diskless replication rdb transfer uses pipe, and writes to sockets form the parent process.

misc:
- handle SSL_has_pending by iterating though these in beforeSleep, and setting timeout of 0 to aeProcessEvents
- fix issue with epoll signaling EPOLLHUP and EPOLLERR only to the write handlers. (needed to detect the rdb pipe was closed)
- add key-load-delay config for testing
- trim connShutdown which is no longer needed
- rioFdsetWrite -> rioFdWrite - simplified since there's no longer need to write to multiple FDs
- don't detect rdb child exited (don't call wait3) until we detect the pipe is closed
- Cleanup bad optimization from rio.c, add another one

17 files changed, 580 insertions, 353 deletions

diff --git a/TLS.md b/TLS.md
index ee24a8df5..90ed08e7c 100644
--- a/TLS.md
+++ b/TLS.md
@@ -81,23 +81,6 @@ implementation details between TLS and TCP.
    difficult, but there are probably other good reasons to improve that part
    anyway.
 
-5. A mechanism to re-trigger read callbacks for connections with unread buffers
-   (the case of reading partial TLS frames):
-
-    a) Before sleep should iterate connections looking for those with a read handler,
-       SSL_pending() != 0 and no read event.
-    b) If found, trigger read handler for these conns.
-    c) After iteration if this state persists, epoll should be called in a way
-       that won't block so the process continues and this behave the same as a
-       level trigerred epoll.
-
-Replication
------------
-
-Diskless master replication is broken, until child/parent connection proxying is
-implemented.
-
-
 TLS Features
 ------------
 
@@ -119,6 +102,10 @@ most actions.
 This will need to be cleaned up for proper TLS support. The best approach is
 probably to migrate to hiredis async mode.
 
+redis-cli
+---------
+1. Support tls in --slave and --rdb
+
 
 Others
 ------
diff --git a/src/ae.c b/src/ae.c
index 53629ef77..54c0d994e 100644
--- a/src/ae.c
+++ b/src/ae.c
@@ -76,6 +76,7 @@ aeEventLoop *aeCreateEventLoop(int setsize) {
     eventLoop->maxfd = -1;
     eventLoop->beforesleep = NULL;
     eventLoop->aftersleep = NULL;
+    eventLoop->flags = 0;
     if (aeApiCreate(eventLoop) == -1) goto err;
     /* Events with mask == AE_NONE are not set. So let's initialize the
      * vector with it. */
@@ -97,6 +98,14 @@ int aeGetSetSize(aeEventLoop *eventLoop) {
     return eventLoop->setsize;
 }
 
+/* Tells the next iteration/s of the event processing to set timeout of 0. */
+void aeDontWait(aeEventLoop *eventLoop, int noWait) {
+    if (noWait)
+        eventLoop->flags |= AE_DONT_WAIT;
+    else
+        eventLoop->flags &= ~AE_DONT_WAIT;
+}
+
 /* Resize the maximum set size of the event loop.
  * If the requested set size is smaller than the current set size, but
  * there is already a file descriptor in use that is >= the requested
@@ -406,6 +415,11 @@ int aeProcessEvents(aeEventLoop *eventLoop, int flags)
             }
         }
 
+        if (eventLoop->flags & AE_DONT_WAIT) {
+            tv.tv_sec = tv.tv_usec = 0;
+            tvp = &tv;
+        }
+
         /* Call the multiplexing API, will return only on timeout or when
          * some event fires. */
         numevents = aeApiPoll(eventLoop, tvp);
diff --git a/src/ae.h b/src/ae.h
index 184fe3d1b..bbb43fe6e 100644
--- a/src/ae.h
+++ b/src/ae.h
@@ -106,6 +106,7 @@ typedef struct aeEventLoop {
     void *apidata; /* This is used for polling API specific data */
     aeBeforeSleepProc *beforesleep;
     aeBeforeSleepProc *aftersleep;
+    int flags;
 } aeEventLoop;
 
