vitess-gh/doc/TopologyService.md

Topology Service

This document describes the Topology Service, a key part of the Vitess architecture. This service is exposed to all Vitess processes, and is used to store small pieces of configuration data about the Vitess cluster, and provide cluster-wide locks. It also supports watches, and master election.

Concretely, the Topology Service features are implemented by a Lock Server, referred to as Topology Server in the rest of this document. We use a plug-in implementation and we support multiple Lock Servers (Zookeeper, etcd, Consul, …) as backends for the service.

Requirements and usage

The Topology Service is used to store information about the Keyspaces, the Shards, the Tablets, the Replication Graph, and the Serving Graph. We store small data structures (a few hundred bytes) per object.

The main contract for the Topology Server is to be very highly available and consistent. It is understood it will come at a higher latency cost and very low throughput.

We never use the Topology Server as an RPC mechanism, nor as a storage system for logs. We never depend on the Topology Server being responsive and fast to serve every query.

The Topology Server must also support a Watch interface, to signal when certain conditions occur on a node. This is used for instance to know when keyspaces topology changes (for resharding for instance).

Global vs local

We differentiate two instances of the Topology Server: the Global instance, and the per-cell Local instance:

• The Global instance is used to store global data about the topology that doesn’t change very often, for instance information about Keyspaces and Shards. The data is independent of individual instances and cells, and needs to survive a cell going down entirely.
• There is one Local instance per cell, that contains cell-specific information, and also rolled-up data from the global + local cell to make it easier for clients to find the data. The Vitess local processes should not use the Global topology instance, but instead the rolled-up data in the Local topology server as much as possible.

The Global instance can go down for a while and not impact the local cells (an exception to that is if a reparent needs to be processed, it might not work). If a Local instance goes down, it only affects the local tablets in that instance (and then the cell is usually in bad shape, and should not be used).

Furthermore, the Vitess processes will not use the Global nor the Local Topology Server to serve individual queries. They only use the Topology Server to get the topology information at startup and in the background, but never to directly serve queries.

Recovery

If a local Topology Server dies and is not recoverable, it can be wiped out. All the tablets in that cell then need to be restarted so they re-initialize their topology records (but they won’t lose any MySQL data).

If the global Topology Server dies and is not recoverable, this is more of a problem. All the Keyspace / Shard objects have to be re-created. Then the cells should recover.

Global data

This section describes the data structures stored in the global instance of the topology server.

Keyspace

The Keyspace object contains various information, mostly about sharding: how is this Keyspace sharded, what is the name of the sharding key column, is this Keyspace serving data yet, how to split incoming queries, …

An entire Keyspace can be locked. We use this during resharding for instance, when we change which Shard is serving what inside a Keyspace. That way we guarantee only one operation changes the keyspace data concurrently.

Shard

A Shard contains a subset of the data for a Keyspace. The Shard record in the global topology contains:

• the Master tablet alias for this shard (that has the MySQL master).
• the sharding key range covered by this Shard inside the Keyspace.
• the tablet types this Shard is serving (master, replica, batch, …), per cell if necessary.
• if during filtered replication, the source shards this shard is replicating from.
• the list of cells that have tablets in this shard.
• shard-global tablet controls, like blacklisted tables no tablet should serve in this shard.

A Shard can be locked. We use this during operations that affect either the Shard record, or multiple tablets within a Shard (like reparenting), so multiple jobs don’t concurrently alter the data.

VSchema data

The VSchema data contains sharding and routing information for the VTGate V3 API.

Local data

This section describes the data structures stored in the local instance (per cell) of the topology server.

Tablets

The Tablet record has a lot of information about a single vttablet process running inside a tablet (along with the MySQL process):

• the Tablet Alias (cell+unique id) that uniquely identifies the Tablet.
• the Hostname, IP address and port map of the Tablet.
• the current Tablet type (master, replica, batch, spare, …).
• which Keyspace / Shard the tablet is part of.
• the sharding Key Range served by this Tablet.
• user-specified tag map (to store per installation data for instance).

