docs(worker): document and expose JetStream storage limits for jobs and KV state

Add `COMPUTE_JOBS_STREAM_MAX_BYTES` and `COMPUTE_JOB_STATES_MAX_BYTES` to
environment example, worker server configuration, and deployment docs. These
variables allow tuning JetStream resource caps for job queue and state storage
to better fit deployment requirements.

compute/worker/.env.example

@@ -27,3 +27,5 @@ S3_FORCE_PATH_STYLE=true
 
 # Queue + execution tuning
 COMPUTE_PREWARM_MODELS=true
+COMPUTE_JOBS_STREAM_MAX_BYTES=268435456
+COMPUTE_JOB_STATES_MAX_BYTES=67108864

compute/worker/src/server.ts

@@ -243,12 +243,14 @@ async function ensureJetStreamResources(
   whisperTimeoutMs: number,
   pdfTimeoutMs: number,
   attempts: number,
+  maxBytes: number,
 ): Promise<void> {
   const streamConfig = {
     name: JOBS_STREAM_NAME,
     subjects: [WHISPER_JOBS_SUBJECT, LAYOUT_JOBS_SUBJECT],
     retention: RetentionPolicy.Workqueue,
     storage: StorageType.File,
+    max_bytes: maxBytes,
   };
 
   try {
@@ -257,6 +259,7 @@ async function ensureJetStreamResources(
     if (!isAlreadyExistsError(error)) throw error;
     await jsm.streams.update(JOBS_STREAM_NAME, {
       subjects: [WHISPER_JOBS_SUBJECT, LAYOUT_JOBS_SUBJECT],
+      max_bytes: maxBytes,
     });
   }
 
@@ -414,6 +417,8 @@ async function main(): Promise<void> {
   const pdfTimeoutMs = readIntEnv('COMPUTE_PDF_TIMEOUT_MS', 90_000);
   const attempts = readIntEnv('COMPUTE_JOB_ATTEMPTS', 2);
   const prewarmModels = parseBoolEnv('COMPUTE_PREWARM_MODELS', true);
+  const jobsStreamMaxBytes = readIntEnv('COMPUTE_JOBS_STREAM_MAX_BYTES', 256 * 1024 * 1024);
+  const jobStatesMaxBytes = readIntEnv('COMPUTE_JOB_STATES_MAX_BYTES', 64 * 1024 * 1024);
 
   const connectOpts: any = { servers: natsUrl };
   const natsCreds = process.env.NATS_CREDS?.trim();
@@ -433,11 +438,12 @@ async function main(): Promise<void> {
   const js: JetStreamClient = jetstream(nc);
   const jsm: JetStreamManager = await jetstreamManager(nc);
 
-  await ensureJetStreamResources(jsm, whisperTimeoutMs, pdfTimeoutMs, attempts);
+  await ensureJetStreamResources(jsm, whisperTimeoutMs, pdfTimeoutMs, attempts, jobsStreamMaxBytes);
 
   const kv = await new Kvm(js).create(JOB_STATES_BUCKET, {
     history: 1,
     ttl: JOB_STATES_TTL_MS,
+    max_bytes: jobStatesMaxBytes,
   });
 
   const s3 = buildS3Client();

docs-site/docs/deploy/compute-worker.md

@@ -45,6 +45,8 @@ Common optional:
 - `COMPUTE_WORKER_PORT=8081`
 - `COMPUTE_LOG_FORMAT=pretty` (default) or `json`
 - `COMPUTE_PREWARM_MODELS=true`
+- `COMPUTE_JOBS_STREAM_MAX_BYTES=268435456` (256MB JetStream jobs stream cap)
+- `COMPUTE_JOB_STATES_MAX_BYTES=67108864` (64MB JetStream KV bucket cap)
 
 ## App server environment variables (worker mode)
 
@@ -102,6 +104,8 @@ Set these in the Railway worker service:
 COMPUTE_WORKER_HOST=0.0.0.0
 COMPUTE_WORKER_PORT=8081
 COMPUTE_WORKER_TOKEN=<long-random-shared-token>
+COMPUTE_JOBS_STREAM_MAX_BYTES=268435456
+COMPUTE_JOB_STATES_MAX_BYTES=67108864
 
 NATS_URL=tls://connect.ngs.global:4222
 NATS_CREDS="-----BEGIN NATS USER JWT-----
@@ -122,6 +126,7 @@ Notes:
 - `NATS_CREDS` should be the full Synadia `.creds` file content, including begin/end markers.
 - Keep `COMPUTE_WORKER_TOKEN` identical between app server and worker.
 - If your platform supports mounted files, you can use `NATS_CREDS_FILE` instead of `NATS_CREDS`.
+- `COMPUTE_JOBS_STREAM_MAX_BYTES` and `COMPUTE_JOB_STATES_MAX_BYTES` are optional; defaults are `268435456` (256MiB) and `67108864` (64MiB).
 
 ### 4. Configure the OpenReader app server (worker mode)