 /* Prototypes */
@@ -128,5 +129,6 @@ void aeSetBeforeSleepProc(aeEventLoop *eventLoop, aeBeforeSleepProc *beforesleep
 void aeSetAfterSleepProc(aeEventLoop *eventLoop, aeBeforeSleepProc *aftersleep);
 int aeGetSetSize(aeEventLoop *eventLoop);
 int aeResizeSetSize(aeEventLoop *eventLoop, int setsize);
+void aeDontWait(aeEventLoop *eventLoop, int noWait);
 
 #endif
diff --git a/src/ae_epoll.c b/src/ae_epoll.c
index 410aac70d..fa197297e 100644
--- a/src/ae_epoll.c
+++ b/src/ae_epoll.c
@@ -121,8 +121,8 @@ static int aeApiPoll(aeEventLoop *eventLoop, struct timeval *tvp) {
 
             if (e->events & EPOLLIN) mask |= AE_READABLE;
             if (e->events & EPOLLOUT) mask |= AE_WRITABLE;
-            if (e->events & EPOLLERR) mask |= AE_WRITABLE;
-            if (e->events & EPOLLHUP) mask |= AE_WRITABLE;
+            if (e->events & EPOLLERR) mask |= AE_WRITABLE|AE_READABLE;
+            if (e->events & EPOLLHUP) mask |= AE_WRITABLE|AE_READABLE;
             eventLoop->fired[j].fd = e->data.fd;
             eventLoop->fired[j].mask = mask;
         }
diff --git a/src/aof.c b/src/aof.c
index eed994bf2..ef34097ed 100644
--- a/src/aof.c
+++ b/src/aof.c
@@ -836,6 +836,8 @@ int loadAppendOnlyFile(char *filename) {
         freeFakeClientArgv(fakeClient);
         fakeClient->cmd = NULL;
         if (server.aof_load_truncated) valid_up_to = ftello(fp);
+        if (server.key_load_delay)
+            usleep(server.key_load_delay);
     }
 
     /* This point can only be reached when EOF is reached without errors.
diff --git a/src/config.c b/src/config.c
index 456fb0226..792e45057 100644
--- a/src/config.c
+++ b/src/config.c
@@ -522,6 +522,12 @@ void loadServerConfigFromString(char *config) {
                 err = "rdb-key-save-delay can't be negative";
                 goto loaderr;
             }
+        } else if (!strcasecmp(argv[0],"key-load-delay") && argc==2) {
+            server.key_load_delay = atoi(argv[1]);
+            if (server.key_load_delay < 0) {
+                err = "key-load-delay can't be negative";
+                goto loaderr;
+            }
         } else if (!strcasecmp(argv[0],"requirepass") && argc == 2) {
             if (strlen(argv[1]) > CONFIG_AUTHPASS_MAX_LEN) {
                 err = "Password is longer than CONFIG_AUTHPASS_MAX_LEN";
@@ -1192,6 +1198,8 @@ void configSetCommand(client *c) {
     } config_set_numerical_field(
       "rdb-key-save-delay",server.rdb_key_save_delay,0,LLONG_MAX) {
     } config_set_numerical_field(
+      "key-load-delay",server.key_load_delay,0,LLONG_MAX) {
+    } config_set_numerical_field(
       "slave-announce-port",server.slave_announce_port,0,65535) {
     } config_set_numerical_field(
       "replica-announce-port",server.slave_announce_port,0,65535) {
@@ -1452,6 +1460,7 @@ void configGetCommand(client *c) {
     config_get_numerical_field("cluster-replica-validity-factor",server.cluster_slave_validity_factor);
     config_get_numerical_field("repl-diskless-sync-delay",server.repl_diskless_sync_delay);
     config_get_numerical_field("rdb-key-save-delay",server.rdb_key_save_delay);
+    config_get_numerical_field("key-load-delay",server.key_load_delay);
     config_get_numerical_field("tcp-keepalive",server.tcpkeepalive);
 
