gateway: persistent PSS key + retry resolveSelfOverlay

Two bugs that together blackholed every reverse-PSS message from the
provider when the api ran more than one replica:

1) `pssKeys = generatePssKeypair()` ran per-container, so each replica
   had a different ECIES identity. Provider encrypts its reply to ONE
   of those keys, only that replica can decrypt — so two replicas =
   ~50% of chats hang forever.

2) `resolveSelfOverlay` swallowed errors and returned 0x00…00 when
   Bee wasn't ready yet at boot. That zeroed overlay landed in the
   gateway context AND in the job_notify envelope sent to the provider
   — provider then sent PSS replies into the void.

Fix: load PSS key from PSS_KEY_PATH (shared volume across replicas);
retry getNodeAddresses up to 5 minutes and throw if still zero.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

src/env.ts

@@ -58,6 +58,12 @@ const EnvSchema = z.object({
   HOSTED_GATEWAY_PORT: z.coerce.number().int().positive().default(8080),
   LOG_LEVEL: z.enum(['fatal', 'error', 'warn', 'info', 'debug', 'trace']).default('info'),
 
+  // Operator PSS keypair file. All `prod_api` and `prod_worker` replicas
+  // share this key so providers can encrypt their PSS replies to a single
+  // identity. The compose mounts a named volume at /data so the key persists
+  // across deploys and is visible to every replica.
+  PSS_KEY_PATH: z.string().default('/data/pss.key'),
+
   // Phase 2 — Stripe (optional)
   STRIPE_SECRET_KEY: z.string().optional(),
   STRIPE_WEBHOOK_SECRET: z.string().optional(),

src/gateway/context.ts

@@ -14,7 +14,7 @@ import { discoverUsableBatchId, downloadChunk, PssTransport } from '../t4t/lib/s
 import { ensureManagedStamp, topUpIfBelow, type ManagedStamp } from '../t4t/lib/stamps';
 import { clientTopic } from '../t4t/lib/envelope';
 import { EciesCipher, jsonDecrypt } from '../t4t/lib/crypto';
-import { generatePssKeypair } from '../t4t/lib/keys';
+import { loadOrCreatePssKey } from '../t4t/lib/keys';
 import { ModelDiscovery } from '../t4t/modes/gateway/models';
 import type {
   JobAckBody,
@@ -89,12 +89,16 @@ async function bootGatewayContext(): Promise<GatewayContext> {
     xbzz: env.XBZZ_ADDRESS as Hex,
   });
 
-  // Operator PSS keypair — non-persistent for now; tenants reuse this single
+  // Operator PSS keypair — persisted to disk so every api replica shares the
-  // operator key (providers see all jobs as coming from the same client PSS
+  // same identity. Without this, each replica generates a fresh keypair at
-  // identity, which matches the on-chain wallet model).
+  // boot and providers encrypt their PSS replies to a key only one replica
-  const pssKeys = generatePssKeypair();
+  // can decrypt, so most chat completions hang.
+  const pssKeys = loadOrCreatePssKey(env.PSS_KEY_PATH);
 
   const cipher = new EciesCipher(pssKeys.privateKey);
+  // Block until Bee returns a non-zero overlay; without this, a replica that
+  // races Bee's boot persists `selfOverlay = 0x00…00` in its gateway context
+  // and providers send PSS replies into the void.
   const selfOverlay = await resolveSelfOverlay(bee);
   const pss = new PssTransport({ bee, postageBatchId, logger, selfAddress: chain.address });
 
@@ -297,10 +301,18 @@ async function resolvePostageBatch(bee: Bee): Promise<string> {
 }
 
 async function resolveSelfOverlay(bee: Bee): Promise<Hex> {
+  const zero = '0x' + '00'.repeat(32);
+  for (let attempt = 0; attempt < 60; attempt++) {
     const info = await bee.getNodeAddresses().catch(() => null);
-  const overlay = info?.overlay?.toString() ?? '00'.repeat(32);
+    const overlay = info?.overlay?.toString();
+    if (overlay && overlay.replace(/^0x/, '') !== '00'.repeat(32)) {
       return ('0x' + overlay.replace(/^0x/, '')) as Hex;
     }
+    logger.warn({ attempt }, 'bee.getNodeAddresses returned no overlay yet — retrying in 5s');
+    await new Promise((r) => setTimeout(r, 5000));
+  }
+  throw new Error('bee did not return a non-zero overlay after 60 attempts — refusing to boot with a zeroed selfOverlay');
+}
 
 /** Update a GatewayJob row. Direct prisma write (not model.gatewayJob.update)
  *  because we don't need fan-out for status transitions in v1. */