If you run into issues or have questions, please post on our forum.
To run Vitess in Docker, you can either use our pre-built images on Docker Hub, or build them yourself.
The vitess/base image contains a full development environment, capable of building Vitess and running integration tests.
The vitess/lite image contains only the compiled Vitess binaries, excluding ZooKeeper. It can run Vitess, but lacks the environment needed to build Vitess or run tests. It's primarily used for the Vitess on Kubernetes guide.
For example, you can directly run
vitess/base, and Docker will download the
image for you:
$ sudo docker run -ti vitess/base bash vitess@32f187ef9351:/vt/src/github.com/youtube/vitess$ make build
Now you can proceed to start a Vitess cluster inside the Docker container you just started. Note that if you want to access the servers from outside the container, you'll need to expose the ports as described in the Docker Engine Reference Guide.
For local testing, you can also access the servers on the local IP address created for the container by Docker:
$ docker inspect 32f187ef9351 | grep IPAddress ### example output: # "IPAddress": "172.17.3.1",
You can also build Vitess Docker images yourself to include your
own patches or configuration data. The
in the root of the Vitess tree builds the
subdirectory contains scripts for building other images, such as
Makefile also contains rules to build the images. For example:
# Create vitess/bootstrap, which prepares everything up to ./bootstrap.sh vitess$ make docker_bootstrap # Create vitess/base from vitess/bootstrap by copying in your local working directory. vitess$ make docker_base
The following sections explain the process for manually building Vitess without Docker.
We currently test Vitess regularly on Ubuntu 14.04 (Trusty) and Debian 8 (Jessie). OS X 10.11 (El Capitan) should work as well, the installation instructions are below.
In addition, Vitess requires the software and libraries listed below.
The Vitess development team currently tests against MariaDB 10.0.21 and MySQL 5.6.27.
If you are installing MariaDB, note that you must install version 10.0 or
higher. If you are using
apt-get, confirm that your repository
offers an option to install that version. You can also download the source
directly from mariadb.org.
If you are using Ubuntu 14.04 with MySQL 5.6, the default install may be
missing a file too,
/usr/share/mysql/my-default.cnf. It would show as an
Could not find my-default.cnf. If you run into this, just add
it with the following contents:
Select a lock service from the options listed below. It is technically possible to use another lock server, but plugins currently exist only for ZooKeeper and etcd.
etcdcommand on your path.
Install the following other tools needed to build and run Vitess:
These can be installed with the following apt-get command:
$ sudo apt-get install make automake libtool python-dev python-virtualenv python-mysqldb libssl-dev g++ mercurial git pkg-config bison curl unzip
If you decided to use ZooKeeper in step 3, you also need to install a Java Runtime, such as OpenJDK.
$ sudo apt-get install openjdk-7-jre
sudo chown -R $(whoami):admin /usr/local
On OS X, MySQL 5.6 has to be used, MariaDB doesn't work for some reason yet. It should be installed from Homebrew (install steps are below).
If Xcode is installed (with Console tools, which should be bundled automatically since the 7.1 version), all the dev dependencies should be satisfied in this step. If no Xcode is present, it is necessery to install pkg-config.
brew install pkg-config
ZooKeeper is used as lock service.
Run the following commands:
brew install go automake libtool python mercurial git bison curl wget homebrew/versions/mysql56 pip install --upgrade pip setuptools pip install virtualenv pip install MySQL-python pip install tox
Install Java runtime from this URL: https://support.apple.com/kb/dl1572?locale=en_US Apple only supports Java 6. If you need to install a newer version, this link might be helpful: http://osxdaily.com/2015/10/17/how-to-install-java-in-os-x-el-capitan/
The Vitess bootstrap script makes some checks for the go runtime, so it is recommended to have the following commands in your ~/.profile or ~/.bashrc or ~/.zshrc:
export PATH=/usr/local/opt/go/libexec/bin:$PATH export GOROOT=/usr/local/opt/go/libexec
There is a problem with installing the enum34 Python package using pip, so the following file has to be edited:
and this line:
has to be commented out:
After running the ./bootstrap.sh script from the next step, you can revert the change.
For the Vitess hostname resolving functions to work correctly, a new entry has to be added into the /etc/hosts file with the current LAN IP address of the computer (preferably IPv4) and the current hostname, which you get by typing the 'hostname' command in the terminal.
It is also a good idea to put the following line to force the Go DNS resolver in your ~/.profile or ~/.bashrc or ~/.zshrc:
Navigate to the directory where you want to download the Vitess
source code and clone the Vitess Github repo. After doing so,
navigate to the
cd $WORKSPACE git clone https://github.com/youtube/vitess.git \ src/github.com/youtube/vitess cd src/github.com/youtube/vitess
MYSQL_FLAVOR environment variable. Choose the appropriate
value for your database. This value is case-sensitive.
export MYSQL_FLAVOR=MariaDB # or (mandatory for OS X) # export MYSQL_FLAVOR=MySQL56
If your selected database installed in a location other than
VT_MYSQL_ROOT variable to the root directory of your
MariaDB installation. For example, if MariaDB is installed in
/usr/local/mysql, run the following command.
export VT_MYSQL_ROOT=/usr/local/mysql # on OS X, this is the correct value: # export VT_MYSQL_ROOT=/usr/local/opt/mysql56
Note that the command indicates that the
mysql executable should
be found at
mysql_config --version and confirm that you
are running the correct version of MariaDB or MySQL. The value should
be 10 or higher for MariaDB and 5.6.x for MySQL.