     /* Bool (yes/no) values */
@@ -2272,6 +2281,7 @@ int rewriteConfig(char *path) {
     rewriteConfigNumericalOption(state,"hz",server.config_hz,CONFIG_DEFAULT_HZ);
     rewriteConfigEnumOption(state,"supervised",server.supervised_mode,supervised_mode_enum,SUPERVISED_NONE);
     rewriteConfigNumericalOption(state,"rdb-key-save-delay",server.rdb_key_save_delay,CONFIG_DEFAULT_RDB_KEY_SAVE_DELAY);
+    rewriteConfigNumericalOption(state,"key-load-delay",server.key_load_delay,CONFIG_DEFAULT_KEY_LOAD_DELAY);
     rewriteConfigStringOption(state,"tls-cert-file",server.tls_cert_file,NULL);
     rewriteConfigStringOption(state,"tls-key-file",server.tls_key_file,NULL);
     rewriteConfigStringOption(state,"tls-dh-params-file",server.tls_dh_params_file,NULL);
diff --git a/src/connection.c b/src/connection.c
index 62c6f3506..601f175bc 100644
--- a/src/connection.c
+++ b/src/connection.c
@@ -140,10 +140,6 @@ void *connGetPrivateData(connection *conn) {
  * move here as we implement additional connection types.
  */
 
-static int connSocketShutdown(connection *conn, int how) {
-    return shutdown(conn->fd, how);
-}
-
 /* Close the connection and free resources. */
 static void connSocketClose(connection *conn) {
     if (conn->fd != -1) {
@@ -298,7 +294,6 @@ static ssize_t connSocketSyncReadLine(connection *conn, char *ptr, ssize_t size,
 ConnectionType CT_Socket = {
     .ae_handler = connSocketEventHandler,
     .close = connSocketClose,
-    .shutdown = connSocketShutdown,
     .write = connSocketWrite,
     .read = connSocketRead,
     .accept = connSocketAccept,
diff --git a/src/connection.h b/src/connection.h
index e3e844f95..3fdcddebe 100644
--- a/src/connection.h
+++ b/src/connection.h
@@ -55,7 +55,6 @@ typedef struct ConnectionType {
     int (*connect)(struct connection *conn, const char *addr, int port, const char *source_addr, ConnectionCallbackFunc connect_handler);
     int (*write)(struct connection *conn, const void *data, size_t data_len);
     int (*read)(struct connection *conn, void *buf, size_t buf_len);
-    int (*shutdown)(struct connection *conn, int how);
     void (*close)(struct connection *conn);
     int (*accept)(struct connection *conn, ConnectionCallbackFunc accept_handler);
     int (*set_write_handler)(struct connection *conn, ConnectionCallbackFunc handler);
@@ -159,10 +158,6 @@ static inline void connClose(connection *conn) {
     conn->type->close(conn);
 }
 
-static inline int connShutdown(connection *conn, int how) {
-    return conn->type->shutdown(conn, how);
-}
-
 /* Returns the last error encountered by the connection, as a string.  If no error,
  * a NULL is returned.
  */
@@ -208,4 +203,8 @@ int connFormatPeer(connection *conn, char *buf, size_t buf_len);
 int connSockName(connection *conn, char *ip, size_t ip_len, int *port);
 const char *connGetInfo(connection *conn, char *buf, size_t buf_len);
 
+/* Helpers for tls special considerations */
+int tlsHasPendingData();
+void tlsProcessPendingData();
+
 #endif  /* __REDIS_CONNECTION_H */
diff --git a/src/networking.c b/src/networking.c
index 2d00b0e74..d52eb1eba 100644
--- a/src/networking.c
+++ b/src/networking.c
@@ -881,7 +881,7 @@ static void acceptCommonHandler(connection *conn, int flags, char *ip) {
         serverLog(LL_WARNING,
                 "Error accepting a client connection: %s (conn: %s)",
                 connGetLastError(conn), connGetInfo(conn, conninfo, sizeof(conninfo)));
-        connClose(conn);
+        freeClient(connGetPrivateData(conn));
         return;
     }
 }
@@ -984,14 +984,21 @@ void unlinkClient(client *c) {
             c->client_list_node = NULL;
         }
 
