In this blog post, I’m gonna talk about MySQL Memory Issues. What causes High memory usage and what is the impact on MySQL database instances.
But first, we can do a quick review on the following topics:
How MySQL uses memory? (Log files, buffers, InnoDB buffer pool size)
How MySQL allocates memory?
Swap memory
Troubleshooting
Best practices.
Globally shared memory is allocated upon the creation of an instance and is shared by all connections Example: thread cache, query cache, buffer pool, and log buffer.
Private memory is used to allocate cache upon connection to MySQL server. It is allocated for each thread. For example; sort buffer, join buffer, and temporary table.
InnoDB maintains one or more buffer pools that cache frequently used data and indexes in the main memory.
So, when a read query is executed from the client program, InnoDB checks to see if the required data pages are in the buffer pool. If it is not in the buffer pool, InnoDB requests the data from the tablespace. Then it will put the data pages in the buffer pool. And then, MySQL will return the results to the client.
When transactions are performed, data changes are written to the log buffer in memory. then every time a transaction is committed, the transaction logs get flushed to the disk, but this behavior can be changed (InnoDB_flush_log_at_trx_commit). If a crash occurs (while the tables are being modified), the redo log files are used for recovery.
On the other hand, Dirty pages in the buffer pool get written to the disk at checkpoints. This is to ensure that the cache has enough free space.
innodb_buffer_pool_size is the most important tuning parameter. It caches table and index data. The buffer pool permits frequently used data to be accessed directly from memory, which speeds up processing.
Setting it too low can degrade the performance. Setting it too high can increase the memory consumption causing the DB to crash.
MySQL allocation is not only from innodb_buffer_pool_size, but also from other buffers in the database such sort_buffer_size, read_buffer_size, read_rnd_buffer, join_buffer_size, and tmp_table_size and this will require additional 5-10% extra memory
Below is a formula to calculate the approximate memory usage for your MySQL:
Maximum MySQL Memory Usage = innodb_buffer_pool_size + key_buffer_size + ((read_buffer_size + read_rnd_buffer_size + sort_buffer_size + join_buffer_size) X max_connections)
Swapping can happen when a system requires more memory than is allocated .A large buffer may lead to swapping in the operating system and make the performance slow.
To avoid your MySQL/MariaDB data being SWAP instead of RAM, you have to play with a kernel parameter called swappiness.
A swappiness value is used to change the balance between swapping out runtimememory and dropping pages from the system page cache. The higher the value, the more the system will swap. The lower the value, the less the system will swap. The maximum value is 100, the minimum is 0, and 60 is the default.
I performed a test in AWS RDS MySQL. I ran a stored procedure that consumed a lot of memory until the server crashed. We will notice that when memory gets low, Swap usage will increase. We will also see a spike in disk IO usage.
When MySQL crashed, we will lose connection to the database. In AWS RDS, we can see the logs and EVENTs to identify the cause of the crash. The log indicates that the database process was killed by the OS due to excessive memory consumption.
I also did the same test in Azure MySQL. The difference between Azure and AWS is that when the DB crash, in Azure, it will failover to the Standby replica, but in AWS, a mysqld process crash would not trigger a failover.
In Azure, under Private DNS Zone, if we see that the IP address changed, then it means that it failed over to the standby replica successfully.
This blog post describes how to deploy a TiDB cluster on GCP Google Kubernetes Engine (GKE). TiDB on Kubernetes is the standard way to deploy TiDB on public clouds.
TiDB Architecture
TiDB is designed to consist of multiple components. These components communicate with each other and form a complete TiDB system. The architecture is as follows:
TiDB server
The TiDB server is a stateless SQL layer that exposes the connection endpoint of the MySQL protocol to the outside. The TiDB server receives SQL requests, performs SQL parsing and optimization, and ultimately generates a distributed execution plan. It is horizontally scalable and provides the unified interface to the outside through the load balancing components such as Linux Virtual Server (LVS), HAProxy, or F5. It does not store data and is only for computing and SQL analyzing, transmitting actual data read request to TiKV nodes (or TiFlash nodes).
Placement Driver (PD) server
The PD server is the metadata managing component of the entire cluster. It stores metadata of real-time data distribution of every single TiKV node and the topology structure of the entire TiDB cluster, provides the TiDB Dashboard management UI, and allocates transaction IDs to distributed transactions. The PD server is “the brain” of the entire TiDB cluster because it not only stores metadata of the cluster, but also sends data scheduling command to specific TiKV nodes according to the data distribution state reported by TiKV nodes in real time. In addition, the PD server consists of three nodes at least and has high availability. It is recommended to deploy an odd number of PD nodes.
Storage servers
Storage servers
TiKV server
The TiKV server is responsible for storing data. TiKV is a distributed transactional key-value storage engine. Region is the basic unit to store data. Each Region stores the data for a particular Key Range which is a left-closed and right-open interval from StartKey to EndKey. Multiple Regions exist in each TiKV node. TiKV APIs provide native support to distributed transactions at the key-value pair level and supports the Snapshot Isolation level isolation by default. This is the core of how TiDB supports distributed transactions at the SQL level. After processing SQL statements, the TiDB server converts the SQL execution plan to an actual call to the TiKV API. Therefore, data is stored in TiKV. All the data in TiKV is automatically maintained in multiple replicas (three replicas by default), so TiKV has native high availability and supports automatic failover.
TiFlash server
The TiFlash Server is a special type of storage server. Unlike ordinary TiKV nodes, TiFlash stores data by column, mainly designed to accelerate analytical processing.
