Make routing discovery less picky on errors

Author: ali-ince

src/v1/internal/connection-providers.js

@@ -211,7 +211,10 @@ export class LoadBalancer extends ConnectionProvider {
         // try next router
         return this._createSessionForRediscovery(currentRouter).then(session => {
           if (session) {
-            return this._rediscovery.lookupRoutingTableOnRouter(session, currentRouter);
+            return this._rediscovery.lookupRoutingTableOnRouter(session, currentRouter).catch(error => {
+              this._log.warn(`unable to fetch routing table because of an error ${error}`);
+              return null;
+            });
           } else {
             // unable to acquire connection and create session towards the current router
             // return null to signal that the next router should be tried

test/internal/connection-providers.test.js

@@ -1117,6 +1117,9 @@ function newLoadBalancerWithSeedRouter(seedRouter, seedRouterResolved,
   loadBalancer._routingTable = new RoutingTable(routers, readers, writers, expirationTime);
   loadBalancer._rediscovery = new FakeRediscovery(routerToRoutingTable);
   loadBalancer._hostNameResolver = new FakeDnsResolver(seedRouterResolved);
+  if (expirationTime === Integer.ZERO) {
+    loadBalancer._useSeedRouter = false;
+  }
   return loadBalancer;
 }
 
@@ -1170,7 +1173,7 @@ class FakeRediscovery {
   }
 
   lookupRoutingTableOnRouter(ignored, router) {
-    return this._routerToRoutingTable[router.asKey()];
+    return Promise.resolve(this._routerToRoutingTable[router.asKey()]);
   }
 }
 

test/internal/node/routing.driver.boltkit.test.js

@@ -198,7 +198,7 @@ describe('routing driver with stub server', () => {
       // When
       const session = driver.session(neo4j.READ);
       session.run("MATCH (n) RETURN n.name").catch(err => {
-        expect(err.code).toEqual(neo4j.error.PROTOCOL_ERROR);
+        expect(err.code).toEqual(neo4j.error.SERVICE_UNAVAILABLE);
 
         session.close();
         driver.close();
@@ -562,8 +562,6 @@ describe('routing driver with stub server', () => {
       // When
       const session1 = driver.session(neo4j.session.READ);
       session1.run("MATCH (n) RETURN n.name").catch(() => {
-        const session2 = driver.session(neo4j.session.READ);
-        session2.run("MATCH (n) RETURN n.name").then(() => {
           driver.close();
           seedServer.exit(code1 => {
             readServer.exit(code2 => {
@@ -572,7 +570,6 @@ describe('routing driver with stub server', () => {
               done();
             });
           });
-        });
       });
     });
   });
@@ -592,8 +589,8 @@ describe('routing driver with stub server', () => {
       const session = driver.session();
       session.run("MATCH (n) RETURN n.name").catch(err => {
         expect(err.code).toEqual(neo4j.error.SERVICE_UNAVAILABLE);
-        expect(err.message.indexOf('Make sure you are connecting to a causal cluster') > 0).toBeTruthy();
-        assertHasRouters(driver, ['127.0.0.1:9001']);
+        expect(err.message).toContain('Could not perform discovery');
+        assertHasRouters(driver, []);
         session.close();
         driver.close();
         server.exit(code => {
@@ -960,31 +957,31 @@ describe('routing driver with stub server', () => {
     });
   });
 
-  it('should throw protocol error when no records', done => {
+  it('should throw error when no records', done => {
     testForProtocolError('./test/resources/boltstub/empty_get_servers_response.script', done);
   });
 
-  it('should throw protocol error when no TTL entry', done => {
+  it('should throw error when no TTL entry', done => {
     testForProtocolError('./test/resources/boltstub/no_ttl_entry_get_servers.script', done);
   });
 
-  it('should throw protocol error when no servers entry', done => {
+  it('should throw error when no servers entry', done => {
     testForProtocolError('./test/resources/boltstub/no_servers_entry_get_servers.script', done);
   });
 
-  it('should throw protocol error when multiple records', done => {
+  it('should throw error when multiple records', done => {
     testForProtocolError('./test/resources/boltstub/unparsable_ttl_get_servers.script', done);
   });
 
-  it('should throw protocol error on unparsable record', done => {
+  it('should throw error on unparsable record', done => {
     testForProtocolError('./test/resources/boltstub/unparsable_servers_get_servers.script', done);
   });
 
-  it('should throw protocol error when no routers', done => {
+  it('should throw error when no routers', done => {
     testForProtocolError('./test/resources/boltstub/no_routers_get_servers.script', done);
   });
 
-  it('should throw protocol error when no readers', done => {
+  it('should throw error when no readers', done => {
     testForProtocolError('./test/resources/boltstub/no_readers_get_servers.script', done);
   });
 
@@ -2125,6 +2122,49 @@ describe('routing driver with stub server', () => {
     });
   })
 
