Merge pull request #11915 from dgquintas/grpclb_drop_fixes

grpclb: deal with all-drops serverlists

src/core/ext/filters/client_channel/lb_policy/grpclb/grpclb.c

@@ -491,11 +491,8 @@ static grpc_lb_addresses *process_serverlist_locked(
   for (size_t i = 0; i < serverlist->num_servers; ++i) {
     if (is_server_valid(serverlist->servers[i], i, true)) ++num_valid;
   }
-  if (num_valid == 0) return NULL;
-
   grpc_lb_addresses *lb_addresses =
       grpc_lb_addresses_create(num_valid, &lb_token_vtable);
-
   /* second pass: actually populate the addresses and LB tokens (aka user data
    * to the outside world) to be read by the RR policy during its creation.
    * Given that the validity tests are very cheap, they are performed again
@@ -503,14 +500,12 @@ static grpc_lb_addresses *process_serverlist_locked(
    * incurr in an allocation due to the arbitrary number of server */
   size_t addr_idx = 0;
   for (size_t sl_idx = 0; sl_idx < serverlist->num_servers; ++sl_idx) {
-    GPR_ASSERT(addr_idx < num_valid);
     const grpc_grpclb_server *server = serverlist->servers[sl_idx];
     if (!is_server_valid(serverlist->servers[sl_idx], sl_idx, false)) continue;
-
+    GPR_ASSERT(addr_idx < num_valid);
     /* address processing */
     grpc_resolved_address addr;
     parse_server(server, &addr);
-
     /* lb token processing */
     void *user_data;
     if (server->has_load_balance_token) {
@@ -596,7 +591,7 @@ static void update_lb_connectivity_status_locked(
         grpc_connectivity_state_name(rr_state), (void *)glb_policy->rr_policy);
   }
   grpc_connectivity_state_set(exec_ctx, &glb_policy->state_tracker, rr_state,
-                              GRPC_ERROR_REF(rr_state_error),
+                              rr_state_error,
                               "update_lb_connectivity_status_locked");
 }
 
@@ -678,11 +673,12 @@ static bool pick_from_internal_rr_locked(
 
 static grpc_lb_policy_args *lb_policy_args_create(grpc_exec_ctx *exec_ctx,
                                                   glb_lb_policy *glb_policy) {
+  grpc_lb_addresses *addresses =
+      process_serverlist_locked(exec_ctx, glb_policy->serverlist);
+  GPR_ASSERT(addresses != NULL);
   grpc_lb_policy_args *args = gpr_zalloc(sizeof(*args));
   args->client_channel_factory = glb_policy->cc_factory;
   args->combiner = glb_policy->base.combiner;
-  grpc_lb_addresses *addresses =
-      process_serverlist_locked(exec_ctx, glb_policy->serverlist);
   // Replace the LB addresses in the channel args that we pass down to
   // the subchannel.
   static const char *keys_to_remove[] = {GRPC_ARG_LB_ADDRESSES};
@@ -727,7 +723,6 @@ static void create_rr_locked(grpc_exec_ctx *exec_ctx, glb_lb_policy *glb_policy,
   /* Connectivity state is a function of the RR policy updated/created */
   update_lb_connectivity_status_locked(exec_ctx, glb_policy, rr_state,
                                        rr_state_error);
-
   /* Add the gRPC LB's interested_parties pollset_set to that of the newly
    * created RR policy. This will make the RR policy progress upon activity on
    * gRPC LB, which in turn is tied to the application's call */
@@ -761,8 +756,8 @@ static void create_rr_locked(grpc_exec_ctx *exec_ctx, glb_lb_policy *glb_policy,
     pp->wrapped_on_complete_arg.client_stats =
         grpc_grpclb_client_stats_ref(glb_policy->client_stats);
     if (GRPC_TRACER_ON(grpc_lb_glb_trace)) {
-      gpr_log(GPR_INFO, "Pending pick about to PICK from 0x%" PRIxPTR "",
+      gpr_log(GPR_INFO, "Pending pick about to (async) PICK from %p",
-              (intptr_t)glb_policy->rr_policy);
+              (void *)glb_policy->rr_policy);
     }
     pick_from_internal_rr_locked(exec_ctx, glb_policy, &pp->pick_args,
                                  true /* force_async */, pp->target,
@@ -788,10 +783,9 @@ static void rr_handover_locked(grpc_exec_ctx *exec_ctx,
                                glb_lb_policy *glb_policy) {
   GPR_ASSERT(glb_policy->serverlist != NULL &&
              glb_policy->serverlist->num_servers > 0);
-
   if (glb_policy->shutting_down) return;
-
   grpc_lb_policy_args *args = lb_policy_args_create(exec_ctx, glb_policy);
+  GPR_ASSERT(args != NULL);
   if (glb_policy->rr_policy != NULL) {
     if (GRPC_TRACER_ON(grpc_lb_glb_trace)) {
       gpr_log(GPR_DEBUG, "Updating Round Robin policy (%p)",
@@ -826,8 +820,8 @@ static void glb_rr_connectivity_changed_locked(grpc_exec_ctx *exec_ctx,
     unref_needed = true;
     gpr_free(rr_connectivity);
   } else { /* rr state != SHUTDOWN && !shutting down: biz as usual */
-    update_lb_connectivity_status_locked(exec_ctx, glb_policy,
+    update_lb_connectivity_status_locked(
-                                         rr_connectivity->state, error);
+        exec_ctx, glb_policy, rr_connectivity->state, GRPC_ERROR_REF(error));
     /* Resubscribe. Reuse the "rr_connectivity_cb" weak ref. */
     grpc_lb_policy_notify_on_state_change_locked(
         exec_ctx, glb_policy->rr_policy, &rr_connectivity->state,
@@ -1089,6 +1083,16 @@ static void glb_shutdown_locked(grpc_exec_ctx *exec_ctx, grpc_lb_policy *pol) {
   }
 }
 
