lambdaapp

const (
	// KindTick is what an EventBridge Scheduler schedule sends, one per target.
	KindTick = "tick"
	// KindRun is what the http function sends when somebody asks for a pass now
	// (internal/port/invoke).
	KindRun = invoke.KindRun
)

The event kinds of a controller function. Both run the same pass; they differ in who asked.

const (
	// OutcomeRan is a pass that ran to its end.
	OutcomeRan = "ran"
	// OutcomeUnknown is a run-now of a target this controller does not run: the
	// notification was sent to every controller because nobody knew which kind
	// the target is, and the other one ends here, cleanly.
	OutcomeUnknown = "unknown"
	// OutcomeMaintenance is a pass that was skipped because the module is under
	// maintenance. A target another runner holds is [OutcomeContended].
	OutcomeMaintenance = "maintenance"
	// OutcomeContended is a target another invocation holds the lease of: the
	// pass is its, and this one ends cleanly. It is not a failure, so the
	// platform neither retries it nor counts it as an error.
	OutcomeContended = "contended"
)

The outcomes of a controller invocation.

const (
	ModuleIssuer     = "issuer"
	ModuleCloudflare = "cloudflare"
	ModuleBackup     = "backup"
	ModuleGitHub     = "github"
	ModuleSlack      = "slack"
	ModuleGoogle     = "google"
)

Module names: the role a function runs as. A zip is built for exactly one of them (cmd/sluis-<module>), and its document must be that module's.

const KindBackup = "backup"

KindBackup is the event an EventBridge Scheduler schedule sends the backup function: {"kind":"backup"} writes a backup in one invocation (format 2). {"kind":"backup","resume":true} was the schedule that continued a paused run of format 1; it still answers, idle, so a schedule left behind does not fail.

const KindCheck = "check"

KindCheck is the event a deploy sends the function after it exists, to prove that every secret the function's document and policy declare is in SSM: {"kind":"check"}. It reads no state and writes nothing; the answer is the deploycheck report (addresses and versions, never a value), and a missing or empty secret is an error of the invocation, naming the addresses, so the deploy fails.

const KindClients = "clients"

KindClients is the event an EventBridge Scheduler schedule sends the issuer function to look after the generated client secrets: {"kind":"clients"}. It reconciles the clients' secrets and nothing else; the directory's snapshots are the Google module's, refreshed by its own function.

Deprecated: KindRefresh is the same schedule under its old name, still accepted for one release.

const KindCloudflare = "cloudflare"

KindCloudflare is the event an EventBridge Scheduler schedule sends the `http` function to rotate the Cloudflare credentials that are due: {"kind":"cloudflare"}. A minute's schedule costs one read of the secrets store per preset until a rotation is due.

const KindRefresh = "refresh"

KindRefresh is the event an EventBridge Scheduler schedule sends the `http` function to take a new snapshot of every connected workspace's directory: {"kind":"refresh"}. The Kubernetes server does it on a ticker; there is no loop on Lambda, and a snapshot that is not refreshed stops being authoritative after the freshness window, which sign-in refuses on.

const KindRestore = "restore"

KindRestore is the event that drives the restore function (the backup zip with `backup.role: restore`, docs/decisions/0072): only an administrator or the break-glass role may invoke that function, by IAM. Invoked asynchronously (the platform's `Event` invocation type) it starts a restore and returns at once; the work goes on in this invocation and, when time runs out, in the next, which the function starts itself:

{"kind":"restore","backup":"<id>"}                    start a restore of the backup
{"kind":"restore","backup":"<id>","overwrite":true}   ... replacing different data in the destination
{"kind":"restore","backup":"<id>","preview":true}     what it would do; writes nothing (invoke it synchronously)
{"kind":"restore","resume":true}                      continue a restore that paused

"by" and "note" are the caller's own words for the status record. A restore that fails is not an error of the invocation, which the platform would retry: the outcome, `restore.status` and the audit trail say so, and every module stays under maintenance until a restore completes.