-        /* In the case of diskless replication the fork is writing to the
-         * sockets and just closing the fd isn't enough, if we don't also
-         * shutdown the socket the fork will continue to write to the slave
-         * and the salve will only find out that it was disconnected when
-         * it will finish reading the rdb. */
-        int need_shutdown = ((c->flags & CLIENT_SLAVE) &&
-                (c->replstate == SLAVE_STATE_WAIT_BGSAVE_END));
-        if (need_shutdown) connShutdown(c->conn, SHUT_RDWR);
+        /* Check if this is a replica waiting for diskless replication (rdb pipe),
+         * in which case it needs to be cleaned from that list */
+        if (c->flags & CLIENT_SLAVE &&
+            c->replstate == SLAVE_STATE_WAIT_BGSAVE_END &&
+            server.rdb_pipe_conns)
+        {
+            int i;
+            for (i=0; i < server.rdb_pipe_numconns; i++) {
+                if (server.rdb_pipe_conns[i] == c->conn) {
+                    rdbPipeWriteHandlerConnRemoved(c->conn);
+                    server.rdb_pipe_conns[i] = NULL;
+                    break;
+                }
+            }
+        }
         connClose(c->conn);
         c->conn = NULL;
     }
@@ -1309,7 +1316,7 @@ int handleClientsWithPendingWrites(void) {
             {
                 ae_flags |= AE_BARRIER;
             }
-            /* TODO: Handle write barriers in connection */
+            /* TODO: Handle write barriers in connection (also see tlsProcessPendingData) */
             if (connSetWriteHandler(c->conn, sendReplyToClient) == C_ERR) {
                 freeClientAsync(c);
             }
diff --git a/src/rdb.c b/src/rdb.c
index 3b98f57bb..59f22916b 100644
--- a/src/rdb.c
+++ b/src/rdb.c
@@ -2211,6 +2211,8 @@ int rdbLoadRio(rio *rdb, rdbSaveInfo *rsi, int loading_aof) {
              * own reference. */
             decrRefCount(key);
         }
+        if (server.key_load_delay)
+            usleep(server.key_load_delay);
 
         /* Reset the state that is key-specified and is populated by
          * opcodes before the key, so that we start from scratch again. */
@@ -2306,8 +2308,6 @@ void backgroundSaveDoneHandlerDisk(int exitcode, int bysignal) {
  * This function covers the case of RDB -> Salves socket transfers for
  * diskless replication. */
 void backgroundSaveDoneHandlerSocket(int exitcode, int bysignal) {
-    uint64_t *ok_slaves;
-
     if (!bysignal && exitcode == 0) {
         serverLog(LL_NOTICE,
             "Background RDB transfer terminated with success");
@@ -2321,79 +2321,6 @@ void backgroundSaveDoneHandlerSocket(int exitcode, int bysignal) {
     server.rdb_child_type = RDB_CHILD_TYPE_NONE;
     server.rdb_save_time_start = -1;
 
-    /* If the child returns an OK exit code, read the set of slave client
-     * IDs and the associated status code. We'll terminate all the slaves
-     * in error state.
-     *
-     * If the process returned an error, consider the list of slaves that
-     * can continue to be empty, so that it's just a special case of the
-     * normal code path. */
-    ok_slaves = zmalloc(sizeof(uint64_t)); /* Make space for the count. */
-    ok_slaves[0] = 0;
-    if (!bysignal && exitcode == 0) {
-        int readlen = sizeof(uint64_t);
-
-        if (read(server.rdb_pipe_read_result_from_child, ok_slaves, readlen) ==
-                 readlen)
-        {
-            readlen = ok_slaves[0]*sizeof(uint64_t)*2;
-
-            /* Make space for enough elements as specified by the first
-             * uint64_t element in the array. */
-            ok_slaves = zrealloc(ok_slaves,sizeof(uint64_t)+readlen);
-            if (readlen &&
-                read(server.rdb_pipe_read_result_from_child, ok_slaves+1,
-                     readlen) != readlen)
-            {
-                ok_slaves[0] = 0;
-            }
-        }
-    }
-
-    close(server.rdb_pipe_read_result_from_child);
-    close(server.rdb_pipe_write_result_to_parent);
-
-    /* We can continue the replication process with all the slaves that
-     * correctly received the full payload. Others are terminated. */
-    listNode *ln;
-    listIter li;
-
-    listRewind(server.slaves,&li);
-    while((ln = listNext(&li))) {
-        client *slave = ln->value;
-
-        if (slave->replstate == SLAVE_STATE_WAIT_BGSAVE_END) {
-            uint64_t j;
-            int errorcode = 0;
-
-            /* Search for the slave ID in the reply. In order for a slave to
-             * continue the replication process, we need to find it in the list,
-             * and it must have an error code set to 0 (which means success). */
-            for (j = 0; j < ok_slaves[0]; j++) {
-                if (slave->id == ok_slaves[2*j+1]) {
-                    errorcode = ok_slaves[2*j+2];
-                    break; /* Found in slaves list. */
-                }
-            }
-            if (j == ok_slaves[0] || errorcode != 0) {
-                serverLog(LL_WARNING,
-                "Closing slave %s: child->slave RDB transfer failed: %s",
-                    replicationGetSlaveName(slave),
-                    (errorcode == 0) ? "RDB transfer child aborted"
-                                     : strerror(errorcode));
-                freeClient(slave);
-            } else {
-                serverLog(LL_WARNING,
-                "Slave %s correctly received the streamed RDB file.",
-                    replicationGetSlaveName(slave));
-                /* Restore the socket as non-blocking. */
-                connNonBlock(slave->conn);
-                connSendTimeout(slave->conn,0);
-            }
-        }
-    }
-    zfree(ok_slaves);
-
     updateSlavesWaitingBgsave((!bysignal && exitcode == 0) ? C_OK : C_ERR, RDB_CHILD_TYPE_SOCKET);
 }
 
