fix: coordinate SMS completion and improve operations diagnostics

api/prisma/migrations/20260916040000_persist_infrastructure_alerts/migration.sql

@@ -0,0 +1,17 @@
+CREATE TABLE "InfrastructureAlertCollection" (
+  "id" TEXT NOT NULL PRIMARY KEY,
+  "observedAt" TIMESTAMP(3) NOT NULL
+);
+CREATE TABLE "InfrastructureAlertEvent" (
+  "id" TEXT NOT NULL PRIMARY KEY,
+  "fingerprint" TEXT NOT NULL,
+  "activeAt" TIMESTAMP(3) NOT NULL,
+  "payload" JSONB NOT NULL,
+  "lastObservedAt" TIMESTAMP(3) NOT NULL,
+  "recoveredAt" TIMESTAMP(3),
+  "clearedAt" TIMESTAMP(3),
+  "clearedBy" TEXT,
+  "createdAt" TIMESTAMP(3) NOT NULL DEFAULT CURRENT_TIMESTAMP
+);
+CREATE UNIQUE INDEX "InfrastructureAlertEvent_fingerprint_activeAt_key" ON "InfrastructureAlertEvent"("fingerprint", "activeAt");
+CREATE INDEX "InfrastructureAlertEvent_clearedAt_activeAt_idx" ON "InfrastructureAlertEvent"("clearedAt", "activeAt");

api/prisma/migrations/20260916041000_attempt_completion_work/migration.sql

@@ -0,0 +1,15 @@
+CREATE TABLE "SmsAttemptCompletionWork" (
+"id" TEXT PRIMARY KEY, "workKey" TEXT NOT NULL, "tenantId" TEXT, "messageRecordId" TEXT NOT NULL, "sourceSubmitRecordId" TEXT,
+"revision" INTEGER NOT NULL DEFAULT 0, "processedRevision" INTEGER NOT NULL DEFAULT 0, "state" TEXT NOT NULL DEFAULT 'pending',
+"leaseOwner" TEXT, "leaseUntil" TIMESTAMP(3), "fenceVersion" INTEGER NOT NULL DEFAULT 0, "attempts" INTEGER NOT NULL DEFAULT 0,
+"nextAttemptAt" TIMESTAMP(3) NOT NULL DEFAULT (CURRENT_TIMESTAMP AT TIME ZONE 'UTC'), "decision" TEXT, "retrySubmitRecordId" TEXT, "lastError" TEXT,
+"createdAt" TIMESTAMP(3) NOT NULL DEFAULT (CURRENT_TIMESTAMP AT TIME ZONE 'UTC'), "updatedAt" TIMESTAMP(3) NOT NULL DEFAULT (CURRENT_TIMESTAMP AT TIME ZONE 'UTC'));
+CREATE UNIQUE INDEX "SmsAttemptCompletionWork_workKey_key" ON "SmsAttemptCompletionWork"("workKey");
+CREATE UNIQUE INDEX "SmsAttemptCompletionWork_sourceSubmitRecordId_key" ON "SmsAttemptCompletionWork"("sourceSubmitRecordId");
+CREATE INDEX "SmsAttemptCompletionWork_state_nextAttemptAt_idx" ON "SmsAttemptCompletionWork"("state", "nextAttemptAt");
+CREATE INDEX "SmsAttemptCompletionWork_state_leaseUntil_idx" ON "SmsAttemptCompletionWork"("state", "leaseUntil");
+CREATE INDEX "SmsAttemptCompletionWork_messageRecordId_idx" ON "SmsAttemptCompletionWork"("messageRecordId");
+CREATE TABLE "SmsCompletionEvent" ("id" TEXT PRIMARY KEY, "eventKey" TEXT NOT NULL, "workId" TEXT NOT NULL REFERENCES "SmsAttemptCompletionWork"("id") ON DELETE RESTRICT,
+"kind" TEXT NOT NULL, "payload" JSONB NOT NULL, "processedAt" TIMESTAMP(3), "createdAt" TIMESTAMP(3) NOT NULL DEFAULT (CURRENT_TIMESTAMP AT TIME ZONE 'UTC'));
+CREATE UNIQUE INDEX "SmsCompletionEvent_eventKey_key" ON "SmsCompletionEvent"("eventKey");
+CREATE INDEX "SmsCompletionEvent_workId_processedAt_createdAt_idx" ON "SmsCompletionEvent"("workId", "processedAt", "createdAt");

api/prisma/schema.prisma

@@ -2840,3 +2840,59 @@ model OpenApiDispatchOutbox {
   updatedAt DateTime @updatedAt
   @@index([status, leaseUntil])
 }
+
+model InfrastructureAlertCollection {
+  id String @id
+  observedAt DateTime
+}
+
+model InfrastructureAlertEvent {
+  id String @id @default(cuid())
+  fingerprint String
+  activeAt DateTime
+  payload Json
+  lastObservedAt DateTime
+  recoveredAt DateTime?
+  clearedAt DateTime?
+  clearedBy String?
+  createdAt DateTime @default(now())
+  @@unique([fingerprint, activeAt])
+  @@index([clearedAt, activeAt])
+}
+
+model SmsAttemptCompletionWork {
+  id String @id @default(cuid())
+  workKey String @unique
+  tenantId String?
+  messageRecordId String
+  sourceSubmitRecordId String? @unique
+  revision Int @default(0)
+  processedRevision Int @default(0)
+  state String @default("pending")
+  leaseOwner String?
+  leaseUntil DateTime?
+  fenceVersion Int @default(0)
+  attempts Int @default(0)
+  nextAttemptAt DateTime @default(dbgenerated("(CURRENT_TIMESTAMP AT TIME ZONE 'UTC')"))
+  decision String?
+  retrySubmitRecordId String?
+  lastError String?
+  createdAt DateTime @default(dbgenerated("(CURRENT_TIMESTAMP AT TIME ZONE 'UTC')"))
+  updatedAt DateTime @updatedAt
+  events SmsCompletionEvent[]
+  @@index([state, nextAttemptAt])
+  @@index([state, leaseUntil])
+  @@index([messageRecordId])
+}
+
+model SmsCompletionEvent {
+  id String @id @default(cuid())
+  eventKey String @unique
+  workId String
+  kind String
+  payload Json
+  processedAt DateTime?
+  createdAt DateTime @default(dbgenerated("(CURRENT_TIMESTAMP AT TIME ZONE 'UTC')"))
+  work SmsAttemptCompletionWork @relation(fields: [workId], references: [id], onDelete: Restrict)
+  @@index([workId, processedAt, createdAt])
+}