Build Vitess using the commands below. Note that the
bootstrap.sh script needs to download some dependencies.
If your machine requires a proxy to access the Internet, you will need
to set the usual environment variables (e.g.
Run the boostrap.sh script:
./bootstrap.sh ### example output: # skipping zookeeper build # go install golang.org/x/tools/cmd/cover ... # Found MariaDB installation in ... # creating git pre-commit hooks # # source dev.env in your shell before building
# Remaining commands to build Vitess . ./dev.env make build
Note: If you are using etcd, set the following environment variable:
The default targets when running
make test contain a full set of
tests intended to help Vitess developers to verify code changes. Those tests
simulate a small Vitess cluster by launching many servers on the local
machine. To do so, they require a lot of resources; a minimum of 8GB RAM
and SSD is recommended to run the tests.
If you want only to check that Vitess is working in your environment, you can run a lighter set of tests:
Attempts to run the full developer test suite (
on an underpowered machine often results in failure. If you still see
the same failures when running the lighter set of tests (
please let the development team know in the
A failed test can leave orphaned processes. If you use the default settings, you can use the following commands to identify and kill those processes:
pgrep -f -l '(vtdataroot|VTDATAROOT)' # list Vitess processes pkill -f '(vtdataroot|VTDATAROOT)' # kill Vitess processes
This error often means your disk is too slow. If you don't have access to an SSD, you can try testing against a ramdisk.
These errors might indicate that the machine ran out of RAM and a server crashed when trying to allocate more RAM. Some of the heavier tests require up to 8GB RAM.
This error might indicate that the machine does not have a Java Runtime installed, which is a requirement if you are using ZooKeeper as the lock server.
Some of the larger tests use up to 4GB of temporary space on disk.
After completing the instructions above to build Vitess, you can use the example scripts in the Github repo to bring up a Vitess cluster on your local machine. These scripts use ZooKeeper as the lock service. ZooKeeper is included in the Vitess distribution.
Check system settings
Some Linux distributions ship with default file descriptor limits that are too low for database servers. This issue could show up as the database crashing with the message "too many open files".
Check the system-wide
file-max setting as well as user-specific
ulimit values. We recommend setting them above 100K to be safe.
The exact procedure
may vary depending on your Linux distribution.
Configure environment variables
If you are still in the same terminal window that you used to run the build commands, you can skip to the next step since the environment variables will already be set.
If you're adapting this example to your own deployment, the only environment
variables required before running the scripts are
VTROOT to the parent of the Vitess source tree. For example, if you
make build while in
then you should set:
VTDATAROOT to the directory where you want data files and logs to
be stored. For example:
Servers in a Vitess cluster find each other by looking for dynamic configuration data stored in a distributed lock service. The following script creates a small ZooKeeper cluster:
$ cd $VTROOT/src/github.com/youtube/vitess/examples/local vitess/examples/local$ ./zk-up.sh ### example output: # Starting zk servers... # Waiting for zk servers to be ready...
After the ZooKeeper cluster is running, we only need to tell each Vitess process how to connect to ZooKeeper. Then, each process can find all of the other Vitess processes by coordinating via ZooKeeper.
Each of our scripts automatically uses the
variable to point to the global ZooKeeper instance. The global instance in
turn is configured to point to the local instance. In our sample scripts,
they are both hosted in the same ZooKeeper service.
The vtctld server provides a web interface that displays all of the coordination information stored in ZooKeeper.
vitess/examples/local$ ./vtctld-up.sh # Starting vtctld # Access vtctld web UI at http://localhost:15000 # Send commands with: vtctlclient -server localhost:15999 ...
http://localhost:15000 to verify that
vtctld is running. There won't be any information
there yet, but the menu should come up, which indicates that
vtctld is running.
The vtctld server also accepts commands from the
which is used to administer the cluster. Note that the port for RPCs
(in this case
15999) is different from the web UI port (
These ports can be configured with command-line flags, as demonstrated
For convenience, we'll use the
lvtctl.sh script in example commands,
to avoid having to type the vtctld address every time.
# List available commands vitess/examples/local$ ./lvtctl.sh help
vitess/examples/local$ ./vttablet-up.sh # Output from vttablet-up.sh is below # Starting MySQL for tablet test-0000000100... # Starting vttablet for test-0000000100... # Access tablet test-0000000100 at http://localhost:15100/debug/status # Starting MySQL for tablet test-0000000101... # Starting vttablet for test-0000000101... # Access tablet test-0000000101 at http://localhost:15101/debug/status # Starting MySQL for tablet test-0000000102... # Starting vttablet for test-0000000102... # Access tablet test-0000000102 at http://localhost:15102/debug/status
After this command completes, refresh the vtctld web UI, and you should
see a keyspace named
test_keyspace with a single shard named
This is what an unsharded keyspace looks like.