+// Cancel a specific pending pick.
+//
+// A grpclb pick progresses as follows:
+// - If there's a Round Robin policy (glb_policy->rr_policy) available, it'll be
+//   handed over to the RR policy (in create_rr_locked()). From that point
+//   onwards, it'll be RR's responsibility. For cancellations, that implies the
+//   pick needs also be cancelled by the RR instance.
+// - Otherwise, without an RR instance, picks stay pending at this policy's
+//   level (grpclb), inside the glb_policy->pending_picks list. To cancel these,
+//   we invoke the completion closure and set *target to NULL right here.
 static void glb_cancel_pick_locked(grpc_exec_ctx *exec_ctx, grpc_lb_policy *pol,
                                    grpc_connected_subchannel **target,
                                    grpc_error *error) {
@@ -1108,9 +1112,23 @@ static void glb_cancel_pick_locked(grpc_exec_ctx *exec_ctx, grpc_lb_policy *pol,
     }
     pp = next;
   }
+  if (glb_policy->rr_policy != NULL) {
+    grpc_lb_policy_cancel_pick_locked(exec_ctx, glb_policy->rr_policy, target,
+                                      GRPC_ERROR_REF(error));
+  }
   GRPC_ERROR_UNREF(error);
 }
 
+// Cancel all pending picks.
+//
+// A grpclb pick progresses as follows:
+// - If there's a Round Robin policy (glb_policy->rr_policy) available, it'll be
+//   handed over to the RR policy (in create_rr_locked()). From that point
+//   onwards, it'll be RR's responsibility. For cancellations, that implies the
+//   pick needs also be cancelled by the RR instance.
+// - Otherwise, without an RR instance, picks stay pending at this policy's
+//   level (grpclb), inside the glb_policy->pending_picks list. To cancel these,
+//   we invoke the completion closure and set *target to NULL right here.
 static void glb_cancel_picks_locked(grpc_exec_ctx *exec_ctx,
                                     grpc_lb_policy *pol,
                                     uint32_t initial_metadata_flags_mask,
@@ -1132,6 +1150,11 @@ static void glb_cancel_picks_locked(grpc_exec_ctx *exec_ctx,
     }
     pp = next;
   }
+  if (glb_policy->rr_policy != NULL) {
+    grpc_lb_policy_cancel_picks_locked(
+        exec_ctx, glb_policy->rr_policy, initial_metadata_flags_mask,
+        initial_metadata_flags_eq, GRPC_ERROR_REF(error));
+  }
   GRPC_ERROR_UNREF(error);
 }
 