A Tablet record is created before a tablet can be running (either by vtctl InitTablet or by passing the init_* parameters to the vttablet process). The only way a Tablet record will be updated is one of:

• The vttablet process itself owns the record while it is running, and can change it.
• At init time, before the tablet starts.
• After shutdown, when the tablet gets deleted.
• If a tablet becomes unresponsive, it may be forced to spare to make it unhealthy when it restarts.

Replication graph

The Replication Graph allows us to find Tablets in a given Cell / Keyspace / Shard. It used to contain information about which Tablet is replicating from which other Tablet, but that was too complicated to maintain. Now it is just a list of Tablets.

Serving graph

The Serving Graph is what the clients use to find the per-cell topology of a Keyspace. It is a roll-up of global data (Keyspace + Shard). vtgates only open a small number of these objects and get all they need quickly.

SrvKeyspace

It is the local representation of a Keyspace. It contains information on what shard to use for getting to the data (but not information about each individual shard):

• the partitions map is keyed by the tablet type (master, replica, batch, …) and the values are list of shards to use for serving.
• it also contains the global Keyspace fields, copied for fast access.

It can be rebuilt by running vtctl RebuildKeyspaceGraph. It is automatically rebuilt when a tablet starts up in a cell and the SrvKeyspace for that cell / keyspace doesn't exist yet. It will also be changed during horizontal and vertical splits.

SrvVSchema

It is the local roll-up for the VSchema. It contains the VSchema for all keyspaces in a single object.

It can be rebuilt by running vtctl RebuildVSchemaGraph. It is automatically rebuilt when using vtctl ApplyVSchema (unless prevented by flags).

Workflows involving the Topology Server

The Topology Server is involved in many Vitess workflows.

When a Tablet is initialized, we create the Tablet record, and add the Tablet to the Replication Graph. If it is the master for a Shard, we update the global Shard record as well.

Administration tools need to find the tablets for a given Keyspace / Shard: first we get the list of Cells that have Tablets for the Shard (global topology Shard record has these) then we use the Replication Graph for that Cell / Keyspace / Shard to find all the tablets then we can read each tablet record.

When a Shard is reparented, we need to update the global Shard record with the new master alias.

Finding a tablet to serve the data is done in two stages: vtgate maintains a health check connection to all possible tablets, and they report which keyspace / shard / tablet type they serve. vtgate also reads the SrvKeyspace object, to find out the shard map. With these two pieces of information, vtgate can route the query to the right vttablet.

During resharding events, we also change the topology a lot. An horizontal split will change the global Shard records, and the local SrvKeyspace records. A vertical split will change the global Keyspace records, and the local SrvKeyspace records.

Exploring the data in a Topology Server

We store the proto3 binary data for each object.

We use the following paths for the data, in all implementations:

Global Cell:

• CellInfo path: cells/<cell name>/CellInfo
• Keyspace: keyspaces/<keyspace>/Keyspace
• Shard: keyspaces/<keyspace>/shards/<shard>/Shard
• VSchema: keyspaces/<keyspace>/VSchema

Local Cell:

• Tablet: tablets/<cell>-<uid>/Tablet
• Replication Graph: keyspaces/<keyspace>/shards/<shard>/ShardReplication
• SrvKeyspace: keyspaces/<keyspace>/SrvKeyspace
• SrvVSchema: SvrVSchema

The vtctl TopoCat utility can decode these files when using the -decode_proto option:

TOPOLOGY="-topo_implementation zk2 -topo_global_server_address global_server1,global_server2 -topo_global_root /vitess/global"

$ vtctl $TOPOLOGY TopoCat -decode_proto -long /keyspaces/*/Keyspace
path=/keyspaces/ks1/Keyspace version=53
sharding_column_name: "col1"
path=/keyspaces/ks2/Keyspace version=55
sharding_column_name: "col2"

The vtctld web tool also contains a topology browser (use the Topology tab on the left side). It will display the various proto files, decoded.

Implementations

The Topology Server interfaces are defined in our code in go/vt/topo/, specific implementations are in go/vt/topo/<name>, and we also have a set of unit tests for it in go/vt/topo/test.

This part describes the implementations we have, and their specific behavior.

