MySQL replication topology manager - agent (daemon)
Go Shell
Switch branches/tags
Clone or download
Latest commit 61af2a9 Oct 28, 2016


MySQL topology agent (daemon)

orchestrator-agent is a sub-project of orchestrator. It is a service that runs on MySQL hosts and communicates with orchestrator.

orchestrator-agent is capable of proving operating system, file system and LVM information to orchestrator, as well as invoke certain commands and scripts.

The primary drive for developing orchestrator-agent was Outbrain's need for a controlled seeding mechanism, where MySQL instances would duplicate onto new/corrupt installations. As such, orchestrator-agent's functionality is tightly coupled with the backup/seed mechanisms used by Outbrain, as described following. Though easily extendible, orchestrator-agent was not developed with a general purpose backup/restore/orchestration capabilities in mind.

Whether or not orchestrator-agent is useful to you depends on your needs.

Generic functionality offered by orchestrator-agent:
  • Detection of the MySQL service, starting and stopping (start/stop/status commands provided via configuration)
  • Detection of MySQL port, data directory (assumes configuration is /etc/my.cnf)
  • Calculation of disk usage on data directory mount point
  • Tailing the error log file
  • Discovery (the mere existence of the orchestrator-agent service on a host may suggest the existence or need of existence of a MySQL service)
Specialized functionality offered by orchestrator-agent:
  • Detection of LVM snapshots on MySQL host (snapshots that are MySQL specific)
  • Creation of new snapshots
  • Mounting/umounting of LVM snapshots
  • Detection of DC-local and DC-agnostic snapshots available for a given cluster
  • Transmitting/receiving seed data

The Outbrain seed method

The following does not discuss hard-backups where binary/logical data is written to an external/remote disk or a tape.

At Outbrain we use LVM snapshots. Each MySQL replication topology has designated slaves which serve as snapshot servers. These servers do LVM snapshots daily, and keep such snapshots open for a few days. Thus it is possible that a server has, say, 5 open (and unmounted) LVM snapshots at a given time.

Upon need, we are able to mount any such snapshot in near zero time and restart MySQL on the host using mounted data directory. We are thus able to recover any one of few days back at speed of InnoDB crash-recovery.

This serves two purposes: an immediate recovery/sanity check for destructive operations (unintentional or malicious UPDATE or DROP) as well as a seed source for new/corrupt servers.

The choice of snapshot slaves is not random; we have multiple data centers and we keep at least one snapshot server per topology per data center, such that upon need we can make DC-local copies. For this purpose, the daily snapshot process reports, upon success, the availability of the snapshot along with any metadata required on cluster/DC.

orchestrator-agent thus depends on external (configurable) commands to:

  • Detect where in local and remote DCs it can find an appropriate snapshot
  • Find said snapshot on server, mount it
  • Stop MySQL on target host, clear data on MySQL data directory
  • Initiate send/receive process
  • Cleanup data after copy (e.g. remove .pid files if any)
  • Unmount snapshot
  • etc.

The orchestrator & orchestrator-agent architecture

orchestrator is a standalone, centralized service/command line tool. When acting as a service, it provides with web API and web interface to allow replication topology refactoring, long query control, and more.

Coupled with orchestrator-agent, orchestrator is further able to assist in seeding new/corrupt servers. orchestrator-agent does not initiate anything by itself, but is in fact controlled by orchestrator.

When started, orchestrator-agent chooses a random, secret token and attempts to connect to the centralized orchestrator service API (location configurable). It then registers at the orchestrator service with its secret token.

orchestrator-agent then serves via HTTP API, and for all but the simplest commands requires the secret token.

At this point orchestrator becomes the major player; having multiple orchestrator-agent registered it is able to coordinate operations such as snapshot mounting, space cleanup, send and receive so as to establish a successful seed (a binary copy of a MySQL data directory).

orchestrator-agent only provides the minimal and required operating system functionality and does not interact with the MySQL service directly (i.e. no credentials required and no SQL queries invoked). Any and all queries are invoked by the centralized orchestrator service.


Orchestrator-agent uses a configuration file, located in either /etc/orchestrator-agent.conf.json or relative path to binary conf/orchestrator-agent.conf.json or orchestrator-agent.conf.json.

Note that the agent will use the config file in its relative conf path first.