@@ -1463,7 +1486,8 @@ static void query_for_backends_locked(grpc_exec_ctx *exec_ctx,
   op++;
   /* take a weak ref (won't prevent calling of \a glb_shutdown if the strong ref
    * count goes to zero) to be unref'd in lb_on_sent_initial_request_locked() */
-  GRPC_LB_POLICY_WEAK_REF(&glb_policy->base, "lb_on_server_status_received");
+  GRPC_LB_POLICY_WEAK_REF(&glb_policy->base,
+                          "lb_on_sent_initial_request_locked");
   call_error = grpc_call_start_batch_and_execute(
       exec_ctx, glb_policy->lb_call, ops, (size_t)(op - ops),
       &glb_policy->lb_on_sent_initial_request);
@@ -1480,8 +1504,9 @@ static void query_for_backends_locked(grpc_exec_ctx *exec_ctx,
   op->reserved = NULL;
   op++;
   /* take a weak ref (won't prevent calling of \a glb_shutdown if the strong ref
-   * count goes to zero) to be unref'd in lb_on_server_status_received */
+   * count goes to zero) to be unref'd in lb_on_server_status_received_locked */
-  GRPC_LB_POLICY_WEAK_REF(&glb_policy->base, "lb_on_server_status_received");
+  GRPC_LB_POLICY_WEAK_REF(&glb_policy->base,
+                          "lb_on_server_status_received_locked");
   call_error = grpc_call_start_batch_and_execute(
       exec_ctx, glb_policy->lb_call, ops, (size_t)(op - ops),
       &glb_policy->lb_on_server_status_received);
@@ -1493,8 +1518,9 @@ static void query_for_backends_locked(grpc_exec_ctx *exec_ctx,
   op->flags = 0;
   op->reserved = NULL;
   op++;
-  /* take another weak ref to be unref'd in lb_on_response_received */
+  /* take another weak ref to be unref'd/reused in
-  GRPC_LB_POLICY_WEAK_REF(&glb_policy->base, "lb_on_response_received");
+   * lb_on_response_received_locked */
+  GRPC_LB_POLICY_WEAK_REF(&glb_policy->base, "lb_on_response_received_locked");
   call_error = grpc_call_start_batch_and_execute(
       exec_ctx, glb_policy->lb_call, ops, (size_t)(op - ops),
       &glb_policy->lb_on_response_received);
@@ -1511,13 +1537,12 @@ static void lb_on_sent_initial_request_locked(grpc_exec_ctx *exec_ctx,
     do_send_client_load_report_locked(exec_ctx, glb_policy);
   }
   GRPC_LB_POLICY_WEAK_UNREF(exec_ctx, &glb_policy->base,
-                            "lb_on_response_received_locked");
+                            "lb_on_sent_initial_request_locked");
 }
 
 static void lb_on_response_received_locked(grpc_exec_ctx *exec_ctx, void *arg,
                                            grpc_error *error) {
   glb_lb_policy *glb_policy = arg;
-
   grpc_op ops[2];
   memset(ops, 0, sizeof(ops));
   grpc_op *op = ops;
@@ -1548,7 +1573,7 @@ static void lb_on_response_received_locked(grpc_exec_ctx *exec_ctx, void *arg,
         }
         /* take a weak ref (won't prevent calling of \a glb_shutdown() if the
          * strong ref count goes to zero) to be unref'd in
-         * send_client_load_report() */
+         * send_client_load_report_locked() */
         glb_policy->client_load_report_timer_pending = true;
         GRPC_LB_POLICY_WEAK_REF(&glb_policy->base, "client_load_report");
         schedule_next_client_load_report(exec_ctx, glb_policy);
@@ -1576,7 +1601,6 @@ static void lb_on_response_received_locked(grpc_exec_ctx *exec_ctx, void *arg,
             gpr_free(ipport);
           }
         }
-
         /* update serverlist */
         if (serverlist->num_servers > 0) {
           if (grpc_grpclb_serverlist_equals(glb_policy->serverlist,
@@ -1611,9 +1635,7 @@ static void lb_on_response_received_locked(grpc_exec_ctx *exec_ctx, void *arg,
                 grpc_dump_slice(response_slice, GPR_DUMP_ASCII | GPR_DUMP_HEX));
       }
     }
-
     grpc_slice_unref_internal(exec_ctx, response_slice);
-
     if (!glb_policy->shutting_down) {
       /* keep listening for serverlist updates */
       op->op = GRPC_OP_RECV_MESSAGE;
@@ -1621,7 +1643,7 @@ static void lb_on_response_received_locked(grpc_exec_ctx *exec_ctx, void *arg,
       op->flags = 0;
       op->reserved = NULL;
       op++;
-      /* reuse the "lb_on_response_received" weak ref taken in
+      /* reuse the "lb_on_response_received_locked" weak ref taken in
        * query_for_backends_locked() */
       const grpc_call_error call_error = grpc_call_start_batch_and_execute(
           exec_ctx, glb_policy->lb_call, ops, (size_t)(op - ops),
@@ -1629,10 +1651,10 @@ static void lb_on_response_received_locked(grpc_exec_ctx *exec_ctx, void *arg,
       GPR_ASSERT(GRPC_CALL_OK == call_error);
     }
   } else { /* empty payload: call cancelled. */
-           /* dispose of the "lb_on_response_received" weak ref taken in
+           /* dispose of the "lb_on_response_received_locked" weak ref taken in
             * query_for_backends_locked() and reused in every reception loop */
     GRPC_LB_POLICY_WEAK_UNREF(exec_ctx, &glb_policy->base,
-                              "lb_on_response_received_empty_payload");
+                              "lb_on_response_received_locked_empty_payload");
   }
 }
 
