From d5d7d3361799787b9c570994276a3e0faa5e880d Mon Sep 17 00:00:00 2001
From: prayagv <76861333+prayagv@users.noreply.github.com>
Date: Thu, 22 Oct 2026 17:56:27 +0530
Subject: [PATCH] I let the drill read a full Grid slot

---
 docs/verification/grid-worker-recovery.md |  2 +-
 scripts/grid-recovery-drill.mjs           |  5 +++--
 scripts/grid-recovery.test.mjs            | 16 +++++++++++++---
 3 files changed, 17 insertions(+), 6 deletions(-)

diff --git a/docs/verification/grid-worker-recovery.md b/docs/verification/grid-worker-recovery.md
index cf9fe75..3bf6abc 100644
--- a/docs/verification/grid-worker-recovery.md
+++ b/docs/verification/grid-worker-recovery.md
@@ -2,7 +2,7 @@
 
 This is an opt-in hosted experiment. Run **Grid Worker Recovery Drill** with GitHub Actions `workflow_dispatch`; it is not part of the regular browser matrix and it does not target a shared or remote Grid. The manual workflow creates a unique Compose project from the pinned Selenium Hub and Chrome image digests, uses one Chrome slot, checks the effective 2 GiB memory and shared-memory limits, and has an eight-minute job timeout with project-scoped teardown.
 
-The drill first validates the Grid `/status` slot inventory and confirms it starts empty. It sends one `quality-drill-unsupported` request with the hub's five-second session queue timeout and one-second retry interval. Only a structured session-creation failure plus a validated inventory without that browser is `CAPABILITY_UNAVAILABLE`; connection trouble is `GRID_UNREACHABLE`.
+The drill first validates the Grid `/status` slot inventory and confirms it starts empty. Grid's `ready` flag describes available capacity: a full one-slot Grid reports `ready: false` while its owned session is running. The harness requires readiness during initial startup, then keeps reading the validated inventory even while that slot is occupied. It sends one `quality-drill-unsupported` request with the hub's five-second session queue timeout and one-second retry interval. Only a structured session-creation failure plus a validated inventory without that browser is `CAPABILITY_UNAVAILABLE`; connection trouble is `GRID_UNREACHABLE`.
 
 For the interruption, the TestNG worker creates one Chrome session through `WebDriverFactory`, visits a harmless local `data:` page, and atomically writes a private receipt before printing its readiness marker. The receipt directory is mode `0700`; the receipt is mode `0600`, and neither is uploaded. The Node harness independently matches the receipt's exact session in the validated Grid inventory, then sends `SIGKILL` only to the detached process group it spawned. It requires the orphan to remain visible before issuing one W3C DELETE for that receipt-authorized session. It rejects stale, mismatched, foreign, or inventory-unavailable receipts and never uses a blanket session delete.
 
diff --git a/scripts/grid-recovery-drill.mjs b/scripts/grid-recovery-drill.mjs
index 73b23de..607ecca 100644
--- a/scripts/grid-recovery-drill.mjs
+++ b/scripts/grid-recovery-drill.mjs
@@ -54,7 +54,7 @@ export function validatePrivateGridUrl(value) {
 
 export function parseGridInventory(payload) {
   const value = payload?.value;
-  if (!value || value.ready !== true || !Array.isArray(value.nodes)) {
+  if (!value || typeof value.ready !== 'boolean' || !Array.isArray(value.nodes)) {
     throw failure('GRID_INVENTORY_SCHEMA_UNAVAILABLE', 'UNKNOWN', 'inventory');
   }
   const sessions = [];
@@ -90,7 +90,7 @@ export function parseGridInventory(payload) {
     }
   }
   return Object.freeze({
-    ready: true,
+    ready: value.ready,
     nodeCount: value.nodes.length,
     sessions: Object.freeze(sessions),
     stereotypes: Object.freeze(stereotypes),
@@ -362,6 +362,7 @@ async function waitForGrid(grid, fetchImpl = fetch) {
     try {
       const inventory = await readGridInventory(grid, fetchImpl);
       if (
+        inventory.ready &&
         inventory.nodeCount > 0 &&
         inventory.stereotypes.includes('chrome')
       ) {
diff --git a/scripts/grid-recovery.test.mjs b/scripts/grid-recovery.test.mjs
index cf9975c..84b0b55 100644
--- a/scripts/grid-recovery.test.mjs
+++ b/scripts/grid-recovery.test.mjs
@@ -16,10 +16,10 @@ const ownedId = 'owned:session-1';
 const origin = 'http://127.0.0.1:4444';
 const grid = { origin, endpoint: `${origin}/wd/hub` };
 
-function statusPayload(sessions = [], browserName = 'chrome') {
+function statusPayload(sessions = [], browserName = 'chrome', ready = true) {
   return {
     value: {
-      ready: true,
+      ready,
       nodes: [{
         id: 'node-test',
         slots: [{
@@ -84,7 +84,17 @@ test('capability classification follows the validated slot inventory and structu
 test('missing or changed Grid inventory schemas fail closed', () => {
   assert.throws(() => parseGridInventory({ value: { ready: true } }), /GRID_INVENTORY_SCHEMA_UNAVAILABLE/);
   assert.throws(() => parseGridInventory({ value: { ready: true, nodes: [{ slots: [{}] }] } }), /GRID_INVENTORY_SCHEMA_UNAVAILABLE/);
-  assert.throws(() => parseGridInventory({ value: { ready: false, nodes: [] } }), /GRID_INVENTORY_SCHEMA_UNAVAILABLE/);
+  assert.throws(() => parseGridInventory({ value: { nodes: [] } }), /GRID_INVENTORY_SCHEMA_UNAVAILABLE/);
+});
+
+test('a full single-slot Grid can be not ready and still expose its live session', () => {
+  const inventory = parseGridInventory(
+    statusPayload([{ sessionId: ownedId }], 'chrome', false),
+  );
+
+  assert.equal(inventory.ready, false);
+  assert.equal(inventory.sessions.length, 1);
+  assert.equal(inventory.sessions[0].sessionId, ownedId);
 });
 
 test('private Grid URL rejects remote hosts, credentials, query strings and fragments', () => {