# Trains ## The list `/trains`. Every `IServiceTrain` the host registers, from [`operations.trains`](/docs/sdk-reference/graphql-api/queries/operations#trains), sorted by any column. | Column | Value | |---|---| | **Train** | Its short name, linking its page; the full name on hover | | **Kind** | Query or Mutation, when the train is exposed through GraphQL | | **Lifetime** | Transient, Scoped or Singleton | | **Roles** | The roles its `[TraxAuthorize]` requires | The filter (`q` in the address) matches part of the full name. **Hide scheduler trains** leaves out the scheduler's own trains. Each row has **Queue** and **Run**. ## A train `/trains/{full name}`. Its full name under the title, **Queue** and **Run**, and a **Range** of the last 24 hours or the last 7 days (`range` in the address) for the parts that read a range. | Part | Shows | |---|---| | Summary | Total runs, success rate (amber under 90%), failed, running, average duration, last success | | **Chain** | The train's declared chain drawn as a graph. Each step carries a red count of the failed runs in the range that stopped there; under it, how many failed in the range, and how many stopped outside a drawn step. "At least" when there were more than 500. | | **Recent runs** | A grid of the newest 20 runs, oldest on the left: the top row is each run's state, and a row per step gives that step's state in that run. Every cell opens its run. **All runs of this train** opens [Runs](/docs/dashboard/runs) filtered to it. | | **Manifests** | Up to 10 manifests that run this train: each one's schedule in words and its next run, or Disabled, or "No run due" | | **Started by machines** | Each state machine state that starts this train on entering it, linking that machine's instances in the state | | **Runs over time** | Completed, failed and cancelled runs over the range, with their totals and the slowest bucket's 95th percentile duration | A train that has never run says so and offers **Run now** and **Queue a run**. A train this host does not register, or whose chain it cannot read, says so in place of the chain. ## Run and queue Both open a dialog with a **Train** picker, one field per public property of the input type (a drop-down for an enum, a checkbox for a boolean, text otherwise), and an **Input JSON** box that overrides the fields; blank is `Unit`. | Button | Calls | Then | |---|---|---| | **Run train** | `runTrain`: runs now through the host's job submitter, with no work queue entry | Opens the new run | | **Queue train** | `queueTrain`, with a **Priority** of 0 to 31 | Opens the new queue entry | A refusal (an unknown train, an input that does not read as the input type, a refusal the train's queue check makes) keeps the dialog open with the API's reason. For a train that sets a subject key, the Queue dialog says the entry waits while another for the same subject is ahead of it or running. The Run dialog warns that Run skips the queue; the API refuses to run such a train for a caller outside a trusted scope, which includes the console, so queue it instead. Why is in [Where Trax checks authorization](/docs/security/where-authorization-is-checked#the-operations-surface-and-the-dashboard). The caller needs the operations gate, and the train's own `[TraxAuthorize]` requirements apply on top of it: passing the gate makes someone an operator, not a holder of every train's role. The same holds for Requeue and Resume on [a run's page](/docs/dashboard/a-run).