fix: prevent race conditions with KV locks in dispatch and group provisioning

server/lib/core/dispatch.ts

@@ -197,60 +197,131 @@ export async function dispatchEvent(
         if (message.updateKey) {
           const groupPrefix = route.groupId ? `${route.groupId}:` : "";
           const kvKey = `msg:${groupPrefix}${route.id}:${message.updateKey}:${targetStr}`;
-          const existingId = await env.KV.get(kvKey);
-          if (existingId) {
-            result = await driver.edit(message, target, env, existingId);
-            if (result.ok) {
-              attempts.push({
-                groupId: route.groupId,
-                routeId: route.id,
-                routeName: route.name,
-                target: targetStr,
-                ok: true,
-              });
-              await recordSend(env.DB, {
-                ...base,
-                ok: true,
-                status: result.status,
-                messageId: existingId,
-                platform: driver.id,
-                attempts: result.attempts,
-                durationMs: Date.now() - started,
-                errorCode: result.errorCode,
-              });
+          const lockKey = `msg:lock:${kvKey}`;
+
+          // Acquire a short-lived lock so concurrent events for the same
+          // updateKey don't race (both read null, both send, the later put
+          // overwrites). KV is eventually consistent so the lock is
+          // best-effort; a retry loop further shrinks the window.
+          let locked = false;
+          for (let attempt = 0; attempt < 3; attempt++) {
+            const holder = await env.KV.get(lockKey);
+            if (holder) {
+              const existing = await env.KV.get(kvKey);
+              if (existing) {
+                result = await driver.edit(message, target, env, existing);
+                if (result.ok || /not modified/i.test(result.error ?? "")) {
+                  const ok = result.ok || /not modified/i.test(result.error ?? "");
+                  attempts.push({
+                    groupId: route.groupId,
+                    routeId: route.id,
+                    routeName: route.name,
+                    target: targetStr,
+                    ok: true,
+                  });
+                  await recordSend(env.DB, {
+                    ...base,
+                    ok: true,
+                    status: result.status,
+                    messageId: existing,
+                    platform: driver.id,
+                    attempts: result.attempts,
+                    durationMs: Date.now() - started,
+                    errorCode: result.errorCode,
+                  });
+                  if (!ok) await env.KV.delete(kvKey);
+                  continue;
+                }
+                await env.KV.delete(kvKey);
+                break;
+              }
+              await new Promise((r) => setTimeout(r, 50 * (attempt + 1)));
               continue;
             }
-            if (/not modified/i.test(result.error ?? "")) {
-              attempts.push({
-                groupId: route.groupId,
-                routeId: route.id,
-                routeName: route.name,
-                target: targetStr,
-                ok: true,
-              });
-              await recordSend(env.DB, {
-                ...base,
-                ok: true,
-                status: result.status,
-                messageId: existingId,
-                platform: driver.id,
-                attempts: result.attempts,
-                durationMs: Date.now() - started,
-                errorCode: result.errorCode,
-              });
-              continue;
-            }
-            await env.KV.delete(kvKey);
+            await env.KV.put(lockKey, "1", { expirationTtl: 60 });
+            locked = true;
+            break;
           }
-          result = await driver.send(message, target, env);
-          if (result.ok && result.messageId) {
-            await env.KV.put(kvKey, result.messageId, { expirationTtl: 604800 });
+
+          try {
+            const existingId = await env.KV.get(kvKey);
+            if (existingId) {
+              result = await driver.edit(message, target, env, existingId);
+              if (result.ok) {
+                attempts.push({
+                  groupId: route.groupId,
+                  routeId: route.id,
+                  routeName: route.name,
+                  target: targetStr,
+                  ok: true,
+                });
+                await recordSend(env.DB, {
+                  ...base,
+                  ok: true,
+                  status: result.status,
+                  messageId: existingId,
+                  platform: driver.id,
+                  attempts: result.attempts,
+                  durationMs: Date.now() - started,
+                  errorCode: result.errorCode,
+                });
+                continue;
+              }
+              if (/not modified/i.test(result.error ?? "")) {
+                attempts.push({
+                  groupId: route.groupId,
+                  routeId: route.id,
+                  routeName: route.name,
+                  target: targetStr,
+                  ok: true,
+                });
+                await recordSend(env.DB, {
+                  ...base,
+                  ok: true,
+                  status: result.status,
+                  messageId: existingId,
+                  platform: driver.id,
+                  attempts: result.attempts,
+                  durationMs: Date.now() - started,
+                  errorCode: result.errorCode,
+                });
+                continue;
+              }
+              await env.KV.delete(kvKey);
+            }
+            result = await driver.send(message, target, env);
+            if (result.ok && result.messageId) {
+              await env.KV.put(kvKey, result.messageId, { expirationTtl: 604800 });
+            }
+          } finally {
+            if (locked) await env.KV.delete(lockKey);
           }
         } else {
           result = await driver.send(message, target, env);
         }
         const durationMs = Date.now() - started;
-        if (!result.ok) throw new Error(result.error ?? "Send failed");
+        if (!result.ok) {
+          const error = result.error ?? "Send failed";
+          attempts.push({
+            groupId: route.groupId,
+            routeId: route.id,
+            routeName: route.name,
+            target: targetStr,
+            ok: false,
+            error,
+          });
+          await recordSend(env.DB, {
+            ...base,
+            ok: false,
+            error,
+            durationMs,
+            status: result.status,
+            platform: driver.id,
+            attempts: result.attempts,
+            errorCode: result.errorCode,
+          });
+          continue;
+        }
         attempts.push({
           groupId: route.groupId,
           routeId: route.id,

