nghttpx: Add --backend-fall and --backend-rise options

These options are analogous to fall and rise parameter found in
haproxy.

gennghttpxfun.py

@@ -129,6 +129,8 @@ OPTIONS = [
     "backend-tls",
     "backend-connections-per-host",
     "error-page",
+    "backend-fall",
+    "backend-rise",
 ]
 
 LOGVARS = [

src/shrpx.cc

@@ -1314,6 +1314,24 @@ Connections:
               --backend-write-timeout options.
   --accept-proxy-protocol
               Accept PROXY protocol version 1 on frontend connection.
+  --backend-fall=<N>
+              If  nghttpx cannot  connect  to a  specific backend  <N>
+              times in a  row, that backend is assumed  to be offline,
+              and  it  is  excluded  from load  balancing.   See  also
+              --backend-rise option.  If <N> is  0, a backend never be
+              excluded  from  load  balancing whatever  times  nghttpx
+              cannot connect to it.
+              Default: )" << get_config()->conn.downstream.fall << R"(
+  --backend-rise=<N>
+              As described  in --backend-fall,  a backend  is excluded
+              from  load  balancing  if  nghttpx assumes  that  it  is
+              offline.  Then  nghttpx periodically attempts to  make a
+              connection to the failed  backend, and if the connection
+              is made successfully <N> times  in a row, the backend is
+              assumed to  be online, and  it is now eligible  for load
+              balancing target.  If <N> is 0, a backend is permanently
+              offline, once it goes in that state.
+              Default: )" << get_config()->conn.downstream.rise << R"(
 
 Performance:
   -n, --workers=<N>
@@ -2519,6 +2537,8 @@ int main(int argc, char **argv) {
         {SHRPX_OPT_BACKEND_CONNECTIONS_PER_HOST.c_str(), required_argument,
          &flag, 121},
         {SHRPX_OPT_ERROR_PAGE.c_str(), required_argument, &flag, 122},
+        {SHRPX_OPT_BACKEND_FALL.c_str(), required_argument, &flag, 123},
+        {SHRPX_OPT_BACKEND_RISE.c_str(), required_argument, &flag, 124},
         {nullptr, 0, nullptr, 0}};
 
     int option_index = 0;
@@ -3094,6 +3114,14 @@ int main(int argc, char **argv) {
         // --error-page
         cmdcfgs.emplace_back(SHRPX_OPT_ERROR_PAGE, StringRef{optarg});
         break;
+      case 123:
+        // --backend-fall
+        cmdcfgs.emplace_back(SHRPX_OPT_BACKEND_FALL, StringRef{optarg});
+        break;
+      case 124:
+        // --backend-rise
+        cmdcfgs.emplace_back(SHRPX_OPT_BACKEND_RISE, StringRef{optarg});
+        break;
       default:
         break;
       }

