[#108] switch-to support

doc/GETTING_STARTED.md

@@ -149,6 +149,7 @@ Commands:
   cancel-shutdown          Cancel the graceful shutdown
   status                   Status of pgagroal
   details                  Detailed status of pgagroal
+  switch-to                Switch to another primary
   reset                    Reset the Prometheus statistics
   reset-server             Reset the state of a server
 ```

doc/man/pgagroal-cli.1.rst

@@ -83,6 +83,9 @@ status
 details
   Detailed status of pgagroal
 
+switch-to
+  Switch to another primary
+
 reset
   Reset the Prometheus statistics
 

src/cli.c

@@ -63,6 +63,7 @@
 #define ACTION_DISABLEDB       9
 #define ACTION_RESET          10
 #define ACTION_RESET_SERVER   11
+#define ACTION_SWITCH_TO      12
 
 static int flush(SSL* ssl, int socket, int32_t mode);
 static int enabledb(SSL* ssl, int socket, char* database);
@@ -75,6 +76,7 @@ static int details(SSL* ssl, int socket);
 static int isalive(SSL* ssl, int socket);
 static int reset(SSL* ssl, int socket);
 static int reset_server(SSL* ssl, int socket, char* server);
+static int switch_to(SSL* ssl, int socket, char* server);
 
 
 static void
@@ -116,6 +118,7 @@ usage()
    printf("  cancel-shutdown          Cancel the graceful shutdown\n");
    printf("  status                   Status of pgagroal\n");
    printf("  details                  Detailed status of pgagroal\n");
+   printf("  switch-to                Switch to another primary\n");
    printf("  reset                    Reset the Prometheus statistics\n");
    printf("  reset-server             Reset the state of a server\n");
    printf("\n");
@@ -322,6 +325,14 @@ main(int argc, char **argv)
             server = argv[argc - 1];
          }
       }
+      else if (!strcmp("switch-to", argv[argc - 1]) || !strcmp("switch-to", argv[argc - 2]))
+      {
+         if (!strcmp("switch-to", argv[argc - 2]))
+         {
+            action = ACTION_SWITCH_TO;
+            server = argv[argc - 1];
+         }
+      }
 
       if (action != ACTION_UNKNOWN)
       {
@@ -440,6 +451,10 @@ main(int argc, char **argv)
       {
          exit_code = reset_server(s_ssl, socket, server);
       }
+      else if (action == ACTION_SWITCH_TO)
+      {
+         exit_code = switch_to(s_ssl, socket, server);
+      }
    }
 
 done:
@@ -639,3 +654,14 @@ reset_server(SSL* ssl, int socket, char* server)
 
    return 0;
 }
+
+static int
+switch_to(SSL* ssl, int socket, char* server)
+{
+   if (pgagroal_management_switch_to(ssl, socket, server))
+   {
+      return 1;
+   }
+
+   return 0;
+}

src/include/management.h

@@ -56,6 +56,7 @@ extern "C" {
 #define MANAGEMENT_RESET_SERVER       14
 #define MANAGEMENT_CLIENT_DONE        15
 #define MANAGEMENT_CLIENT_FD          16
+#define MANAGEMENT_SWITCH_TO          17
 
 /**
  * Read the management header
@@ -282,6 +283,16 @@ pgagroal_management_client_done(void* shmem, pid_t pid);
 int
 pgagroal_management_client_fd(void* shmem, int32_t slot, pid_t pid);
 
+/**
+ * Management operation: Switch to
+ * @param ssl The SSL connection
+ * @param socket The socket
+ * @param server The server
+ * @return 0 upon success, otherwise 1
+ */
+int
+pgagroal_management_switch_to(SSL* ssl, int socket, char* server);
+
 #ifdef __cplusplus
 }
 #endif

src/include/server.h

@@ -91,6 +91,15 @@ pgagroal_server_force_failover(void* shmem, int server);
 int
 pgagroal_server_reset(void* shmem, char* server);
 