@@ -2425,9 +2352,6 @@ void killRDBChild(void) {
 /* Spawn an RDB child that writes the RDB to the sockets of the slaves
  * that are currently in SLAVE_STATE_WAIT_BGSAVE_START state. */
 int rdbSaveToSlavesSockets(rdbSaveInfo *rsi) {
-    connection **conns;
-    uint64_t *clientids;
-    int numconns;
     listNode *ln;
     listIter li;
     pid_t childpid;
@@ -2436,35 +2360,30 @@ int rdbSaveToSlavesSockets(rdbSaveInfo *rsi) {
 
     if (server.aof_child_pid != -1 || server.rdb_child_pid != -1) return C_ERR;
 
-    /* Before to fork, create a pipe that will be used in order to
-     * send back to the parent the IDs of the slaves that successfully
-     * received all the writes. */
+    /* Even if the previous fork child exited, don't start a new one until we
+     * drained the pipe. */
+    if (server.rdb_pipe_conns) return C_ERR;
+
+    /* Before to fork, create a pipe that is used to transfer the rdb bytes to
+     * the parant, we can't let it write directly to the sockets, since in case
+     * of TLS we must let the parent handle a contineous TLS state when the
+     * child terminates and parent takes over. */
     if (pipe(pipefds) == -1) return C_ERR;
-    server.rdb_pipe_read_result_from_child = pipefds[0];
-    server.rdb_pipe_write_result_to_parent = pipefds[1];
+    server.rdb_pipe_read = pipefds[0];
+    server.rdb_pipe_write = pipefds[1];
+    anetNonBlock(NULL, server.rdb_pipe_read);
 
-    /* Collect the file descriptors of the slaves we want to transfer
+    /* Collect the connections of the replicas we want to transfer
      * the RDB to, which are i WAIT_BGSAVE_START state. */
-    conns = zmalloc(sizeof(connection *)*listLength(server.slaves));
-    /* We also allocate an array of corresponding client IDs. This will
-     * be useful for the child process in order to build the report
-     * (sent via unix pipe) that will be sent to the parent. */
-    clientids = zmalloc(sizeof(uint64_t)*listLength(server.slaves));
-    numconns = 0;
-
+    server.rdb_pipe_conns = zmalloc(sizeof(connection *)*listLength(server.slaves));
+    server.rdb_pipe_numconns = 0;
+    server.rdb_pipe_numconns_writing = 0;
     listRewind(server.slaves,&li);
     while((ln = listNext(&li))) {
         client *slave = ln->value;
-
         if (slave->replstate == SLAVE_STATE_WAIT_BGSAVE_START) {
-            clientids[numconns] = slave->id;
-            conns[numconns++] = slave->conn;
+            server.rdb_pipe_conns[server.rdb_pipe_numconns++] = slave->conn;
             replicationSetupSlaveForFullResync(slave,getPsyncInitialOffset());
-            /* Put the socket in blocking mode to simplify RDB transfer.
-             * We'll restore it when the children returns (since duped socket
-             * will share the O_NONBLOCK attribute with the parent). */
-            connBlock(slave->conn);
-            connSendTimeout(slave->conn,server.repl_timeout*1000);
         }
     }
 
