It's a common problem to have instances consuming lots of memory or running into out-of-memory (OOM) events. A database instance running with high memory utilization often causes performance issues, stalls, or even database downtime.
Some MySQL memory blocks are used globally. This means that all query workloads share memory locations, are occupied all the time, and are released only when the MySQL process stops. Some memory blocks are session based, which means that as soon as the session closes, memory used by that session is also released back to the system.
Whenever there is high memory usage by a Cloud SQL for MySQL instance, Cloud SQL recommends that you identify the query or process that's using a lot of memory and release it. MySQL memory consumption is divided into three major parts:
- Threads and process memory consumption
- Buffer memory consumption
- Cache memory consumption
Threads and process memory consumption
Each user session consumes memory depending on the queries running, buffers, or cache used by that session and is controlled by the session parameters of MySQL. The major parameters include:
thread_stacknet_buffer_lengthread_buffer_sizeread_rnd_buffer_sizesort_buffer_sizejoin_buffer_sizemax_heap_table_sizetmp_table_size
If there are N number of queries running at a particular time, then each query consumes memory according to these parameters during the session.
Buffer memory consumption
This part of memory is common for all the queries and is controlled by parameters
such as innodb_buffer_pool_size, innodb_log_buffer_size, and key_buffer_size.
The InnoDB buffer pool, which is configured by the innodb_buffer_pool_size flag,
occupies a significant amount of memory on your
Cloud SQL for MySQL instance and serves as a
cache for improving performance. To reduce the risk of
out-of-memory (OOM) events, you can
enable managed buffer pool.
Cache memory consumption
Cache memory includes a query cache, which is used to save the queries and their
results for faster data retrieval of the same subsequent queries. It also
includes the binlog cache to hold the changes made to the binary log while the
transaction is running, and is controlled by binlog_cache_size.
Other memory consumption
Memory is also used by join and sort operations. If your queries use join or sort
operations, those queries use memory on the basis of join_buffer_size and
sort_buffer_size.
Apart from this, if you enable the performance schema, it consumes memory. To check the memory usage by the performance schema, use the following query:
SELECT *
FROM
performance_schema.memory_summary_global_by_event_name
WHERE EVENT_NAME LIKE 'memory/performance_schema/%';
There are many instruments available in MySQL that you can set up to monitor the memory usage through the performance schema. To learn more, see the MySQL documentation.
The MyISAM-related parameter for bulk data insertion is bulk_insert_buffer_size.
To learn about how MySQL uses memory, see the MySQL documentation.
Recommendations
The following sections offer some recommendations for optimal memory usage.
Enable managed buffer pool
Enabling managed buffer pool helps you reduce the memory consumption of the
InnoDB buffer pool (or innodb_buffer_pool_size) when instance memory is high.
This reduction frees up memory that other database processes can then consume.
If your instance memory usage is high, then your instance can experience
out-of-memory (OOM) events. We recommend that you enable managed buffer pool on
your instance to help prevent OOM events.
When memory usage stabilizes at a lower value for 10 minutes or longer, MySQL
increases the value of innodb_buffer_pool_size incrementally to its original
value. You can also increase the value of the innodb_buffer_pool_size flag to
a chosen value after memory usage has stabilized.
Eligibility criteria
You can't enable managed buffer pool for shared core instances, or for MySQL 5.6 or MySQL 5.7.
Enable the feature
To enable managed buffer pool for your instance, set the
innodb_cloudsql_managed_buffer_pool flag to on. For more information about
setting database flags, see Set a database
flag.
Changing the value of the innodb_cloudsql_managed_buffer_pool flag doesn't
require a restart of the Cloud SQL instance.
If you've enabled the managed buffer pool and the memory consumption of your
instance exceeds the default threshold percentage of its allocated memory, then
Cloud SQL begins to reduce the size of its innodb_buffer_pool_size.
This default threshold percentage varies between 90% and 97% depending on the
RAM capacity of your instance. To modify the threshold, set the
innodb_cloudsql_managed_buffer_pool_threshold_pct flag to a different
percentage value. For example, to adjust the threshold to 97%, use the following
command:
gcloud sql instances patch INSTANCE_NAME \
--database-flags=EXISTING_FLAGS,innodb_cloudsql_managed_buffer_pool=on,\
innodb_cloudsql_managed_buffer_pool_threshold_pct=97
You can set the innodb_cloudsql_managed_buffer_pool_threshold_pct flag to an
integer value between 50 and 99. Changing the value of the memory usage
threshold doesn't require a restart of the Cloud SQL instance.
Adjustment logic
Managed buffer pool doesn't shrink the innodb_buffer_pool_size to a fixed,
predetermined minimum size. Instead, it reduces the size iteratively and
dynamically until total instance memory utilization drops back below the
configured threshold percentage
(innodb_cloudsql_managed_buffer_pool_threshold_pct). It shrinks the buffer
pool by adjusting the value of the innodb_buffer_pool_size flag, leveraging
InnoDB's built-in buffer pool resize functionality.
To prevent the innodb_buffer_pool_size from shrinking to a size that would
severely impact performance when memory usage remains high despite reduction,
the feature uses an internal safety floor. Values represent the percentage of
total instance memory that must be allocated to the buffer pool.
| MySQL container size | Minimum buffer pool size |
|---|---|
| 1025–2048 MB | 35% |
| 2049–6528 MB | 30% |
| 6529–11315 MB | 40% |
| 11316–22630 MB | 45% |
| Other sizes (default) | 50% |
The amount by which innodb_buffer_pool_size decreases depends on the memory
capacity of your database instance. The following table shows the percentage
decrease for each container size:
| MySQL container size | Percentage decrease |
|---|---|
| 1025–2048 MB | 15% |
| 2049–6528 MB | 11% |
| 6529–11315 MB | 8% |
| 11316–22630 MB | 6% |
| Other sizes (default) | 5% |
After calculating the new, reduced value, the managed buffer pool rounds
innodb_buffer_pool_size to the nearest multiple of the
innodb_buffer_pool_instances and innodb_buffer_pool_chunk_size values.