If starting from scratch, please use the zk2, etcd2 or consul implementations. We deprecated the old zookeeper and etcd implementations. See the migration section below if you want to migrate.

Zookeeper zk2 implementation

This is the current implementation when using Zookeeper. (The old zookeeper implementation is deprecated).

The global cell typically has around 5 servers, distributed one in each cell. The local cells typically have 3 or 5 servers, in different server racks / sub-networks for higher resilience. For our integration tests, we use a single ZK server that serves both global and local cells.

We provide the zk utility for easy access to the topology data in Zookeeper. It can list, read and write files inside any Zoopeeker server. Just specify the -server parameter to point to the Zookeeper servers. Note the vtctld UI can also be used to see the contents of the topology data.

To configure a Zookeeper installation, let's start with the global cell service. It is described by the addresses of the servers (comma separated list), and by the root directory to put the Vitess data in. For instance, assuming we want to use servers global_server1,global_server2 in path /vitess/global:

# The root directory in the global server will be created
# automatically, same as when running this command:
# zk -server global_server1,global_server2 touch -p /vitess/global

# Set the following flags to let Vitess use this global server:
# -topo_implementation zk2
# -topo_global_server_address global_server1,global_server2
# -topo_global_root /vitess/global

Then to add a cell whose local topology servers cell1_server1,cell1_server2 will store their data under the directory /vitess/cell1:

TOPOLOGY="-topo_implementation zk2 -topo_global_server_address global_server1,global_server2 -topo_global_root /vitess/global"

# Reference cell1 in the global topology service:
vtctl $TOPOLOGY AddCellInfo \
  -server_address cell1_server1,cell1_server2 \
  -root /vitess/cell1 \
  cell1

If only one cell is used, the same Zookeeper instance can be used for both global and local data. A local cell record still needs to be created, just use the same server address, and very importantly a different root directory.

Zookeeper Observers can also be used to limit the load on the global Zookeeper. They are configured by specifying the adresses of the observers in the server address, after a |, for instance: global_server1:p1,global_server2:p2|observer1:po1,observer2:po2.

Implementation details

We use the following paths for Zookeeper specific data, in addition to the regular files:

• Locks sub-directory: locks/ (for instance: keyspaces/<keyspace>/Keyspace/locks/ for a keyspace)
• Master election path: elections/<name>

Both locks and master election are implemented using ephemeral, sequential files which are stored in their respective directory.

etcd etcd2 implementation (new version of etcd)

This topology service plugin is meant to use etcd clusters as storage backend for the topology data. This topology service supports version 3 and up of the etcd server.

This implementation is named etcd2 because it supersedes our previous implementation etcd. Note that the storage format has been changed with the etcd2 implementation, i.e. existing data created by the previous etcd implementation must be migrated manually (See migration section below).

To configure an etcd2 installation, let's start with the global cell service. It is described by the addresses of the servers (comma separated list), and by the root directory to put the Vitess data in. For instance, assuming we want to use servers http://global_server1,http://global_server2 in path /vitess/global:

# Set the following flags to let Vitess use this global server,
# and simplify the example below:
# -topo_implementation etcd2
# -topo_global_server_address http://global_server1,http://global_server2
# -topo_global_root /vitess/global
TOPOLOGY="-topo_implementation etcd2 -topo_global_server_address http://global_server1,http://global_server2 -topo_global_root /vitess/global

Then to add a cell whose local topology servers http://cell1_server1,http://cell1_server2 will store their data under the directory /vitess/cell1:

# Reference cell1 in the global topology service:
# (the TOPOLOGY variable is defined in the previous section)
vtctl $TOPOLOGY AddCellInfo \
  -server_address http://cell1_server1,http://cell1_server2 \
  -root /vitess/cell1 \
  cell1

If only one cell is used, the same etcd instances can be used for both global and local data. A local cell record still needs to be created, just use the same server address and, very importantly, a different root directory.

Implementation details

For locks, we use a subdirectory named locks in the directory to lock, and an ephemeral file in that subdirectory (it is associated with a lease, whose TTL can be set with the -topo_etcd_lease_duration flag, defaults to 30 seconds). The ephemeral file with the lowest ModRevision has the lock, the others wait for files with older ModRevisions to disappear.

