fix(supervisor): treat a missing cancel route as unsupported, not an error

Author: nicktrn

apps/supervisor/src/workloadServer/index.ts

@@ -63,7 +63,7 @@ const checkpointDeleteRequests = new Counter({
 const checkpointCancelRequests = new Counter({
   name: "checkpoint_cancel_requests_total",
   help: "Checkpoint cancel requests attempted when a run continues, by outcome",
-  labelNames: ["result"], // "sent" | "no_client" | "not_applicable" | "http_error"
+  labelNames: ["result"],
   registers: [register],
 });
 
@@ -306,16 +306,22 @@ export class WorkloadServer extends EventEmitter<WorkloadServerEvents> {
       return;
     }
 
-    const [error, accepted] = await tryCatch(
+    const [error, outcome] = await tryCatch(
       this.checkpointClient.cancelCheckpoints({ runFriendlyId })
     );
 
-    if (error || !accepted) {
+    if (error || outcome === "failed") {
       checkpointCancelRequests.inc({ result: "http_error" });
       this.logger.error("Failed to request checkpoint cancel", { runFriendlyId, error });
       return;
     }
 
+    if (outcome === "unsupported") {
+      checkpointCancelRequests.inc({ result: "unsupported" });
+      this.logger.debug("Checkpoint cancel not supported", { runFriendlyId });
+      return;
+    }
+
     checkpointCancelRequests.inc({ result: "sent" });
   }
 

packages/core/src/v3/serverOnly/checkpointClient.ts

@@ -158,20 +158,32 @@ export class CheckpointClient {
     return true;
   }
 
-  async cancelCheckpoints({ runFriendlyId }: { runFriendlyId: string }): Promise<boolean> {
+  /**
+   * cancelCheckpoints returns "unsupported" when the route is absent, which is expected while a
+   * newer caller runs against an older checkpoint service.
+   */
+  async cancelCheckpoints({
+    runFriendlyId,
+  }: {
+    runFriendlyId: string;
+  }): Promise<"ok" | "unsupported" | "failed"> {
     const res = await fetch(
       new URL(`/api/v1/runs/${runFriendlyId}/checkpoints/cancel`, this.opts.apiUrl),
       { method: "POST" }
     );
 
+    if (res.status === 404) {
+      return "unsupported";
+    }
+
     if (!res.ok) {
       this.logger.error("[CheckpointClient] Cancel checkpoints request failed", {
         runFriendlyId,
         status: res.status,
       });
-      return false;
+      return "failed";
     }
 
-    return true;
+    return "ok";
   }
 }