Connector state

Take a connector hold

Takes the connector's hold for process, or renews it if process already holds it. Only the process holding a live hold can replace the state document or write agreements.

POST
/connectors/{id}/hold

Authorization

bearerAuth
headerAuthorizationBearer <token>

Authenticate with an API key (marfa_k1_…) or an OAuth access token (marfa_at_…).

Path parameters

id*string

The ID of the connector.

Header parameters

X-Marfa-Instance-Id?string

The ID of the instance you expect, as GET / returns it. If another instance answers at this address, Marfa returns 409 instance_changed and doesn't write. Leave it out to send the request to whichever instance answers.

Match^[0-9a-f]{8}-[0-9a-f]{4}-7[0-9a-f]{3}-[89ab][0-9a-f]{3}-[0-9a-f]{12}$

Request body

application/json
  1. body
process*string

A name the process chose for itself, such as a UUID made at start. Marfa treats it as opaque.

Length1 <= length <= 100

Response body

About errors

Returns the hold. It lapses at expires_at unless the process renews it first.

application/json
  1. response
expires_at*string

When the hold lapses: the hold window after Marfa took or renewed it. The window is three minutes unless the instance sets another.

ttl_seconds*integer

The hold window in seconds, so you can schedule the next renewal without reading Marfa's clock.

renewed*boolean

true if this process's hold was still live when the call arrived; false on a first take or after a lapse. If it's false and you believed you held the connector, read the state and agreements again before writing.

  • missing_required_field: process is missing.
  • validation_error: process isn't 1 to 100 characters, or the body has a top-level field this endpoint doesn't take.
application/json
  1. response

An error response.

error*

What went wrong.

unauthorized: the request has no credential, or its credential is not valid. For a key that has expired, details.expired_at says when it expired.

application/json
  1. response

An error response.

error*

What went wrong.

  • forbidden: your credential isn't the connector's own key. Management permission does not authorize connector activity.
application/json
  1. response

An error response.

error*

What went wrong.

  • connector_not_found: no connector has this ID.
application/json
  1. response

An error response.

error*

What went wrong.

request_timeout: the request body didn't arrive whole within 180 seconds. Nothing changed, and Marfa closed the connection. Send the request again. A body larger than about 21 MiB can't arrive in that time over a 1 Mbit/s link, so send it in smaller requests.

application/json
  1. response

An error response.

error*

What went wrong.

  • connector_held: another process holds the connector until details.expires_at. The hold doesn't move.
application/json
  1. response

An error response.

error*

What went wrong.

contract_mismatch: X-Marfa-Contract names another contract version than this instance speaks, or is not a whole number. Nothing was done.

application/json
  1. response

An error response.

error*

What went wrong.

request_too_large: the request body is larger than this instance accepts.

application/json
  1. response

An error response.

error*

What went wrong.

rate_limited: you sent too many requests. Wait for the number of seconds in Retry-After, then try again.

application/json
  1. response

An error response.

error*

What went wrong.

internal_error: Marfa failed in a way it didn't expect, and the request may not have completed. Read what you changed before you repeat a write.

application/json
  1. response

An error response.

error*

What went wrong.

write_contention: the database was busy, and Marfa couldn't complete the request in time. Nothing changed. Try the request again. While a restore runs, details.reason is restore_running.

application/json
  1. response

An error response.

error*

What went wrong.

insufficient_storage: the disk that holds the instance's data has no room for the request, or the request would leave less free than the instance keeps in reserve. Nothing changed, unless details.write_outcome is unknown, which means the write may have landed: read what you changed before you repeat it. Free space on the disk, then try the request again.

application/json
  1. response

An error response.

error*

What went wrong.

curl -X POST 'http://localhost:8600/connectors/string/hold' \  -H 'Content-Type: application/json' \  -d '{  "process": "string"}'
About errors
{  "expires_at": "2026-10-03T10:25:00.150Z",  "ttl_seconds": 180,  "renewed": false}