server/lib/drivers/discord/interactions.ts

@@ -33,6 +33,19 @@ const GITHUB_COMMENT_RE =
   /github\.com\/([^/\s]+)\/([^/\s]+)\/(?:issues|pull)\/\d+#issuecomment-(\d+)/;
 const GITHUB_ISSUE_RE = /github\.com\/([^/\s]+)\/([^/\s]+)\/(?:issues|pull)\/(\d+)/;
 
+/**
+ * Split a pipe-delimited custom-id suffix into [owner, repo, number]. The
+ * number is always the last segment (arbitrary-length owner/repo can contain
+ * pipes on non-GitHub forges), so we split from the right.
+ */
+function splitThree(rest: string): [string, string, string] {
+  const parts = rest.split("|");
+  const number = parts.pop();
+  const repo = parts.pop();
+  const owner = parts.join("|");
+  return [owner ?? "", repo ?? "", number ?? ""];
+}
+
 interface Interaction {
   id: string;
   token: string;
@@ -101,7 +114,8 @@ export async function handleInteractionRequest(request: Request, env: Env): Prom
     );
     return new Response("Invalid signature", { status: 401 });
   }
-  if (Math.abs(Math.floor(Date.now() / 1000) - Number(timestamp)) > TIMESTAMP_TOLERANCE_SECONDS) {
+  const ts = Number(timestamp);
+  if (!Number.isFinite(ts) || Math.abs(Math.floor(Date.now() / 1000) - ts) > TIMESTAMP_TOLERANCE_SECONDS) {
     return new Response("Invalid signature", { status: 401 });
   }
 
@@ -287,7 +301,7 @@ async function handleButton(
     return;
   }
 
-  const [owner, repo, number] = rest.split("|");
+  const [owner, repo, number] = splitThree(rest);
   if (!owner || !repo || !number) return;
 
   // Acknowledge first (deferred, ephemeral) so the clicker sees a spinner
@@ -494,7 +508,7 @@ async function modalSubmit(
 
   // ghc|add|owner|repo|issueNumber
   if (customId.startsWith(MODAL_ADD)) {
-    const [owner, repo, number] = customId.slice(MODAL_ADD.length).split("|");
+    const [owner, repo, number] = splitThree(customId.slice(MODAL_ADD.length));
     if (!owner || !repo || !number)
       return respond(env, id, token, "内部错误：无法解析目标 issue。");
     try {
@@ -514,7 +528,7 @@ async function modalSubmit(
 