+/**
+ * Switch server
+ * @param shmem The shared memory segment
+ * @param server The server
+ * @return 0 upon success, otherwise 1
+ */
+int
+pgagroal_server_switch(void* shmem, char* server);
+
 #ifdef __cplusplus
 }
 #endif

src/libpgagroal/management.c

@@ -182,6 +182,7 @@ pgagroal_management_read_payload(int socket, signed char id, int* payload_i, cha
          *payload_s = s;
          break;
       case MANAGEMENT_RESET_SERVER:
+      case MANAGEMENT_SWITCH_TO:
          s = malloc(MISC_LENGTH);
          memset(s, 0, MISC_LENGTH);
          if (read_complete(NULL, socket, s, MISC_LENGTH))
@@ -1119,6 +1120,35 @@ pgagroal_management_client_fd(void* shmem, int32_t slot, pid_t pid)
    return 1;
 }
 
+int
+pgagroal_management_switch_to(SSL* ssl, int fd, char* server)
+{
+   char name[MISC_LENGTH];
+
+   if (write_header(ssl, fd, MANAGEMENT_SWITCH_TO, -1))
+   {
+      ZF_LOGW("pgagroal_management_switch_to: write: %d", fd);
+      errno = 0;
+      goto error;
+   }
+
+   memset(&name[0], 0, MISC_LENGTH);
+   memcpy(&name[0], server, strlen(server));
+
+   if (write_complete(ssl, fd, &name[0], sizeof(name)))
+   {
+      ZF_LOGW("pgagroal_management_switch_to: write: %d %s", fd, strerror(errno));
+      errno = 0;
+      goto error;
+   }
+
+   return 0;
+
+error:
+
+   return 1;
+}
+
 static int
 read_complete(SSL* ssl, int socket, void* buf, size_t size)
 {

src/libpgagroal/remote.c

@@ -113,6 +113,7 @@ pgagroal_remote_management(int client_fd, char* address, void* shmem, void* pipe
             break;
          case MANAGEMENT_FLUSH:
          case MANAGEMENT_RESET_SERVER:
+         case MANAGEMENT_SWITCH_TO:
             status = pgagroal_read_timeout_message(client_ssl, client_fd, config->authentication_timeout, &msg);
             if (status != MESSAGE_STATUS_OK)
             {

src/libpgagroal/server.c

@@ -289,6 +289,43 @@ pgagroal_server_reset(void* shmem, char* server)
    return 1;
 }
 
+int
+pgagroal_server_switch(void* shmem, char* server)
+{
+   int old_primary;
+   int new_primary;
+   signed char state;
+   struct configuration* config = NULL;
+
+   config = (struct configuration*)shmem;
+
+   old_primary = -1;
+   new_primary = -1;
+
+   for (int i = 0; i < config->number_of_servers; i++)
+   {
+      state = atomic_load(&config->servers[i].state);
+
+      if (state == SERVER_PRIMARY)
+      {
+         old_primary = i;
+      }
+      else if (!strcmp(config->servers[i].name, server))
+      {
+         new_primary = i;
+      }
+   }
+
+   if (old_primary != -1 && new_primary != -1)
+   {
+      atomic_store(&config->servers[old_primary].state, SERVER_FAILED);
+      atomic_store(&config->servers[new_primary].state, SERVER_PRIMARY);
+      return 0;
+   }
+
+   return 1;
+}
+
 static int
 failover(void* shmem, int old_primary)
 {

src/main.c

@@ -1217,6 +1217,17 @@ accept_mgt_cb(struct ev_loop *loop, struct ev_io *watcher, int revents)
             remove_client(p);
          }
          break;
+      case MANAGEMENT_SWITCH_TO:
+         ZF_LOGD("pgagroal: Management switch to");
+         if (!pgagroal_server_switch(ai->shmem, payload_s))
+         {
+            if (!fork())
+            {
+               pgagroal_flush(ai->shmem, FLUSH_GRACEFULLY);
+            }
+            pgagroal_prometheus_failed_servers(ai->shmem);
+         }
+         break;
       default:
          ZF_LOGD("pgagroal: Unknown management id: %d", id);
          break;