By default, Looker uses a HyperSQL in-memory database to store its configuration, users, and other data. On a busy instance, this database can grow to be gigabytes in size, which can lead to performance issues, Java memory pressure, and long startup times.
We recommend that you replace the HyperSQL database with a full MySQL database backend when the internal HyperSQL database exceeds 600 MB in size. To check the size of the HyperSQL database, view the size of the
looker.script file exceeds 600 MB in size, follow the following procedures to migrate to an external MySQL database.
This procedure assumes a deployment in AWS EC2. For local deployments, systems should be sized comparably to the equivalent AWS instances.Customers updating to Looker 6.6 or later from a version earlier than Looker 6.6 cannot perform the Looker update and migrate to a MySQL backend DB at the same time. If you are updating Looker and migrating to MySQL, we recommend that you complete the Looker update before performing the migration to MySQL.
Provision a MySQL instance
Provision a MySQL 8.0.x instance to use as the backend. Looker also supports MySQL version 5.7.x. MySQL versions before 5.7.x are not supported because they do not support UTF8mb4 encoding.
In AWS RDS, an instance of class
db.m5.large is probably sufficient as a backend for a single Looker instance. Even though the database’s actual usage will likely be in the 5-10 GB range, it’s a good idea to provision 100-150 GB of SSD storage because the provisioned IOPS is based on the amount of storage requested.
max_allowed_packet size is too small for database migration and can cause the migration to fail with a
PACKET_TOO_LARGE error. Set
max_allowed_packet to the maximum allowed value of
In addition, set the following default parameters to use UTF8mb4, which supports UTF8 character sets. See the article In MySQL, never use “utf8”. Use “utf8mb4”. for information about why we recommend the use of UTF8mb4 — not UTF8 — with MySQL.
On Amazon RDS instances, you apply this setting by creating or modifying a parameter group and editing the appropriate settings. We recommend that you copy the current parameter group and make the changes on the copy, especially if you are sharing parameter groups across several RDS instances. After saving the parameter group, apply it to the RDS instance. A reboot may be required.
Set your replica scheme
Looker relies on functionality that necessitates a
row binlog. If you are hosting your own MySQL instance, set your
row by issuing one of the following commands:
Create a database and user
Create a user and a database on the database instance, replacing
<DB_password> with the actual values for the user and database. Also replace
<DB_collation> with the chosen character set and collation that matches the RDS instance param group settings (for true UTF8 support, we recommend
looker_tmp database on the last line doesn’t have to actually exist, but the
grant statement is needed for internal reporting.
Create a database credentials file
Looker needs to know which MySQL database to talk to and which credentials to use. In the Looker directory, create a file named
looker-db.yml with the following contents, replacing
<DB_name> with values for your database:
If your MySQL database requires an SSL connection, add the following line to
If you also want to enable verification of the SSL certificate, add the following line to
Optionally, you can also specify any other additional JDBC parameters that are supported by the MariaDB JDBC Driver by adding
jdbc_additional_params. For example, if you need to use a specific Trust Store file, you can add the following parameter to the MySQL JDBC connection string:
Security recommendation: Follow best-practice security considerations when saving credentials to a file. Ideally, set the
looker-db.ymlfile permissions to 600, owned by the Linux “user” account under which the Looker application is executed. This file should never be checked into a Git repository.
Under Looker’s encryption scheme, all sensitive data in the database is encrypted at rest. Even if someone were to gain access to plaintext database credentials and access the database, Looker encrypts or hashes sensitive data before storing. This applies to passwords, analytics database credentials, query cache, and so on. However, if you do not want to store the cleartext password for this configuration in the
looker-db.yml file on disk, you can configure the environment variable
LOOKER_DB to contain a list of keys/values for each line in the
looker-db.yml file. For example:
Security recommendation: Limit use of your MySQL user account to the IP address used by your Looker server.
Back up the
Back up the
.db directory, which contains the files needed to build the in-memory HyperSQL database, in case you need to restore HyperSQL:
Migrate the database
Migrating the database to MySQL can take hours on a medium or large instance, especially if the HyperSQL database is 1 GB or more. We recommend that you temporarily upgrade the EC2 instance to an
m5.2xlarge (with 32 GB RAM to allow the 26 GB heap specified in the steps) during the migration, which reduces the time required to ~10 minutes.
On the Looker host:cd looker ./looker stop vi looker
In the Looker startup script, make a new second line in the file:exit
Stop the instance in the AWS console. Once it stops, change the EC2 instance size to
m5.2xlarge. Then start the instance back up again.
SSH to the host as the Looker user. First make sure Java isn’t running; then run:cd looker java -Xms26000m -Xmx26000m -jar looker.jar migrate_internal_data looker-db.yml
When running the
libcryptmay not be found and a stack trace will appear, starting with this:NotImplementedError: getppid unsupported or native support failed to load ppid at org/jruby/RubyProcess.java:752 ppid at org/jruby/RubyProcess.java:749
If this happens, set the
LD_LIBRARY_PATHmanually before executing the Java command:export LD_LIBRARY_PATH=$HOME/looker/.tmp/:$LD_LIBRARY_PATH
Once that successfully completes, stop the instance from the AWS console.
You can now restore the instance to its original size.
Start the instance again.
Edit the Looker startup script and delete the
exitline you added earlier.
Ensure that there are no arguments defined in
LOOKERARGSin the startup script. Instead, any arguments should move to the
lookerstart.cfgfile so that they will not be overwritten by new versions of the startup script. Save and exit the startup script.
lookerstart.cfg. It should look similar to the following:LOOKERARGS="-d looker-db.yml"
If there were any other arguments in the Looker startup script, add them to the
.dbdirectory, if it is not archived already.mv .db .db-backup tar -zcvf db-backup.tar.gz ./.db-backup rm -rf ./.db-backup/
Start Looker:./looker start
Verify that Looker is using the new database
If Looker is successfully using the backend MySQL, you should see network connections between the Looker instance and the new database instance. To check this, run the following command on the Looker instance:
You should see some connections to the database instance. Below is a sample output, showing a DB instance at IP address
Backing up Looker
After you migrate to a MySQL backend, Looker’s automated S3 backups will no longer function. We recommend at least nightly backups of the MySQL database along with nightly file system backups of the Looker working directory. The
looker/log/ directory may be excluded from the file system backups. See the Creating backups documentation page for more information.