Don't forget to test decommissioned nodes

oxen-io · Sep 6, 2019 · bab0ad5 · bab0ad5
1 parent d6633a2
commit bab0ad5
Show file tree

Hide file tree

Showing 5 changed files with 23 additions and 19 deletions.
diff --git a/httpserver/https_client.cpp b/httpserver/https_client.cpp
@@ -23,7 +23,7 @@ void make_https_request(boost::asio::io_context& ioc,
 #else
 
     if (sn_address == "0.0.0.0") {
-        LOKI_LOG(warn, "Could not initiate request to snode (we don't know "
+        LOKI_LOG(debug, "Could not initiate request to snode (we don't know "
                        "their IP yet).");
 
         cb(sn_response_t{SNodeError::NO_REACH, nullptr});

diff --git a/httpserver/reachability_testing.cpp b/httpserver/reachability_testing.cpp
@@ -21,31 +21,31 @@ constexpr std::chrono::minutes UNREACH_GRACE_PERIOD = 120min;
 
 bool reachability_records_t::record_unreachable(const sn_pub_key_t& sn) {
 
-    auto it = offline_nodes_.find(sn);
+    const auto it = offline_nodes_.find(sn);
 
     if (it == offline_nodes_.end()) {
-        LOKI_LOG(info, "adding a new node to UNREACHABLE: {}", sn);
+        /// TODO: change this to debug
+        LOKI_LOG(debug, "Adding a new node to UNREACHABLE: {}", sn);
         offline_nodes_.insert({sn, {}});
     } else {
-        LOKI_LOG(info, "node is ALREAY known to be UNREACHABLE: {}", sn);
+        LOKI_LOG(debug, "Node is ALREAY known to be UNREACHABLE: {}", sn);
 
         it->second.last_tested = steady_clock::now();
 
         const auto elapsed = it->second.last_tested - it->second.first_failure;
         const auto elapsed_sec =
             std::chrono::duration_cast<std::chrono::seconds>(elapsed).count();
-        LOKI_LOG(info, "    - first time failed {} seconds ago", elapsed_sec);
+        LOKI_LOG(debug, "First time failed {} seconds ago", elapsed_sec);
 
         /// TODO: Might still want to report as unreachable since this status
         /// gets reset to `true` on Lokid restart
-        if (elapsed > UNREACH_GRACE_PERIOD && !it->second.reported) {
-            LOKI_LOG(warn, "    - will REPORT this node to Lokid!");
+        if (it->second.reported) {
+            LOKI_LOG(debug, "Already reported node: {}", sn);
+        } else if (elapsed > UNREACH_GRACE_PERIOD) {
+            LOKI_LOG(debug, "Will REPORT this node to Lokid!");
             return true;
-        } else {
-            if (it->second.reported) {
-                LOKI_LOG(warn, "    - Already reported node: {}", sn);
-            }
         }
+
     }
 
     return false;
@@ -58,13 +58,13 @@ bool reachability_records_t::record_reachable(const sn_pub_key_t& sn) {
 bool reachability_records_t::expire(const sn_pub_key_t& sn) {
 
     if (offline_nodes_.erase(sn)) {
-        LOKI_LOG(warn, "    - removed entry for {}", sn);
+        LOKI_LOG(debug, "Removed entry for {}", sn);
     }
 }
 
 void reachability_records_t::set_reported(const sn_pub_key_t& sn) {
 
-    auto it = offline_nodes_.find(sn);
+    const auto it = offline_nodes_.find(sn);
     if (it != offline_nodes_.end()) {
         it->second.reported = true;
     }
@@ -82,7 +82,7 @@ boost::optional<sn_pub_key_t> reachability_records_t::next_to_test() {
         return boost::none;
     } else {
 
-        LOKI_LOG(warn, "~~~ Selecting to be re-tested: {}", it->first);
+        LOKI_LOG(debug, "Selecting to be re-tested: {}", it->first);
 
         return it->first;
     }

diff --git a/httpserver/reachability_testing.h b/httpserver/reachability_testing.h
@@ -11,12 +11,12 @@ namespace detail {
 /// TODO: make this class "private"?
 class reach_record_t {
 
-    // The time the node failed for the first time
-    // (and hasn't come back online)
 
     using time_point_t = std::chrono::time_point<std::chrono::steady_clock>;
 
   public:
+    // The time the node failed for the first time
+    // (and hasn't come back online)
     time_point_t first_failure;
     time_point_t last_tested;
     // whether it's been reported to Lokid

diff --git a/httpserver/service_node.cpp b/httpserver/service_node.cpp
@@ -1006,12 +1006,12 @@ void ServiceNode::report_node_reachability(const sn_pub_key_t& sn_pk,
 
     auto cb = [this, sn_pk, reachable](const sn_response_t&& res) {
         if (res.error_code != SNodeError::NO_ERROR) {
-            LOKI_LOG(error, "Could not report node status");
+            LOKI_LOG(warn, "Could not report node status");
             return;
         }
 
         if (!res.body) {
-            LOKI_LOG(error, "Empty body on Lokid report node status");
+            LOKI_LOG(warn, "Empty body on Lokid report node status");
             return;
         }
 
@@ -1026,7 +1026,7 @@ void ServiceNode::report_node_reachability(const sn_pub_key_t& sn_pk,
             if (status == "OK") {
                 success = true;
             } else {
-                LOKI_LOG(error, "Could not report node. Status: {}", status);
+                LOKI_LOG(warn, "Could not report node. Status: {}", status);
             }
         } catch (...) {
             LOKI_LOG(error,

diff --git a/httpserver/swarm.cpp b/httpserver/swarm.cpp
@@ -192,6 +192,10 @@ void Swarm::update_state(const all_swarms_t& swarms,
             all_funded_nodes_.push_back(sn);
         }
     }
+
+    for (const auto& sn : decommissioned) {
+        all_funded_nodes_.push_back(sn);
+    }
 }
 
 boost::optional<sn_record_t> Swarm::choose_funded_node() const {