Backfill subject last_seen_at from connection activity

apps/cloud/scripts/backfill-subjects.ts

@@ -14,13 +14,17 @@
 //
 //   created_at   ← the tenant+subject's oldest connection, the closest honest
 //                  answer to "when did this user first appear".
-//   last_seen_at ← NULL. The column means "seen on a request", no request has
-//                  been recorded, and the console renders NULL as "Never" —
-//                  which is true since the sighting code deployed.
+//   last_seen_at ← the tenant+subject's newest connection `updated_at`, the
+//                  latest activity storage can attest to. Rendering every
+//                  pre-0012 user as "Never" was literally true of the sighting
+//                  column but reads as "inactive user" to an operator, which
+//                  is the wrong answer for a tenant full of active people. The
+//                  first real sighting overwrites this with request truth.
 //
-// Idempotent — ON CONFLICT on the (tenant, external_id) unique index leaves
-// rows the runtime has since minted (or a prior run inserted) untouched, so
-// re-running is safe and never rewinds a live row's fields.
+// Idempotent — the insert's ON CONFLICT on the (tenant, external_id) unique
+// index leaves rows the runtime has since minted (or a prior run inserted)
+// untouched, and the last_seen_at repair only fills NULLs — so re-running is
+// safe and never rewinds a live row's fields.
 // Pass --dry-run to print the plan without writing.
 // ---------------------------------------------------------------------------
 
@@ -37,8 +41,12 @@ if (!connectionString) {
 
 const sql = postgres(connectionString, { max: 1, prepare: false, ssl: "require" });
 
-const rows = await sql<{ tenant: string; subject: string; first_connection_at: Date }[]>`
-  SELECT tenant, subject, MIN(created_at) AS first_connection_at
+const rows = await sql<
+  { tenant: string; subject: string; first_connection_at: Date; last_activity_at: Date }[]
+>`
+  SELECT tenant, subject,
+         MIN(created_at) AS first_connection_at,
+         MAX(updated_at) AS last_activity_at
   FROM connection
   WHERE subject <> ''
   GROUP BY tenant, subject
@@ -47,20 +55,33 @@ const rows = await sql<{ tenant: string; subject: string; first_connection_at: D
 console.log(`${rows.length} (tenant, subject) pair(s) with connections`);
 
 let inserted = 0;
+let repaired = 0;
 for (const row of rows) {
   console.log(
-    `${row.tenant}  ${row.subject}  first connection ${row.first_connection_at.toISOString()}`,
+    `${row.tenant}  ${row.subject}  first connection ${row.first_connection_at.toISOString()}  last activity ${row.last_activity_at.toISOString()}`,
   );
   if (dryRun) continue;
+  const lastSeenMs = row.last_activity_at.getTime();
   const result = await sql`
     INSERT INTO subject (row_id, tenant, external_id, created_at, last_seen_at, status)
-    VALUES (${createId()}, ${row.tenant}, ${row.subject}, ${row.first_connection_at}, NULL, NULL)
+    VALUES (${createId()}, ${row.tenant}, ${row.subject}, ${row.first_connection_at}, ${lastSeenMs}, NULL)
     ON CONFLICT (tenant, external_id) DO NOTHING
   `;
   inserted += result.count;
+  if (result.count > 0) continue;
+  // The row predates this run but may still carry NULL from a run of the
+  // pre-repair version of this script. Fill it in; a sighted row keeps its
+  // request-truth timestamp.
+  const repair = await sql`
+    UPDATE subject SET last_seen_at = ${lastSeenMs}
+    WHERE tenant = ${row.tenant} AND external_id = ${row.subject} AND last_seen_at IS NULL
+  `;
+  repaired += repair.count;
 }
 
 console.log(
-  dryRun ? `dry run — would insert up to ${rows.length} row(s)` : `inserted ${inserted} row(s)`,
+  dryRun
+    ? `dry run — would insert up to ${rows.length} row(s)`
+    : `inserted ${inserted} row(s), repaired last_seen_at on ${repaired} row(s)`,
 );
 await sql.end();