storage_service: Replicate and advertise tokens early in the boot up process

When a node is restarted, there is a race between gossip starts (other
nodes will mark this node up again and send requests) and the tokens are
replicated to other shards. Here is an example:

- n1, n2
- n2 is down, n1 think n2 is down
- n2 starts again, n2 starts gossip service, n1 thinks n2 is up and sends
  reads/writes to n2, but n2 hasn't replicated the token_metadata to all
  the shards.
- n2 complains:
  token_metadata - sorted_tokens is empty in first_token_index!
  token_metadata - sorted_tokens is empty in first_token_index!
  token_metadata - sorted_tokens is empty in first_token_index!
  token_metadata - sorted_tokens is empty in first_token_index!
  token_metadata - sorted_tokens is empty in first_token_index!
  token_metadata - sorted_tokens is empty in first_token_index!
  storage_proxy - Failed to apply mutation from $ip#4: std::runtime_error
  (sorted_tokens is empty in first_token_index!)

The code path looks like below:

0 stoarge_service::init_server
1    prepare_to_join()
2          add gossip application state of NET_VERSION, SCHEMA and so on.
3         _gossiper.start_gossiping().get()
4    join_token_ring()
5           _token_metadata.update_normal_tokens(tokens, get_broadcast_address());
6           replicate_to_all_cores().get()
7           storage_service::set_gossip_tokens() which adds the gossip application state of TOKENS and STATUS

The race talked above is at line 3 and line 6.

To fix, we can replicate the token_metadata early after it is filled
with the tokens read from system table before gossip starts. So that
when other nodes think this restarting node is up, the tokens are
already replicated to all the shards.

In addition, this patch also fixes the issue that other nodes might see
a node miss the TOKENS and STATUS application state in gossip if that
node failed in the middle of a restarting process, i.e., it is killed
after line 3 and before line 7. As a result we could not replace the
node.

Tests: update_cluster_layout_tests.py
Fixes: #4709
Fixes: #4723
(cherry picked from commit 3b39a59135)

service/storage_service.cc

@@ -409,6 +409,14 @@ void storage_service::prepare_to_join(std::vector<inet_address> loaded_endpoints
         }
     }
 
+    // If this is a restarting node, we should update tokens before gossip starts
+    auto my_tokens = db::system_keyspace::get_saved_tokens().get0();
+    bool restarting_normal_node = db::system_keyspace::bootstrap_complete() && !db().local().is_replacing() && !my_tokens.empty();
+    if (restarting_normal_node) {
+        slogger.info("Restarting a node in NORMAL status");
+        _token_metadata.update_normal_tokens(my_tokens, get_broadcast_address());
+    }
+
     // have to start the gossip service before we can see any info on other nodes.  this is necessary
     // for bootstrap to get the load info it needs.
     // (we won't be part of the storage ring though until we add a counterId to our state, below.)
@@ -419,6 +427,12 @@ void storage_service::prepare_to_join(std::vector<inet_address> loaded_endpoints
     }).get();
     auto features = get_config_supported_features();
     _token_metadata.update_host_id(local_host_id, get_broadcast_address());
+
+    // Replicate the tokens early because once gossip runs other nodes
+    // might send reads/writes to this node. Replicate it early to make
+    // sure the tokens are valid on all the shards.
+    replicate_to_all_cores().get();
+
     auto broadcast_rpc_address = utils::fb_utilities::get_broadcast_rpc_address();
     app_states.emplace(gms::application_state::NET_VERSION, value_factory.network_version());
     app_states.emplace(gms::application_state::HOST_ID, value_factory.host_id(local_host_id));
@@ -429,6 +443,10 @@ void storage_service::prepare_to_join(std::vector<inet_address> loaded_endpoints
     app_states.emplace(gms::application_state::SCHEMA_TABLES_VERSION, versioned_value(db::schema_tables::version));
     app_states.emplace(gms::application_state::RPC_READY, value_factory.cql_ready(false));
     app_states.emplace(gms::application_state::VIEW_BACKLOG, versioned_value(""));
+    if (restarting_normal_node) {
+        app_states.emplace(gms::application_state::TOKENS, value_factory.tokens(my_tokens));
+        app_states.emplace(gms::application_state::STATUS, value_factory.normal(my_tokens));
+    }
     slogger.info("Starting up server gossip");
 
     auto& gossiper = gms::get_local_gossiper();
@@ -697,6 +715,7 @@ void storage_service::bootstrap(std::unordered_set<token> tokens) {
     } else {
         // Dont set any state for the node which is bootstrapping the existing token...
         _token_metadata.update_normal_tokens(tokens, get_broadcast_address());
+        replicate_to_all_cores().get();
         auto replace_addr = db().local().get_replace_address();
         if (replace_addr) {
             slogger.debug("Removing replaced endpoint {} from system.peers", *replace_addr);
@@ -1473,6 +1492,7 @@ future<> storage_service::init_server(int delay, bind_messaging_port do_bind) {
             auto tokens = db::system_keyspace::get_saved_tokens().get0();
             if (!tokens.empty()) {
                 _token_metadata.update_normal_tokens(tokens, get_broadcast_address());
+                replicate_to_all_cores().get();
                 // order is important here, the gossiper can fire in between adding these two states.  It's ok to send TOKENS without STATUS, but *not* vice versa.
                 gossiper.add_local_application_state({
                     { gms::application_state::TOKENS, value_factory.tokens(tokens) },