start

start#

Start the openYuanrong cluster, supporting master and agent modes.

Usage#

yr start [OPTIONS]

Parameters#

  • --master: Start in master mode (defaults to agent mode without this parameter).

  • -s, --set KEY=VALUE: Override configuration from the command line; can be specified multiple times. VALUE must be a valid TOML literal.

  • --master_address http(s)://host:port: Only available in agent mode. Before startup, pulls service discovery information from the specified function_master and automatically converts it to -s overrides.

  • --log-dir-prefix <path>: Prefix directory for session and log output (default /tmp/yr_sessions). The session dir <prefix>/<timestamp>/, latest symlink, session.json, yr_current_master_info, and component logs <deploy_path>/logs/ are all created under this prefix, migrating the whole output tree together without splitting paths.

Note

Common -s/--set examples:

  • String: -s 'values.fs.log.level="DEBUG"'

  • Number: -s 'values.ds_master.port=12123'

  • Array of tables: -s 'values.etcd.address=[{ip="10.88.0.9",peer_port="32380",port="32379"}]'

--master_address usage restrictions:

  • Only supported by yr start (agent mode).

  • Using it together with --master will result in an error and exit.

  • If service discovery fails, it will exit directly.

  • The parameter value must include a protocol prefix, i.e., http:// or https://.

  • When using https://, you need to provide values.fs.tls.ca_file, values.fs.tls.cert_file, values.fs.tls.key_file via --config or -s (can be used together with values.fs.tls.base_path).

Configuration Merge Rules#

When yr start starts, the final configuration is merged according to the following priority (high to low):

  1. Command-line overrides: -s/--set

  2. User configuration file: defaults to /etc/yuanrong/config.toml (can be specified via -c)

  3. Built-in defaults: api/python/yr/cli/values.toml + api/python/yr/cli/config.toml.jinja

Startup Behavior#

After executing yr start, the CLI will:

  1. Create a deployment directory <prefix>/<timestamp>/ (<prefix> defaults to /tmp/yr_sessions, overridable via --log-dir-prefix)

  2. Update the symbolic link <prefix>/latest -> <prefix>/<timestamp>/

  3. Start the components corresponding to the mode and perform health checks

  4. Write the session file <prefix>/latest/session.json

Where:

  • master starts by default: etcd, ds_master, ds_worker, function_master, function_proxy, function_agent

  • agent starts by default: ds_worker, function_proxy, function_agent

Examples#

Start the master node:

yr start --master

After successful startup, the terminal will print the recommended command for worker nodes to join the cluster (including -s 'values.*=...' overrides).

On worker nodes, you can directly execute this command to join the cluster, for example:

yr start -s 'values.etcd.address=[{ip="10.88.0.4",peer_port="32380",port="32379"}]' \
  -s 'values.ds_master.ip="10.88.0.4"' \
  -s 'values.ds_master.port="12123"' \
  -s 'values.function_master.ip="10.88.0.4"' \
  -s 'values.function_master.global_scheduler_port="22770"'

Or use the auto-discovery mode (agent):

yr start --master_address http://10.88.0.4:22770

Specify a log output prefix (migrate all sessions and logs under /data/yr_logs):

yr start --master --log-dir-prefix /data/yr_logs

Note

--log-dir-prefix only replaces the prefix; the <timestamp>/ subdirectory, logs/, session.json, etc. beneath it keep their structure. Repeated starts accumulate multiple timestamped directories under the same prefix, and the latest symlink always points to the most recent one.