When managed buffer pool makes adjustments to value of
innodb_buffer_pool_size, the changes aren't reflected in Cloud de Confiance console. To
view the current value of innodb_buffer_pool_size when managed buffer pool is
enabled, you can use the MySQL client:
mysql> SHOW GLOBAL VARIABLES LIKE 'innodb_buffer_pool_size';
Limitations
Reducing the buffer pool size can't prevent OOMs in all cases. For example, some workloads might consume memory unsustainably or increase at a sudden rate, some Cloud SQL instances might be under provisioned, or the buffer pool might not be warmed up. Cloud SQL might not be able to free up memory quickly enough to accommodate sudden changes in memory workload. In addition, Cloud SQL can't accommodate misconfigured values of other memory flags.
Monitoring
You can monitor managed buffer pool in the MySQL error log. In Logs Explorer, you can filter the mysql.err log for entries with the prefix Managed Buffer Pool Plugin: or Tuner Plugin: to find the latest adjustment events.
When managed buffer pool first starts up, it emits a log like the following:
[Note] [MY-000000] [Server] Managed Buffer Pool Plugin: MySQL Instance memory limit: 29533, Current MySQL memory usage: 2663641088, Max Allowed MySQL memory usage: 30732730368 ...
The following logs show examples of automated decreases to the innodb_buffer_pool_size:
[Note] [MY-000000] [Server] Managed Buffer Pool Plugin: Decreasing InnoDB Buffer Pool Size.
[Note] [MY-000000] [Server] Managed Buffer Pool Plugin: Updated innodb_buffer_pool_size=805306368 bytes.
You can configure log-based metrics to track managed buffer pool adjustment events over time.
Use Metrics Explorer to identify the memory usage
You can review the memory usage of an instance with the database/memory/components.usage metric in Metrics Explorer.
In general, if you have less than 10% memory in database/memory/components.cache and
database/memory/components.free combined, the risk of an OOM event is high.
To monitor the memory usage and to prevent OOM events,
we recommend that you set up an alerting policy
with a metric threshold condition in database/memory/components.usage.
The following table shows the relationship between your instance memory and the recommended alerting threshold:
| Instance memory | Recommended alerting threshold |
|---|---|
| Less than or equal to 16 GB | 90% |
| Greater than 16 GB | 95% |
Calculate memory consumption
Calculate the maximum memory usage by your MySQL database to select the appropriate instance type for your MySQL database. Use the following formula:
Maximum MySQL memory usage = innodb_buffer_pool_size + innodb_additional_mem_pool_size + innodb_log_buffer_size + tmp_table_size + key_buffer_size + ((read_buffer_size + read_rnd_buffer_size + sort_buffer_size + join_buffer_size) x max_connections)
Here are the parameters used in the formula:
innodb_buffer_pool_size: The size in bytes of the buffer pool, the memory area where InnoDB caches table and index data.innodb_additional_mem_pool_size: The size in bytes of a memory pool InnoDB uses to store data dictionary information and other internal data structures.innodb_log_buffer_size: The size in bytes of the buffer that InnoDB uses to write to the log files on disk.tmp_table_size: The maximum size of internal in-memory temporary tables created by the MEMORY storage engine and, as of MySQL 8.0.28, the TempTable storage engine.key_buffer_size: The size of the buffer used for index blocks. Index blocks for MyISAM tables are buffered and are shared by all threads.read_buffer_size: Each thread that does a sequential scan for a MyISAM table allocates a buffer of this size (in bytes) for each table it scans.read_rnd_buffer_size: This variable is used for reads from MyISAM tables, for any storage engine, and for Multi-Range Read optimization.sort_buffer_size: Each session that must perform a sort allocates a buffer of this size. sort_buffer_size is not specific to any storage engine and applies in a general manner for optimization.join_buffer_size: The minimum size of the buffer that's used for plain index scans, range index scans, and joins that don't use indexes and thus perform full table scans.max_connections: The maximum permitted number of simultaneous client connections.
Troubleshoot high memory consumption
Run
SHOW PROCESSLISTto see the ongoing queries that are consuming memory. It displays all connected threads and their running SQL statements and tries to optimize them. Pay attention to the state and duration columns.mysql> SHOW [FULL] PROCESSLIST;Check
SHOW ENGINE INNODB STATUSin the sectionBUFFER POOL AND MEMORYto see the current buffer pool and memory usage, which can help you set your buffer pool size.mysql> SHOW ENGINE INNODB STATUS \G ---------------------- BUFFER POOL AND MEMORY ---------------------- Total memory allocated 398063986; in additional pool allocated 0 Dictionary memory allocated 12056 Buffer pool size 89129 Free buffers 45671 Database pages 1367 Old database pages 0 Modified db pages 0Use MySQL's
SHOW variablescommand to check the counter values, which give you information such as number of temporary tables, number of threads, number of table caches, dirty pages, open tables, and buffer pool usage.mysql> SHOW variables like 'VARIABLE_NAME'
Apply changes
After you analyze the memory usage by different components, set the appropriate flag in your MySQL database. To change the flag in Cloud SQL for MySQL instance, you can use Cloud de Confiance console or gcloud CLI. To change the flag value using the Cloud de Confiance console, edit the Flags section, select the flag, and enter the new value.
Lastly, if the memory usage is still high and you feel running queries and flag values are optimized, then consider increasing the instance size to avoid OOM.