Keep Microsoft workload adds serial

Author: RhysSullivan

packages/plugins/microsoft/src/sdk/plugin.test.ts

@@ -806,7 +806,7 @@ describe("Microsoft Graph per-workload add flow", () => {
     ),
   );
 
-  it.effect("addWorkloads runs workload spec fetches in bounded parallel", () =>
+  it.effect("addWorkloads runs workload spec fetches serially", () =>
     Effect.scoped(
       Effect.gen(function* () {
         const workloads = [{ presetId: "mail" }, { presetId: "calendar" }, { presetId: "files" }];
@@ -841,7 +841,7 @@ describe("Microsoft Graph per-workload add flow", () => {
           workloads.map((workload) => workload.presetId),
         );
         expect(specFetchCount).toBe(workloads.length);
-        expect(maxActiveSpecFetches).toBe(workloads.length);
+        expect(maxActiveSpecFetches).toBe(1);
       }),
     ),
   );

packages/plugins/microsoft/src/sdk/plugin.ts

@@ -140,7 +140,11 @@ export interface MicrosoftPluginOptions {
 }
 
 const DEFAULT_MICROSOFT_SLUG = "microsoft_graph";
-const MICROSOFT_ADD_WORKLOADS_CONCURRENCY = 4;
+// Serial on purpose: every workload add stream-compiles the 37MB Graph spec,
+// and concurrent compiles contend on event-loop CPU and the request's single
+// DB connection (measured: parity at best, 20-40x stalls at tail) while
+// stacking multiple 37MB spec strings inside the 128MB Workers isolate.
+const MICROSOFT_ADD_WORKLOADS_CONCURRENCY = 1;
 
 const isMicrosoftPresetWorkloadConfig = (
   workload: MicrosoftWorkloadConfig,