+  it('should revert to initial router if the only known router returns invalid routing table', done => {
+    if (!boltStub.supported) {
+      done();
+      return;
+    }
+
+    // the first seed to get the routing table
+    // the returned routing table includes a non-reachable read-server and points to only one router
+    // which will return an invalid routing table
+    const seedServer1 = boltStub.start('./test/resources/boltstub/acquire_endpoints_v3_point_to_empty_router.script', 9001);
+    // returns an empty routing table
+    const failingSeedServer2 = boltStub.start('./test/resources/boltstub/acquire_endpoints_v3_empty.script', 9004);
+    // returns a normal routing table
+    const seedServer3 = boltStub.start('./test/resources/boltstub/acquire_endpoints_v3_three_servers.script', 9003);
+    // ordinary read server
+    const readServer = boltStub.start('./test/resources/boltstub/read_server_v3_read_tx.script', 9002);
+
+    boltStub.run(() => {
+      const driver = boltStub.newDriver('bolt+routing://my.virtual.host:8080', {
+        resolver: address => ['127.0.0.1:9001', '127.0.0.1:9003']
+      });
+
+      const session = driver.session(neo4j.session.READ);
+      session.readTransaction(tx => tx.run('MATCH (n) RETURN n.name')).then(res => {
+        session.close();
+        driver.close();
+        seedServer1.exit(code1 => {
+          failingSeedServer2.exit(code2 => {
+            seedServer3.exit(code3 => {
+              readServer.exit(code4 => {
+                expect(code1).toEqual(0);
+                expect(code2).toEqual(0);
+                expect(code3).toEqual(0);
+                expect(code4).toEqual(0);
+                done();
+              });
+            });
+          });
+        });
+      }).catch(error => done.fail(error));
+    });
+  });
+
   function testAddressPurgeOnDatabaseError(query, accessMode, done) {
     if (!boltStub.supported) {
       done();
@@ -2254,7 +2294,7 @@ describe('routing driver with stub server', () => {
 
       const session = driver.session();
       session.run('MATCH (n) RETURN n.name').catch(error => {
-        expect(error.code).toEqual(neo4j.error.PROTOCOL_ERROR);
+        expect(error.code).toEqual(neo4j.error.SERVICE_UNAVAILABLE);
 
         session.close();
         driver.close();

test/resources/boltstub/acquire_endpoints_v3_empty.script

@@ -0,0 +1,9 @@
+!: BOLT 3
+!: AUTO HELLO
+!: AUTO RESET
+
+C: RUN "CALL dbms.cluster.routing.getRoutingTable($context)" {"context": {}} {}
+   PULL_ALL
+S: SUCCESS {"fields": ["ttl", "servers"]}
+   RECORD [9223372036854775807, []]
+   SUCCESS {}

test/resources/boltstub/acquire_endpoints_v3_point_to_empty_router.script

@@ -0,0 +1,12 @@
+!: BOLT 3
+!: AUTO HELLO
+!: AUTO RESET
+
+C: RUN "CALL dbms.cluster.routing.getRoutingTable($context)" {"context": {}} {}
+   PULL_ALL
+S: SUCCESS {"fields": ["ttl", "servers"]}
+   RECORD [9223372036854775807, [{"addresses": ["127.0.0.1:9010"],"role": "WRITE"}, {"addresses": ["127.0.0.1:9011"], "role": "READ"},{"addresses": ["127.0.0.1:9004"], "role": "ROUTE"}]]
+   SUCCESS {}
+C: RESET
+S: SUCCESS {}
+   <EXIT>

test/resources/boltstub/acquire_endpoints_v3_three_servers.script

@@ -0,0 +1,11 @@
+!: BOLT 3
+!: AUTO HELLO
+
+C: RUN "CALL dbms.cluster.routing.getRoutingTable($context)" {"context": {}} {}
+   PULL_ALL
+S: SUCCESS {"fields": ["ttl", "servers"]}
+   RECORD [9223372036854775807, [{"addresses": ["127.0.0.1:9001"],"role": "WRITE"}, {"addresses": ["127.0.0.1:9002","127.0.0.1:9003"], "role": "READ"},{"addresses": ["127.0.0.1:9001","127.0.0.1:9002","127.0.0.1:9003"], "role": "ROUTE"}]]
+   SUCCESS {}
+C: RESET
+S: SUCCESS {}
+   <EXIT>

test/resources/boltstub/read_server_v3_read.script

@@ -1,8 +1,7 @@
 !: BOLT 3
+!: AUTO HELLO
 !: AUTO RESET
 
-C: HELLO {"scheme": "basic", "principal": "neo4j", "credentials": "password", "user_agent": "neo4j-javascript/0.0.0-dev"}
-S: SUCCESS {"server": "Neo4j/9.9.9", "connection_id": "bolt-123456789"}
 C: RUN "MATCH (n) RETURN n.name" {} {"mode": "r"}
    PULL_ALL
 S: SUCCESS {"fields": ["n.name"]}

test/resources/boltstub/read_server_v3_read_tx.script

@@ -1,8 +1,7 @@
 !: BOLT 3
+!: AUTO HELLO
 !: AUTO RESET
 
-C: HELLO {"scheme": "basic", "principal": "neo4j", "credentials": "password", "user_agent": "neo4j-javascript/0.0.0-dev"}
-S: SUCCESS {"server": "Neo4j/9.9.9", "connection_id": "bolt-123456789"}
 C: BEGIN {"mode": "r"}
 S: SUCCESS {}
 C: RUN "MATCH (n) RETURN n.name" {} {"mode": "r"}

test/v1/driver.test.js

@@ -211,7 +211,7 @@ describe('driver', () => {
 
     // Expect
     driver.onError = error => {
-      expect(error.message).toEqual(`Server at localhost:7687 can't perform routing. Make sure you are connecting to a causal cluster`);
+      expect(error.message).toContain('Could not perform discovery. No routing servers available.');
       expect(error.code).toEqual(neo4j.error.SERVICE_UNAVAILABLE);
       done();
     };