Master elections also use a subdirectory, named after the election Name, and use a similar method as the locks, with ephemeral files.

We store the proto3 binary data for each object (as the v3 API allows us to store binary data).

Consul consul implementation

This topology service plugin is meant to use Consul clusters as storage backend for the topology data.

To configure a consul installation, let's start with the global cell service. It is described by the address of a server, and by the root node path to put the Vitess data in (it cannot start with /). For instance, assuming we want to use servers global_server:global_port with node path vitess/global:

# Set the following flags to let Vitess use this global server,
# and simplify the example below:
# -topo_implementation consul
# -topo_global_server_address global_server:global_port
# -topo_global_root vitess/global
TOPOLOGY="-topo_implementation consul -topo_global_server_address global_server:global_port -topo_global_root vitess/global

Then to add a cell whose local topology server cell1_server1:cell1_port will store their data under the directory vitess/cell1:

# Reference cell1 in the global topology service:
# (the TOPOLOGY variable is defined in the previous section)
vtctl $TOPOLOGY AddCellInfo \
  -server_address cell1_server1:cell1_port \
  -root vitess/cell1 \
  cell1

If only one cell is used, the same consul instances can be used for both global and local data. A local cell record still needs to be created, just use the same server address and, very importantly, a different root node path.

Implementation details

For locks, we use a file named Lock in the directory to lock, and the regular Consul Lock API.

Master elections use a single lock file (the Election path) and the regular Consul Lock API. The contents of the lock file is the ID of the current master.

Watches use the Consul long polling Get call. They cannot be interrupted, so we use a long poll whose duration is set by the -topo_consul_watch_poll_duration flag. Canceling a watch may have to wait until the end of a polling cycle with that duration before returning.

Running in only one cell

The topology service is meant to be distributed across multiple cells, and survive single cell outages. However, one common usage is to run a Vitess cluster in only one cell / region. This part explains how to do this, and later on upgrade to multiple cells / regions.

If running in a single cell, the same topology service can be used for both global and local data. A local cell record still needs to be created, just use the same server address and, very importantly, a different root node path.

In that case, just running 3 servers for topology service quorum is probably sufficient. For instance, 3 etcd servers. And use their address for the local cell as well. Let's use a short cell name, like local, as the local data in that topology server will later on be moved to a different topology service, which will have the real cell name.

Extending to more cells

To then run in multiple cells, the current topology service needs to be split into a global instance and one local instance per cell. Whereas, the initial setup had 3 topology servers (used for global and local data), we recommend to run 5 global servers across all cells (for global topology data) and 3 local servers per cell (for per-cell topology data).

To migrate to such a setup, start by adding the 3 local servers in the second cell and run vtctl AddCellinfo as was done for the first cell. Tablets and vtgates can now be started in the second cell, and used normally.

vtgate can then be configured with a list of cells to watch for tablets using the -cells_to_watch command line parameter. It can then use all tablets in all cells to route traffic. Note this is necessary to access the master in another cell.

After the extension to two cells, the original topo service contains both the global topology data, and the first cell topology data. The more symetrical configuration we're after would be to split that original service into two: a global one that only contains the global data (spread across both cells), and a local one to the original cells. To achieve that split:

• Start up a new local topology service in that original cell (3 more local servers in that cell).
• Pick a name for that cell, different from local.
• Use vtctl AddCellInfo to configure it.
• Make sure all vtgates can see that new local cell (again, using -cells_to_watch).
• Restart all vttablets to be in that new cell, instead of the local cell name used before.
• Use vtctl RemoveKeyspaceCell to remove all mentions of the local cell in all keyspaces.
• Use vtctl RemoveCellInfo to remove the global configurations for that local cell.
• Remove all remaining data in the global topology service that are in the old local server root.

After this split, the configuration is completely symetrical:

• a global topology service, with servers in all cells. Only contains global topology data about Keyspaces, Shards and VSchema. Typically it has 5 servers across all cells.
• a local topology service to each cell, with servers only in that cell. Only contains local topology data about Tablets, and roll-ups of global data for efficient access. Typically, it has 3 servers in each cell.

