gpc train

Config-driven staged rollout pipeline. The release train automates multi-stage rollout progression with configurable delays and vitals gates.

Synopsis

gpc train <subcommand>

Config File

Create .gpcrc-train.json in your project root:

{
  "stages": [
    { "track": "internal", "rollout": 100 },
    { "track": "alpha", "rollout": 100, "after": "2d" },
    { "track": "production", "rollout": 10, "after": "7d" },
    { "track": "production", "rollout": 100, "after": "14d" }
  ],
  "gates": {
    "crashes": { "max": 1.5 },
    "anr": { "max": 0.5 }
  }
}

Stage fields:

Field	Description
track	Target track name (internal, alpha, beta, production)
rollout	Rollout percentage (1–100)
after	Minimum delay before this stage runs: "2d" (days), "4h" (hours), "30m" (minutes)

Gate fields:

Gate	Description
crashes.max	Maximum crash rate % (e.g. 1.5 = 1.5%)
anr.max	Maximum ANR rate %

Train gates are percentages, unlike gpc vitals crashes --threshold and gpc watch, which take fractions (0.015 = 1.5%).

Gates read the worst value of the most recent day of vitals. If a gate fails during advance, the train is automatically paused and the command exits 6 (TRAIN_CRASH_GATE_FAILED or TRAIN_ANR_GATE_FAILED). If a gate has no data yet (vitals lag 2 to 4 days), the train stays on its current stage and exits 6 with TRAIN_GATE_NO_DATA; run advance again later.

Subcommands

train start

Start a release train from a config file.

gpc train start --config .gpcrc-train.json
gpc train start --force  # restart a running train

Stage 0 executes immediately on start.

train status

Show the current state of the release train.

gpc train status
gpc train status --output json

train advance

Advance to the next stage (after checking delay and gates). Nothing advances on its own: run this on a schedule, for example a CI cron job.

gpc train advance

GPC checks:

1. Has the after delay elapsed since the previous stage?
2. Do vitals gates pass?

If both pass, the next stage is executed.

train pause

Pause a running train. advance will not move it. Pausing does not change any rollout on Google Play. There is no resume command: restart with gpc train start --force, which runs stage 0 again.

gpc train pause

train abort

Abort and clear the train state. Prompts for confirmation. Aborting does not halt the rollout; to stop a release, halt it:

gpc train abort
gpc releases rollout halt --track production

State File

Train state is persisted to ~/.cache/gpc/train-<packageName>.json. This allows you to resume or inspect a train across terminal sessions.

Example Workflow

# 1. Start the train (pushes to internal immediately)
gpc train start

# 2. After 2 days, advance to alpha
gpc train advance

# 3. After 7 days, advance to production at 10%
gpc train advance

# 4. After 14 days, complete rollout
gpc train advance

See Also

• gpc releases rollout — manual rollout control
• gpc vitals — monitor vitals that power the gates
• gpc watch — real-time rollout monitoring with threshold alerts