src/shrpx_config.cc

@@ -893,6 +893,7 @@ enum {
   SHRPX_OPTID_BACKEND_ADDRESS_FAMILY,
   SHRPX_OPTID_BACKEND_CONNECTIONS_PER_FRONTEND,
   SHRPX_OPTID_BACKEND_CONNECTIONS_PER_HOST,
+  SHRPX_OPTID_BACKEND_FALL,
   SHRPX_OPTID_BACKEND_HTTP_PROXY_URI,
   SHRPX_OPTID_BACKEND_HTTP1_CONNECTIONS_PER_FRONTEND,
   SHRPX_OPTID_BACKEND_HTTP1_CONNECTIONS_PER_HOST,
@@ -908,6 +909,7 @@ enum {
   SHRPX_OPTID_BACKEND_READ_TIMEOUT,
   SHRPX_OPTID_BACKEND_REQUEST_BUFFER,
   SHRPX_OPTID_BACKEND_RESPONSE_BUFFER,
+  SHRPX_OPTID_BACKEND_RISE,
   SHRPX_OPTID_BACKEND_TLS,
   SHRPX_OPTID_BACKEND_TLS_SNI_FIELD,
   SHRPX_OPTID_BACKEND_WRITE_TIMEOUT,
@@ -1187,6 +1189,9 @@ int option_lookup_token(const char *name, size_t namelen) {
       }
       break;
     case 'e':
+      if (util::strieq_l("backend-ris", name, 11)) {
+        return SHRPX_OPTID_BACKEND_RISE;
+      }
       if (util::strieq_l("host-rewrit", name, 11)) {
         return SHRPX_OPTID_HOST_REWRITE;
       }
@@ -1194,6 +1199,11 @@ int option_lookup_token(const char *name, size_t namelen) {
         return SHRPX_OPTID_HTTP2_BRIDGE;
       }
       break;
+    case 'l':
+      if (util::strieq_l("backend-fal", name, 11)) {
+        return SHRPX_OPTID_BACKEND_FALL;
+      }
+      break;
     case 'y':
       if (util::strieq_l("client-prox", name, 11)) {
         return SHRPX_OPTID_CLIENT_PROXY;
@@ -2672,6 +2682,10 @@ int parse_config(const StringRef &opt, const StringRef &optarg,
                       opt, optarg);
   case SHRPX_OPTID_ERROR_PAGE:
     return parse_error_page(mod_config()->http.error_pages, opt, optarg);
+  case SHRPX_OPTID_BACKEND_FALL:
+    return parse_uint(&mod_config()->conn.downstream.fall, opt, optarg);
+  case SHRPX_OPTID_BACKEND_RISE:
+    return parse_uint(&mod_config()->conn.downstream.rise, opt, optarg);
   case SHRPX_OPTID_CONF:
     LOG(WARN) << "conf: ignored";
 

src/shrpx_config.h

@@ -275,6 +275,8 @@ constexpr auto SHRPX_OPT_BACKEND_TLS = StringRef::from_lit("backend-tls");
 constexpr auto SHRPX_OPT_BACKEND_CONNECTIONS_PER_HOST =
     StringRef::from_lit("backend-connections-per-host");
 constexpr auto SHRPX_OPT_ERROR_PAGE = StringRef::from_lit("error-page");
+constexpr auto SHRPX_OPT_BACKEND_FALL = StringRef::from_lit("backend-fall");
+constexpr auto SHRPX_OPT_BACKEND_RISE = StringRef::from_lit("backend-rise");
 
 constexpr size_t SHRPX_OBFUSCATED_NODE_LENGTH = 8;
 
@@ -615,6 +617,8 @@ struct ConnectionConfig {
     size_t connections_per_frontend;
     size_t request_buffer_size;
     size_t response_buffer_size;
+    size_t fall;
+    size_t rise;
     // Address family of backend connection.  One of either AF_INET,
     // AF_INET6 or AF_UNSPEC.  This is ignored if backend connection
     // is made via Unix domain socket.

src/shrpx_live_check.cc

@@ -294,7 +294,9 @@ void LiveCheck::on_success() {
   LOG(WARN) << "Liveness check for " << util::to_numeric_addr(&addr_->addr)
             << " succeeded " << success_count_ << " time(s) in a row";
 
-  if (success_count_ < 3) {
+  auto &downstreamconf = get_config()->conn.downstream;
+
+  if (success_count_ < downstreamconf.rise) {
     disconnect();
 
     schedule();

src/shrpx_worker.cc

@@ -475,12 +475,21 @@ void downstream_failure(DownstreamAddr *addr) {
 
   auto fail_count = connect_blocker->get_fail_count();
 
-  if (fail_count >= 3) {
+  auto &downstreamconf = get_config()->conn.downstream;
+
+  if (downstreamconf.fall == 0) {
+    return;
+  }
+
+  if (fail_count >= downstreamconf.fall) {
     LOG(WARN) << "Could not connect to " << util::to_numeric_addr(&addr->addr)
               << " " << fail_count << " times in a row; considered as offline";
 
     connect_blocker->offline();
-    addr->live_check->schedule();
+
+    if (downstreamconf.rise) {
+      addr->live_check->schedule();
+    }
   }
 }