@@ -1699,7 +1721,7 @@ static void lb_on_server_status_received_locked(grpc_exec_ctx *exec_ctx,
                     &glb_policy->lb_on_call_retry, now);
   }
   GRPC_LB_POLICY_WEAK_UNREF(exec_ctx, &glb_policy->base,
-                            "lb_on_server_status_received");
+                            "lb_on_server_status_received_locked");
 }
 
 static void glb_update_locked(grpc_exec_ctx *exec_ctx, grpc_lb_policy *policy,

src/core/ext/filters/client_channel/lb_policy/round_robin/round_robin.c

@@ -584,10 +584,16 @@ static void rr_connectivity_changed_locked(grpc_exec_ctx *exec_ctx, void *arg,
   // Dispose of outdated subchannel lists.
   if (sd->subchannel_list != p->subchannel_list &&
       sd->subchannel_list != p->latest_pending_subchannel_list) {
-    // sd belongs to an outdated subchannel_list: get rid of it.
+    char *reason = NULL;
+    if (sd->subchannel_list->shutting_down) {
+      reason = "sl_outdated_straggler";
+      rr_subchannel_list_unref(exec_ctx, sd->subchannel_list, reason);
+    } else {
+      reason = "sl_outdated";
       rr_subchannel_list_shutdown_and_unref(exec_ctx, sd->subchannel_list,
-                                          "sl_outdated");
+                                            reason);
-    GRPC_LB_POLICY_WEAK_UNREF(exec_ctx, &p->base, "sl_outdated");
+    }
+    GRPC_LB_POLICY_WEAK_UNREF(exec_ctx, &p->base, reason);
     return;
   }
   // Now that we're inside the combiner, copy the pending connectivity
@@ -753,6 +759,7 @@ static void rr_update_locked(grpc_exec_ctx *exec_ctx, grpc_lb_policy *policy,
   for (size_t i = 0; i < addresses->num_addresses; i++) {
     if (!addresses->addresses[i].is_balancer) ++num_addrs;
   }
+  rr_subchannel_list *subchannel_list = rr_subchannel_list_create(p, num_addrs);
   if (num_addrs == 0) {
     grpc_connectivity_state_set(
         exec_ctx, &p->state_tracker, GRPC_CHANNEL_TRANSIENT_FAILURE,
@@ -761,18 +768,16 @@ static void rr_update_locked(grpc_exec_ctx *exec_ctx, grpc_lb_policy *policy,
     if (p->subchannel_list != NULL) {
       rr_subchannel_list_shutdown_and_unref(exec_ctx, p->subchannel_list,
                                             "sl_shutdown_empty_update");
-      p->subchannel_list = NULL;
     }
+    p->subchannel_list = subchannel_list;  // empty list
     return;
   }
   size_t subchannel_index = 0;
-  rr_subchannel_list *subchannel_list = rr_subchannel_list_create(p, num_addrs);
   if (p->latest_pending_subchannel_list != NULL && p->started_picking) {
     if (GRPC_TRACER_ON(grpc_lb_round_robin_trace)) {
       gpr_log(GPR_DEBUG,
               "[RR %p] Shutting down latest pending subchannel list %p, about "
-              "to be "
+              "to be replaced by newer latest %p",
-              "replaced by newer latest %p",
               (void *)p, (void *)p->latest_pending_subchannel_list,
               (void *)subchannel_list);
     }
@@ -876,10 +881,10 @@ static grpc_lb_policy *round_robin_create(grpc_exec_ctx *exec_ctx,
                                           grpc_lb_policy_args *args) {
   GPR_ASSERT(args->client_channel_factory != NULL);
   round_robin_lb_policy *p = gpr_zalloc(sizeof(*p));
-  rr_update_locked(exec_ctx, &p->base, args);
   grpc_lb_policy_init(&p->base, &round_robin_lb_policy_vtable, args->combiner);
   grpc_connectivity_state_init(&p->state_tracker, GRPC_CHANNEL_IDLE,
                                "round_robin");
+  rr_update_locked(exec_ctx, &p->base, args);
   if (GRPC_TRACER_ON(grpc_lb_round_robin_trace)) {
     gpr_log(GPR_DEBUG, "[RR %p] Created with %lu subchannels", (void *)p,
             (unsigned long)p->subchannel_list->num_subchannels);