If you click on the shard box, you'll see a list of tablets in that shard. Note that it's normal for the tablets to be unhealthy at this point, since you haven't initialized them yet.
You can also click the STATUS link on each tablet to be taken to its
status page, showing more details on its operation. Every Vitess server has
a status page served at
/debug/status on its web port.
Initialize MySQL databases
Next, designate one of the tablets to be the initial master.
Vitess will automatically connect the other slaves' mysqld instances so
that they start replicating from the master's mysqld.
This is also when the default database is created. Since our keyspace is
test_keyspace, the MySQL database will be named
vitess/examples/local$ ./lvtctl.sh InitShardMaster -force test_keyspace/0 test-100 ### example output: # master-elect tablet test-0000000100 is not the shard master, proceeding anyway as -force was used # master-elect tablet test-0000000100 is not a master in the shard, proceeding anyway as -force was used
Note: Since this is the first time the shard has been started,
the tablets are not already doing any replication, and there is no
existing master. The
InitShardMaster command above uses the
to bypass the usual sanity checks that would apply if this wasn't a
brand new shard.
After running this command, go back to the Shard Status page in the vtctld web interface. When you refresh the page, you should see that one vttablet is the master and the other two are replicas.
You can also see this on the command line:
vitess/examples/local$ ./lvtctl.sh ListAllTablets test ### example output: # test-0000000100 test_keyspace 0 master localhost:15100 localhost:33100  # test-0000000101 test_keyspace 0 replica localhost:15101 localhost:33101  # test-0000000102 test_keyspace 0 replica localhost:15102 localhost:33102 
Create a table
vtctlclient tool can be used to apply the database schema across all
tablets in a keyspace. The following command creates the table defined in
# Make sure to run this from the examples/local dir, so it finds the file. vitess/examples/local$ ./lvtctl.sh ApplySchema -sql "$(cat create_test_table.sql)" test_keyspace
The SQL to create the table is shown below:
CREATE TABLE messages ( page BIGINT(20) UNSIGNED, time_created_ns BIGINT(20) UNSIGNED, message VARCHAR(10000), PRIMARY KEY (page, time_created_ns) ) ENGINE=InnoDB
Take a backup
Now that the initial schema is applied, it's a good time to take the first backup. This backup will be used to automatically restore any additional replicas that you run, before they connect themselves to the master and catch up on replication. If an existing tablet goes down and comes back up without its data, it will also automatically restore from the latest backup and then resume replication.
vitess/examples/local$ ./lvtctl.sh Backup test-0000000102
After the backup completes, you can list available backups for the shard:
vitess/examples/local$ ./lvtctl.sh ListBackups test_keyspace/0 ### example output: # 2016-05-06.072724.test-0000000102
Note: In this single-server example setup, backups are stored at
$VTDATAROOT/backups. In a multi-server deployment, you would usually mount
an NFS directory there. You can also change the location by setting the
-file_backup_storage_root flag on vtctld and vttablet, as demonstrated
Initialize Vitess Routing Schema
In the examples, we are just using a single database with no specific configuration. So we just need to make that (empty) configuration visible for serving. This is done by running the following command:
vitess/examples/local$ ./lvtctl.sh RebuildVSchemaGraph
(As it works, this command will not display any output.)
Vitess uses vtgate to route each client query to the correct vttablet. This local example runs a single vtgate instance, though a real deployment would likely run multiple vtgate instances to share the load.
client.py file is a simple sample application
that connects to vtgate and executes some queries.
To run it, you need to either:
Add the Vitess Python packages to your
client.sh wrapper script, which temporarily
sets up the environment and then runs
vitess/examples/local$ ./client.sh ### example output: # Inserting into master... # Reading from master... # (5L, 1462510331910124032L, 'V is for speed') # (15L, 1462519383758071808L, 'V is for speed') # (42L, 1462510369213753088L, 'V is for speed') # ...
There are also sample clients in the same directory for Java, PHP, and Go. See the comments at the top of each sample file for usage instructions.
If so, you can skip the tear-down since the sharding guide picks up right here. If not, continue to the clean-up steps below.
-up.sh script has a corresponding
-down.sh script to stop the servers.
vitess/examples/local$ ./vtgate-down.sh vitess/examples/local$ ./vttablet-down.sh vitess/examples/local$ ./vtctld-down.sh vitess/examples/local$ ./zk-down.sh
Note that the
-down.sh scripts will leave behind any data files created.
If you're done with this example data, you can clear out the contents of
$ cd $VTDATAROOT /path/to/vtdataroot$ rm -rf *
If anything goes wrong, check the logs in your
for error messages. There are also some tablet-specific logs, as well as
MySQL logs in the various
If you need help diagnosing a problem, send a message to our
In addition to any errors you see at the command-line, it would also help to
upload an archive of your
VTDATAROOT directory to a file sharing service
and provide a link to it.