Prerequisites
Before deploying a TiDB cluster on GCP GKE, make sure the following requirements are satisfied
1) Create a project
2) Enable Kubernetes Engine API
3) Activate Cloud Shell
Ensure that you have the available quote for Compute Engine CPU in your cluster’s region.
4) Configure the GCP service
Configure your GCP project and default region.
gcloud config set core/project
gcloud config set compute/region
Example:
gcloud config set core/project erudite-spot-326413
gcloud config set compute/zone us-west1-a
TiDB Operator uses Custom Resource Definition (CRD) to extend Kubernetes. Therefore, to use TiDB Operator, you must first create the TidbCluster CRD, which is a one-time job in your Kubernetes cluster.
kubectl get pods --namespace tidb-admin -l app.kubernetes.io/instance=tidb-admin
kubectl get pods --namespace tidb-admin -o wide
Deploy a TiDB Cluster and the Monitoring Component
This section describes how to deploy a TiDB cluster and its monitoring services.
Create namespace
kubectl create namespace tidb-cluster
Note: A namespace is a virtual cluster backed by the same physical cluster. This document takes tidb-cluster as an example. If you want to use other namespace, modify the corresponding arguments of -n or –namespace.
Download the sample TidbCluster and TidbMonitor configuration files
Wait until all Pods for all services are started. As soon as you see Pods of each type (-pd, -tikv, and -tidb) are in the “Running” state, you can press Ctrl+C to get back to the command line and go on to connect to your TiDB cluster.
View the cluster status
kubectl get pods -n tidb-cluster
Get list of services in the tidb-cluster
kubectl get svc -n tidb-cluster
Access the TiDB database
After you deploy a TiDB cluster, you can access the TiDB database via MySQL client.
Prepare a bastion host
The LoadBalancer created for your TiDB cluster is an intranet LoadBalancer. You can create a bastion host in the cluster VPC to access the database.
Note: You can also create the bastion host in other zones in the same region.
Note: In the regional cluster, the nodes are created in 3 zones. Therefore, after scaling out, the number of nodes is 2 * 3 = 6.
After that, execute kubectl edit tc basic -n tidb-cluster and modify each component’s replicas to the desired number of replicas. The scaling-out process is then completed.
kubectl edit tc basic -n tidb-cluster
Deploy TiFlash and TiCDC
TiFlash is the columnar storage extension of TiKV.
TiCDC is a tool for replicating the incremental data of TiDB by pulling TiKV change logs.
The amount of resources that can be used can be controlled from the operating system perspective. Each user has limits that are set, but for that particular user, the limits are applied individually to each of its processes.
Limits can either be hard or soft. Only the root user can set the Hardlimits. Other users can set Soft limits, but it cannot be more than the hard limit.
The default value of ulimit open files limit is 1024. This is very low for a typical web server environment that hosts many have database-driven sites.
MySQL/MariaDB also uses this setting. The open_files_limit is set by MySQL/MariaDB to the system’s ulimit. Default is 1024.
NOTE: MySQL/MariaDB can only set its open_files_limit lower than what is specified under ulimit ‘open files’. It cannot be set higher than that.
Examine Current Limits
To inspect current limits
ulimit -a
# -a will show all current limits including hard, soft, open files, etc.
To inspect the current hard and soft limits.
# Hard Limits
ulimit -Ha
# Soft Limits
ulimit -Sa
# H for hard limits, or S for soft limits.
To check current open file limits.
ulimit -n
# –n for number of open files limits
Set ‘open files’ Limit Persistently
Open /etc/security/limits.conf using the text editor of your choice, and add the following lines, then save it.
* soft nofile 102400
* hard nofile 102400
* soft nproc 10240
* hard nproc 10240
Edit the file /etc/security/limits.d/90-nproc.conf using the text editor of your choice, and add the following lines, then save it.
* soft nofile 1024000
* hard nofile 1024000
* soft nproc 10240
* hard nproc 10240
root soft nproc unlimited
Set open_files_limit in my.cnf (MySQL)
Open and edit /etc/my.cnf
Insert the lines below under [mysql], then save it.
[mysqld]
open_files_limit = 102400
Run the command below to see if there are any .conf files being used by MySQL that overrides the values for open limits.
systemctl status mysqld
Below is something you will see after running the command above.
innochecksum prints checksums for InnoDB files. This tool reads an InnoDB tablespace file, calculates the checksum for each page, compares the calculated checksum to the stored checksum, and reports mismatches, which indicate damaged pages.
It was originally developed to speed up verifying the integrity of tablespace files after power outages but can also be used after file copies. Because checksum mismatches cause InnoDB to deliberately shut down a running server, it may be preferable to use this tool rather than waiting for an in-production server to encounter the damaged pages.
Below is a script you can use to get corrupted tables. Replace the directory path where your data files are located. (Stop MySQL/MariaDB service first, before running it)
INNOCKSM_LOG=/mysql/backup/innochecksum_`date +%Y%m%d_%H%M%S`.log
for DB in `ls -1vd /mysql/data/*/ | grep -wv '/mysql/data/mysql/\|/mysql/data/performance_schema/\|/mysql/data/lost+found/'`
do
for IBD in `ls -1v $DB | grep .ibd`
do
innochecksum ${DB}${IBD}
if [ $? -ne 0 ]; then
echo ${DB}${IBD} >> $INNOCKSM_LOG
innochecksum ${DB}${IBD} >> $INNOCKSM_LOG 2>&1
fi
done
done