test/cpp/end2end/grpclb_end2end_test.cc

@@ -73,8 +73,8 @@ extern "C" {
 
 using std::chrono::system_clock;
 
-using grpc::lb::v1::LoadBalanceResponse;
 using grpc::lb::v1::LoadBalanceRequest;
+using grpc::lb::v1::LoadBalanceResponse;
 using grpc::lb::v1::LoadBalancer;
 
 namespace grpc {
@@ -647,7 +647,6 @@ class UpdatesTest : public GrpclbEnd2endTest {
 TEST_F(UpdatesTest, UpdateBalancers) {
   const std::vector<int> first_backend{GetBackendPorts()[0]};
   const std::vector<int> second_backend{GetBackendPorts()[1]};
-
   ScheduleResponseForBalancer(
       0, BalancerServiceImpl::BuildResponseForBackends(first_backend, 0, 0), 0);
   ScheduleResponseForBalancer(
@@ -934,6 +933,45 @@ TEST_F(SingleBalancerTest, Drop) {
   EXPECT_EQ(1U, balancer_servers_[0].service_->response_count());
 }
 
+TEST_F(SingleBalancerTest, DropAllFirst) {
+  // All registered addresses are marked as "drop".
+  ScheduleResponseForBalancer(
+      0, BalancerServiceImpl::BuildResponseForBackends({}, 1, 1), 0);
+  const auto& statuses_and_responses = SendRpc(kMessage_, 1);
+  for (const auto& status_and_response : statuses_and_responses) {
+    const Status& status = status_and_response.first;
+    EXPECT_FALSE(status.ok());
+    EXPECT_EQ(status.error_message(), "Call dropped by load balancing policy");
+  }
+}
+
+TEST_F(SingleBalancerTest, DropAll) {
+  ScheduleResponseForBalancer(
+      0, BalancerServiceImpl::BuildResponseForBackends(GetBackendPorts(), 0, 0),
+      0);
+  ScheduleResponseForBalancer(
+      0, BalancerServiceImpl::BuildResponseForBackends({}, 1, 1), 1000);
+
+  // First call succeeds.
+  auto statuses_and_responses = SendRpc(kMessage_, 1);
+  for (const auto& status_and_response : statuses_and_responses) {
+    const Status& status = status_and_response.first;
+    const EchoResponse& response = status_and_response.second;
+    EXPECT_TRUE(status.ok()) << "code=" << status.error_code()
+                             << " message=" << status.error_message();
+    EXPECT_EQ(response.message(), kMessage_);
+  }
+  // But eventually, the update with only dropped servers is processed and calls
+  // fail.
+  do {
+    statuses_and_responses = SendRpc(kMessage_, 1);
+    ASSERT_EQ(statuses_and_responses.size(), 1UL);
+  } while (statuses_and_responses[0].first.ok());
+  const Status& status = statuses_and_responses[0].first;
+  EXPECT_FALSE(status.ok());
+  EXPECT_EQ(status.error_message(), "Call dropped by load balancing policy");
+}
+
 class SingleBalancerWithClientLoadReportingTest : public GrpclbEnd2endTest {
  public:
   SingleBalancerWithClientLoadReportingTest() : GrpclbEnd2endTest(4, 1, 2) {}