---
title: "startup in config.yaml"
url: https://devlanes.run/docs/reference/config/user/startup/
description: "Reference for the startup key in ~/.devlanes/config.yaml."
---

# startup in config.yaml

A top-level key of [`~/.devlanes/config.yaml`](/docs/reference/config/user/).

## `startup`

Type: mapping.

How many services may be starting at the same time across all lanes and projects, so a large start does not overload the machine. Defaults to one start per CPU core.

```yaml devlanes-user
startup:
  max: 4
  mode: fixed
```

### `startup.initial`

Type: integer.

Adaptive starting window; defaults to `max`, so starts run at full speed until the machine is under pressure. Set it lower to ramp up from a few slots instead; never above `max`.

```yaml devlanes-user
startup:
  initial: 2
```

### `startup.max`

Type: integer.

Most services starting at once: the constant window in `fixed` mode and the ceiling in `adaptive` mode. Defaults to the number of CPU cores; `all` ignores it.

```yaml devlanes-user
startup:
  max: 8
```

### `startup.mode`

Type: string. Default: `adaptive`.

`adaptive`, `fixed` or `all`; default `adaptive`. A starting service holds its slot until its ready check (`tcp`, `http`, `log` or `cmd`) passes, it fails, or `slot_timeout` passes; one without a check (`ready: started`) holds it until it settles (`settle_cpu`).

Values:

- `adaptive`: Begin each burst of starts at `initial` (by default `max`, full speed). Halve the slots after 3 s above 90% CPU or when the system reports memory pressure, start nothing while that pressure is critical, and add a slot every 2 s while system CPU stays below `target_cpu`, up to `max`.
- `fixed`: Allow `max` starts at a time, never more or fewer. A start also waits while free memory cannot fit what it needs.
- `all`: Start every service immediately, with no limit and no memory check.

```yaml devlanes-user
startup:
  mode: adaptive
```

### `startup.reserve_memory`

Type: [byte size](/docs/reference/config/#byte-sizes).

Free memory every new start must leave, on top of the memory the services already starting are expected to still need (each one's past startup peak, or 512 MiB without history). Defaults to 10% of RAM, at most `2GiB`; `0B` turns it off. A start with nothing else starting never waits for it. `all` ignores it.

```yaml devlanes-user
startup:
  reserve_memory: 2GiB
```

### `startup.settle_cpu`

Type: [percentage](/docs/reference/config/#percentages). Default: `10%`.

Only for services without a ready check (`ready: started`, the default for a service without `hosts`): such a service gives its slot back once its process tree uses less than this share of one core for 2 s, meaning it has finished starting up. Services with a `tcp`, `http`, `log` or `cmd` check keep their slot until the check passes instead. Defaults to `10%`.

```yaml devlanes-user
startup:
  settle_cpu: 10%
```

### `startup.slot_timeout`

Type: [duration](/docs/reference/config/#durations). Default: `20s`.

Longest a starting service holds its slot; after it, the next queued service may start even if this one is not ready, so services waiting on each other cannot block the queue. Defaults to `20s` and must be above zero; the service itself keeps starting.

```yaml devlanes-user
startup:
  slot_timeout: 15s
```

### `startup.target_cpu`

Type: [percentage](/docs/reference/config/#percentages). Default: `75%`.

System CPU use below which `adaptive` mode adds slots; defaults to `75%`, and must be above 0% and at most 100%. Lower it to keep the machine responsive during large starts; other modes ignore it.

```yaml devlanes-user
startup:
  target_cpu: 75%
```
