage.
--------------------
I think this behavior potentially relate to:
A regression in VIEW execution or optimization
Changes in how JOINs inside VIEWs are planned or executed
Optimizer or InnoDB behavior changes introduced between 10.11.2 and 10.11.8
Thank you very much for your time and any guidance you can provide.
Kind regards