feat: add subscription retries

src/GraphQL/Client/BaseClients/Apollo.js

@@ -48,16 +48,44 @@ const createClientWithWebsockets = function (opts) {
     },
   });
 
+  // Close codes that indicate expected/benign disconnects
+  // 1000: Normal closure, 1001: Going away (tab close, navigation, mobile backgrounding)
+  const expectedCloseCodes = new Set([1000, 1001]);
+
   const wsLink = new GraphQLWsLink(
     createWsClient({
       webSocketImpl: globalThis.WebSocket,
       url: opts.websocketUrl,
       timeout: 30000,
+      retryAttempts: Infinity,
+      shouldRetry: () => true,
       connectionParams: {
         headers: opts.authToken
           ? { Authorization: `Bearer ${opts.authToken}` }
           : {},
       },
+      on: {
+        closed: (/** @type {CloseEvent} */ event) => {
+          const code = /** @type {number | undefined} */ (event?.code);
+          const reason = /** @type {string | undefined} */ (event?.reason);
+          const wasClean = /** @type {boolean | undefined} */ (event?.wasClean);
+          if (expectedCloseCodes.has(code)) {
+            return;
+          }
+          console.warn("[graphql-ws] Socket closed unexpectedly", {
+            code,
+            reason,
+            wasClean,
+            url: opts.websocketUrl,
+          });
+        },
+        error: (error) => {
+          console.error("[graphql-ws] WebSocket error", {
+            error,
+            url: opts.websocketUrl,
+          });
+        },
+      },
     }),
   );
 

src/GraphQL/Client/Query.js

@@ -0,0 +1,25 @@
+// Error codes where the request definitely never reached the server
+const noConnectionErrorCodes = new Set([
+  "ECONNREFUSED",
+  "ENETUNREACH",
+]);
+
+// Error codes where the connection may have been established
+// before failing, so the server may have received the request
+const ambiguousConnectionErrorCodes = new Set([
+  "ETIMEDOUT",
+  "ECONNRESET",
+]);
+
+const getCode = (error) => {
+  const code = error?.cause?.code;
+  return typeof code === "string" ? code : "";
+};
+
+export const isNoConnectionError = (error) =>
+  noConnectionErrorCodes.has(getCode(error));
+
+export const isConnectionError = (error) => {
+  const code = getCode(error);
+  return noConnectionErrorCodes.has(code) || ambiguousConnectionErrorCodes.has(code);
+};

src/GraphQL/Client/Query.purs

@@ -43,6 +43,9 @@ import GraphQL.Client.Types (class GqlQuery, class QueryClient, Client(..), GqlR
 import GraphQL.Client.Variables (class VarsTypeChecked, getVarsJson, getVarsTypeNames)
 import Type.Proxy (Proxy(..))
 
+foreign import isConnectionError :: Error -> Boolean
+foreign import isNoConnectionError :: Error -> Boolean
+
 -- | Run a graphQL query with a custom decoder and custom options
 queryOptsWithDecoder
   :: forall client directives schema query returns queryOpts mutationOpts sr
@@ -199,8 +202,12 @@ runQuery
   -> Aff returns
 runQuery decodeFn opts client _ queryNameUnsafe q =
   addErrorInfo (Proxy @schema) queryName q do
-    json <- clientQuery opts client queryName (getVarsTypeNames (Proxy :: _ schema) q <> toGqlQueryString q)
-      (getVarsJson (Proxy :: _ schema) q)
+    let doQuery = clientQuery opts client queryName (getVarsTypeNames (Proxy :: _ schema) q <> toGqlQueryString q)
+          (getVarsJson (Proxy :: _ schema) q)
+    json <- doQuery `catchError` \err ->
+      -- Retry once on connection-level errors (ETIMEDOUT, ECONNREFUSED, etc.)
+      if isConnectionError err then doQuery
+      else throwError err
     decodeJsonData decodeFn json
   where
   queryName = safeQueryName queryNameUnsafe
@@ -218,8 +225,12 @@ runMutation
   -> Aff returns
 runMutation decodeFn opts client _ queryNameUnsafe q =
   addErrorInfo (Proxy @schema) queryName q do
-    json <- clientMutation opts client queryName (getVarsTypeNames (Proxy :: _ schema) q <> toGqlQueryString q)
-      (getVarsJson (Proxy :: _ schema) q)
+    let doMutation = clientMutation opts client queryName (getVarsTypeNames (Proxy :: _ schema) q <> toGqlQueryString q)
+          (getVarsJson (Proxy :: _ schema) q)
+    json <- doMutation `catchError` \err ->
+      -- Only retry on ECONNREFUSED/ENETUNREACH where the request never reached the server
+      if isNoConnectionError err then doMutation
+      else throwError err
     decodeJsonData decodeFn json
   where
   queryName = safeQueryName queryNameUnsafe