Migration between implementations

We provide the topo2topo binary file to migrate between one implementation and another of the topology service.

The process to follow in that case is:

• Start from a stable topology, where no resharding or reparenting is on-going.
• Configure the new topology service so it has at least all the cells of the source topology service. Make sure it is running.
• Run the topo2topo program with the right flags. -from_implementation, -from_root, -from_server describe the source (old) topology service. -to_implementation, -to_root, -to_server describe the destination (new) topology service.
• Run vtctl RebuildKeyspaceGraph for each keyspace using the new topology service flags.
• Run vtctl RebuildVSchemaGraph using the new topology service flags.
• Restart all vtgate using the new topology service flags. They will see the same keyspaces / shards / tablets / vschema as before, as the topology was copied over.
• Restart all vttablet using the new topology service flags. They may use the same ports or not, but they will update the new topology when they start up, and be visible from vtgate.
• Restart all vtctld processes using the new topology service flags. So that the UI also shows the new data.

Sample commands to migrate from deprecated zookeeper to zk2 topology would be:

# Let's assume the zookeeper client config file is already
# exported in $ZK_CLIENT_CONFIG, and it contains a global record
# pointing to: global_server1,global_server2
# an a local cell cell1 pointing to cell1_server1,cell1_server2
#
# The existing directories created by Vitess are:
# /zk/global/vt/...
# /zk/cell1/vt/...
#
# The new zk2 implementation can use any root, so we will use:
# /vitess/global in the global topology service, and:
# /vitess/cell1 in the local topology service.

# Create the new topology service roots in global and local cell.
zk -server global_server1,global_server2 touch -p /vitess/global
zk -server cell1_server1,cell1_server2 touch -p /vitess/cell1

# Store the flags in a shell variable to simplify the example below.
TOPOLOGY="-topo_implementation zk2 -topo_global_server_address global_server1,global_server2 -topo_global_root /vitess/global"

# Reference cell1 in the global topology service:
vtctl $TOPOLOGY AddCellInfo \
  -server_address cell1_server1,cell1_server2 \
  -root /vitess/cell1 \
  cell1

# Now copy the topology. Note the old zookeeper implementation doesn't need
# any server or root parameter, as it reads ZK_CLIENT_CONFIG.
topo2topo \
  -from_implementation zookeeper \
  -to_implementation zk2 \
  -to_server global_server1,global_server2 \
  -to_root /vitess/global \

# Rebuild SvrKeyspace objects in new service, for each keyspace.
vtctl $TOPOLOGY RebuildKeyspaceGraph keyspace1
vtctl $TOPOLOGY RebuildKeyspaceGraph keyspace2

# Rebuild SrvVSchema objects in new service.
vtctl $TOPOLOGY RebuildVSchemaGraph

# Now restart all vtgate, vttablet, vtctld processes replacing:
# -topo_implementation zookeeper
# With:
# -topo_implementation zk2
# -topo_global_server_address global_server1,global_server2
# -topo_global_root /vitess/global
#
# After this, the ZK_CLIENT_CONF file and environment variables are not needed
# any more.

Migration using the Tee implementation

If your migration is more complex, or has special requirements, we also support a 'tee' implementation of the topo service interface. It is defined in go/vt/topo/helpers/tee.go. It allows communicating to two topo services, and the migration uses multiple phases:

• Start with the old topo service implementation we want to replace.
• Bring up the new topo service, with the same cells.
• Use topo2topo to copy the current data from the old to the new topo.
• Configure a Tee topo implementation to maintain both services.
  • Note we don't expose a plugin for this, so a small code change is necessary.
  • all updates will go to both services.
  • the primary topo service is the one we will get errors from, if any.
  • the secondary topo service is just kept in sync.
  • at first, use the old topo service as primary, and the new one as secondary.
  • then, change the configuration to use the new one as primary, and the old one as secondary. Reverse the lock order here.
  • then rollout a configuration to just use the new service.