   // ghc|edit|owner|repo|commentId
   if (customId.startsWith(MODAL_EDIT)) {
-    const [owner, repo, commentId] = customId.slice(MODAL_EDIT.length).split("|");
+    const [owner, repo, commentId] = splitThree(customId.slice(MODAL_EDIT.length));
     if (!owner || !repo || !commentId)
       return respond(env, id, token, "内部错误：无法解析目标评论。");
     try {

server/lib/web/groups.ts

@@ -120,36 +120,57 @@ export async function ensureInstallationGroup(
   installationId: number,
   accountLogin: string,
 ): Promise<Group | null> {
-  const groups = await loadGroups(kv);
-  const existing = groups.find((g) => g.installationId === installationId);
-  if (existing) return existing;
-
-  const login = accountLogin.trim().toLowerCase();
-  const candidates = groups.filter(
-    (g) =>
-      g.installationId == null &&
-      login.length > 0 &&
-      (g.owners ?? []).some((o) => o.trim().toLowerCase() === login),
-  );
-  if (candidates.length > 0) {
-    const next = groups.map((g) => (candidates.includes(g) ? { ...g, installationId } : g));
-    await saveGroups(kv, next);
-    return next.find((g) => g.id === candidates[0]!.id) ?? null;
+  // A per-installation lock serializes concurrent `installation.created`
+  // webhooks: without it two requests both pass the "no group bound" check,
+  // both create `inst-{id}`, and the second save overwrites the first.
+  const lockKey = `inst:lock:${installationId}`;
+  for (let attempt = 0; attempt < 5; attempt++) {
+    const holder = await kv.get(lockKey);
+    if (!holder) break;
+    const groups = await loadGroups(kv);
+    const existing = groups.find((g) => g.installationId === installationId);
+    if (existing) return existing;
+    await new Promise((r) => setTimeout(r, 40 * (attempt + 1)));
   }
 
-  const gid = `inst-${installationId}`;
-  const dedicated = groups.find((g) => g.id === gid);
-  const group: Group = {
-    id: gid,
-    name: accountLogin.trim() || `Installation ${installationId}`,
-    adminIds: [],
-    installationId,
-  };
-  await saveGroups(
-    kv,
-    dedicated ? groups.map((g) => (g.id === gid ? { ...g, ...group } : g)) : [...groups, group],
-  );
-  return group;
+  let locked = false;
+  try {
+    await kv.put(lockKey, "1", { expirationTtl: 60 });
+    locked = true;
+
+    const groups = await loadGroups(kv);
+    const existing = groups.find((g) => g.installationId === installationId);
+    if (existing) return existing;
+
+    const login = accountLogin.trim().toLowerCase();
+    const candidates = groups.filter(
+      (g) =>
+        g.installationId == null &&
+        login.length > 0 &&
+        (g.owners ?? []).some((o) => o.trim().toLowerCase() === login),
+    );
+    if (candidates.length > 0) {
+      const next = groups.map((g) => (candidates.includes(g) ? { ...g, installationId } : g));
+      await saveGroups(kv, next);
+      return next.find((g) => g.id === candidates[0]!.id) ?? null;
+    }
+
+    const gid = `inst-${installationId}`;
+    const dedicated = groups.find((g) => g.id === gid);
+    const group: Group = {
+      id: gid,
+      name: accountLogin.trim() || `Installation ${installationId}`,
+      adminIds: [],
+      installationId,
+    };
+    await saveGroups(
+      kv,
+      dedicated ? groups.map((g) => (g.id === gid ? { ...g, ...group } : g)) : [...groups, group],
+    );
+    return group;
+  } finally {
+    if (locked) await kv.delete(lockKey);
+  }
 }
 
 export interface AccessScope {

server/lib/webhook.ts

@@ -50,7 +50,16 @@ export async function processWebhook(
     return { status: 400, body: { error: "Unknown webhook provider" } };
   }
 
-  if (!(await provider.verify(body, headers, effectiveEnv))) {
+  let verified = false;
+  try {
+    verified = await provider.verify(body, headers, effectiveEnv);
+  } catch (err) {
+    // A malformed secret or an unavailable crypto implementation must fail as
+    // a clean 401, never an uncaught 500.
+    log.warn({ provider: provider.id, err: String(err) }, "Webhook signature verification failed");
+    return { status: 401, body: { error: "Invalid signature" } };
+  }
+  if (!verified) {
     // Log the actual cause: a missing provider secret is a deployment problem,
     // while a mismatched signature usually means the sender used the wrong secret.
     const secret =