The following is a complete list of configuration parameters:

  • SnapshotMountPoint (string), a known mountpoint onto which a mount command will mount snapshot volumes
  • ContinuousPollSeconds (uint), internal clocking interval (default 60 seconds)
  • ResubmitAgentIntervalMinutes (uint), interval at which the agent re-submits itself to orchestrator daemon
  • CreateSnapshotCommand (string), command which creates new LVM snapshot of MySQL data
  • AvailableLocalSnapshotHostsCommand (string), command which returns list of hosts in local DC on which recent snapshots are available
  • AvailableSnapshotHostsCommand (string), command which returns list of hosts in all DCs on which recent snapshots are available
  • SnapshotVolumesFilter (string), free text which identifies MySQL data snapshots (as opposed to other, unrelated snapshots)
  • MySQLDatadirCommand (string), command which returns the data directory (e.g. grep datadir /etc/my.cnf | head -n 1 | awk -F= '{print $2}')
  • MySQLPortCommand (string), command which returns the MySQL port
  • MySQLDeleteDatadirContentCommand (string), command which purges the MySQL data directory
  • MySQLServiceStopCommand (string), command which stops the MySQL service (e.g. service mysql stop)
  • MySQLServiceStartCommand (string), command which starts the MySQL service
  • MySQLServiceStatusCommand (string), command that checks status of service (expecting exit code 1 when service is down)
  • ReceiveSeedDataCommand (string), command which listen on data, must accept arguments: target directory, listen port
  • SendSeedDataCommand (string), command which sends data, must accept arguments: source directory, target host, target port
  • PostCopyCommand (string), command to be executed after the seed is complete (cleanup)
  • AgentsServer (string), Required URL of your orchestrator daemon, You must add the port the orchestrator server expects to talk to agents to (see below, e.g. https://my.orchestrator.daemon:3001)
  • HTTPPort (uint), Port to listen on
  • HTTPAuthUser (string), Basic auth user (default empty, meaning no auth)
  • HTTPAuthPassword (string), Basic auth password
  • UseSSL (bool), If true then serving via https protocol
  • SSLSkipVerify (bool), When connecting to orchestrator via SSL, whether to ignore certification error
  • SSLPrivateKeyFile (string), When serving via https, location of SSL private key file
  • SSLCertFile (string), When serving via https, location of SSL certification file
  • HttpTimeoutSeconds (int), HTTP GET request timeout (when connecting to orchestrator)

An example configuration file may be:

    "SnapshotMountPoint": "/var/tmp/mysql-mount",
    "AgentsServer": "https://my.orchestrator.daemon:3001",
    "ContinuousPollSeconds" : 60,
    "ResubmitAgentIntervalMinutes": 60,
    "CreateSnapshotCommand":                "/path/to/snapshot-command.bash",
    "AvailableLocalSnapshotHostsCommand":   "/path/to/snapshot-local-availability-command.bash",
    "AvailableSnapshotHostsCommand":        "/path/to/snapshot-availability-command.bash",
    "SnapshotVolumesFilter":                "mysql-snap",
    "MySQLDatadirCommand":                  "set $(grep datadir /etc/my.cnf | head -n 1 | awk -F= '{print $2}') ; echo $1",
    "MySQLPortCommand":                     "set $(grep ^port /etc/my.cnf | head -n 1 | awk -F= '{print $2}') ; echo $1",
    "MySQLDeleteDatadirContentCommand":     "set $(grep datadir /etc/my.cnf | head -n 1 | awk -F= '{print $2}') ; rm --preserve-root -rf $1/*",
    "MySQLServiceStopCommand":      "/etc/init.d/mysql stop",
    "MySQLServiceStartCommand":     "/etc/init.d/mysql start",
    "MySQLServiceStatusCommand":    "/etc/init.d/mysql status",
    "ReceiveSeedDataCommand":       "/path/to/data-receive.bash",
    "SendSeedDataCommand":          "/path/to/data-send.bash",
    "PostCopyCommand":              "set $(grep datadir /etc/my.cnf | head -n 1 | awk -F= '{print $2}') ; rm -f $1/*.pid",
    "HTTPPort": 3002,
    "HTTPAuthUser": "",
    "HTTPAuthPassword": "",
    "UseSSL": false,
    "SSLSkipVerify": false,
    "SSLCertFile": "",
    "SSLPrivateKeyFile": "",
    "HttpTimeoutSeconds": 10

Necessary matching configuration on the Orchestrator Server side

If you initially deployed orchestrator with a minimally working configuration, you will need to make some changes on the server side to prepare it for newly deployed agents. The configuration lines needed on the server side to support agents are

  • ServeAgentsHttp (bool), Must be set to true to get the orchestrator server listening for agents
  • AgentsServerPort (String), The port on which the server should listen to agents. Shoult match the port you define for agents in AgentsServer.


  • Linux, 64bit. Tested on CentOS 5 and Ubuntu Server 12.04+
  • MySQL 5.1+
  • LVM, free space in volume group, if snapshot functionality is required
  • orchestrator-agent assumes a single MySQL running on the machine

Extending orchestrator-agent

Yes please. orchestrator-agent is open to pull-requests. Desired functionality is for example the initiation and immediate transfer of backup data via xtrabackup. The same can be done via mysqldump or mydumper etc.

Authored by Shlomi Noach at Outbrain