fix(server): thread list reads no longer freeze the server on large databases

apps/server/src/orchestration-v2/ShellStream.ts

@@ -12,6 +12,7 @@ import {
   OrchestrationProjectShell as ProjectShellSchema,
   OrchestrationV2ThreadShell as ThreadShellSchema,
 } from "@t3tools/contracts";
+import * as NodeSqliteClient from "@t3tools/shared/nodeSqliteClient";
 import * as Clock from "effect/Clock";
 import * as Effect from "effect/Effect";
 import * as Schema from "effect/Schema";
@@ -46,13 +47,16 @@ export const loadShellSnapshotParts = <E1, E2, E3, E4>(input: {
   readonly latestSequence: Effect.Effect<number, E4>;
 }) =>
   Effect.gen(function* () {
-    const read = yield* input.sql.withTransaction(
-      Effect.all({
-        decodeThreads: input.readThreads,
-        projects: input.listProjects,
-        snapshotSequence: input.latestSequence,
-      }),
-    );
+    // Read on the reader worker, off the event loop and the writer.
+    const read = yield* input.sql
+      .withTransaction(
+        Effect.all({
+          decodeThreads: input.readThreads,
+          projects: input.listProjects,
+          snapshotSequence: input.latestSequence,
+        }),
+      )
+      .pipe(NodeSqliteClient.readOnly);
     return {
       projects: read.projects,
       threads: yield* read.decodeThreads,

apps/server/src/orchestration-v2/ThreadPullRequestService.ts

@@ -11,6 +11,7 @@ import {
   type ThreadLinkedPullRequest,
 } from "@t3tools/contracts";
 import { makeDrainableWorker } from "@t3tools/shared/DrainableWorker";
+import * as NodeSqliteClient from "@t3tools/shared/nodeSqliteClient";
 import * as Cause from "effect/Cause";
 import * as Context from "effect/Context";
 import * as Crypto from "effect/Crypto";
@@ -127,10 +128,14 @@ export const make = Effect.gen(function* () {
    */
   const readThreadSnapshot = ({ threadId, backfill }: RefreshRequest) =>
     threadId === null
-      ? orchestrator.getShellSnapshot({
-          location: "active",
-          unsettledOnly: !(backfill || pendingBackfill.size > 0),
-        })
+      ? orchestrator
+          .getShellSnapshot({
+            location: "active",
+            unsettledOnly: !(backfill || pendingBackfill.size > 0),
+          })
+          // Reads every active thread once a minute: run it off the event loop. Writes keep
+          // landing during the read, and the snapshot's sequence guards the sync against them.
+          .pipe(NodeSqliteClient.readOnly)
       : Effect.gen(function* () {
           // Read the sequence first. The thread is then at least this new, so a
           // sync guarded by the sequence is rejected rather than missing a change.

apps/server/src/persistence/Sqlite.ts

@@ -45,6 +45,7 @@ export const layerFromPath = Effect.fn("makeSqlitePersistenceLive")(function* (d
     layerSetup,
     NodeSqliteClient.layer({
       filename: dbPath,
+      readerWorker: true,
       spanAttributes: {
         "db.name": path.basename(dbPath),
         "service.name": "t3code-server",