The manifest now carries an authority block naming the api credential service, so the marketplace view shows the slot instead of relying on the description text. The client takes the full Authorization header: the api slot accepts a basic username/password or a bearer token, setting both is an error. Egress stays operator-supplied at accept time because the instance host comes from the instance_url setting, which a static egressHosts list cannot express.
ServiceNow connector
Records incidents and change requests on one ServiceNow instance, hears change approvals and rejections back as events, and answers the CMDB and change-calendar lookups. Built against a developer instance (PDI).
Each area of ServiceNow is its own file, so the file list is the map:
| File | Area | Tables |
|---|---|---|
client.go |
Table API client: instance URL, basic auth, JSON in and out | all |
incident.go |
Incident delivery and incident.updated event |
incident |
change.go |
Change request deliveries and change.approved / change.rejected events |
change_request |
cmdb.go |
Owner, environment and data-class lookups | cmdb_ci_server, cmdb_ci_appl, cmdb_ci |
calendar.go |
Blackouts lookup for the Timing guard | cmn_schedule_span (spans of schedule.type=blackout) |
listen.go |
Event polling, cursor over both tables | change_request, incident |
servicenow.go |
Wiring: config, Describe, delivery and lookup dispatch |
Configuration
Settings (see package.yaml):
instance_url(required):https://<instance>.service-now.com.poll_interval_seconds: how often events are polled. Defaults to 30.
Credentials: the api slot takes a basic credential (username and password,
resolved to api.username and api.password) or a bearer credential (a
token, resolved to api.token); setting both is an error. The slot is declared
in package.yaml under authority.credentialServices. Egress is confined to
the instance host, entered when the installation is accepted.
Deliveries
Deliveries are records, not judged actions: hard rules apply, guards don't. The host retries a failed delivery with the same idempotency key, so every delivery honours it.
incident— create an incident, or update the one named bynumber, optionally recordingnoteas the customer-visible comment. A create stores the idempotency key ascorrelation_id, so a retry finds the incident it already raised. Returns the incident number.change_request— raise a change request whosedescriptioncarries the findings, verdicts and exact actions the approver is approving. The idempotency key is thecorrelation_id. Returns the change number.change_request_withdraw— cancel the change the key raised, with the reason as a work note. Used when an approval expires.
Events
listen.go polls change_request and incident by sys_updated_on, bounded
above by the current time so future-dated sample rows cannot push the cursor
past the present. The cursor is chg:<sys_updated_on>|<sys_id>;inc:..., stored
per emitted batch, so a restart resumes after the last stored batch and an
in-place update (an approver's decision moves sys_updated_on, not sys_id)
is still caught.
change.approved,change.rejected— the change'sapprovalfield. Fields:number,approval,state,short_description.LinkKeyis the change number, so the answer links to the run that raised it.incident.updated— fields:number,state,short_description.
Lookups
All CMDB lookups take {"target": "<host or application name>"} and resolve
through cmdb_ci_server, then cmdb_ci_appl, then cmdb_ci. The whole CI row
is read with display values and the fields are picked locally, because which of
owner, delegate, environment, data_class exist depends on the
instance's CMDB schema. found=false means no CI matched, which core treats as
an unknown-owner finding.
cmdb_owner—owner,delegate,support_group, plussource(which CI answered).cmdb_environment—valuefrom the CI'senvironment(oru_environment).cmdb_data_class—valuefrom the CI'sdata_class(oru_data_class).blackouts— takes{"at": "<RFC 3339>"}, empty for now; returns the active blackout spans as{name, schedule, start, end}, read fromcmn_schedule_spanwhereschedule.type=blackoutand the span covers the time. The Timing guard reads this as a freeze.
Done when (archie #1983)
Archie raises a change request, approving it on the instance continues the run, rejecting it drops those actions, the owner asked is the CMDB application owner, and a blackout in the change calendar reaches the Timing guard.