@@ -2474,16 +2393,15 @@ int rdbSaveToSlavesSockets(rdbSaveInfo *rsi) {
     if ((childpid = fork()) == 0) {
         /* Child */
         int retval;
-        rio slave_sockets;
+        rio rdb;
 
-        rioInitWithConnset(&slave_sockets,conns,numconns);
-        zfree(conns);
+        rioInitWithFd(&rdb,server.rdb_pipe_write);
 
         closeListeningSockets(0);
         redisSetProcTitle("redis-rdb-to-slaves");
 
-        retval = rdbSaveRioWithEOFMark(&slave_sockets,NULL,rsi);
-        if (retval == C_OK && rioFlush(&slave_sockets) == 0)
+        retval = rdbSaveRioWithEOFMark(&rdb,NULL,rsi);
+        if (retval == C_OK && rioFlush(&rdb) == 0)
             retval = C_ERR;
 
         if (retval == C_OK) {
@@ -2497,48 +2415,9 @@ int rdbSaveToSlavesSockets(rdbSaveInfo *rsi) {
 
             server.child_info_data.cow_size = private_dirty;
             sendChildInfo(CHILD_INFO_TYPE_RDB);
-
-            /* If we are returning OK, at least one slave was served
-             * with the RDB file as expected, so we need to send a report
-             * to the parent via the pipe. The format of the message is:
-             *
-             * <len> <slave[0].id> <slave[0].error> ...
-             *
-             * len, slave IDs, and slave errors, are all uint64_t integers,
-             * so basically the reply is composed of 64 bits for the len field
-             * plus 2 additional 64 bit integers for each entry, for a total
-             * of 'len' entries.
-             *
-             * The 'id' represents the slave's client ID, so that the master
-             * can match the report with a specific slave, and 'error' is
-             * set to 0 if the replication process terminated with a success
-             * or the error code if an error occurred. */
-            void *msg = zmalloc(sizeof(uint64_t)*(1+2*numconns));
-            uint64_t *len = msg;
-            uint64_t *ids = len+1;
-            int j, msglen;
-
-            *len = numconns;
-            for (j = 0; j < numconns; j++) {
-                *ids++ = clientids[j];
-                *ids++ = slave_sockets.io.connset.state[j];
-            }
-
-            /* Write the message to the parent. If we have no good slaves or
-             * we are unable to transfer the message to the parent, we exit
-             * with an error so that the parent will abort the replication
-             * process with all the childre that were waiting. */
-            msglen = sizeof(uint64_t)*(1+2*numconns);
-            if (*len == 0 ||
-                write(server.rdb_pipe_write_result_to_parent,msg,msglen)
-                != msglen)
-            {
-                retval = C_ERR;
-            }
-            zfree(msg);
         }
-        zfree(clientids);
-        rioFreeConnset(&slave_sockets);
+        rioFreeFd(&rdb);
+        close(server.rdb_pipe_write); /* wake up the reader, tell it we're done. */
         exitFromChild((retval == C_OK) ? 0 : 1);
     } else {
         /* Parent */
@@ -2552,17 +2431,16 @@ int rdbSaveToSlavesSockets(rdbSaveInfo *rsi) {
             listRewind(server.slaves,&li);
             while((ln = listNext(&li))) {
                 client *slave = ln->value;
-                int j;
-
-                for (j = 0; j < numconns; j++) {
-                    if (slave->id == clientids[j]) {
-                        slave->replstate = SLAVE_STATE_WAIT_BGSAVE_START;
-                        break;
-                    }
+                if (slave->replstate == SLAVE_STATE_WAIT_BGSAVE_END) {
+                    slave->replstate = SLAVE_STATE_WAIT_BGSAVE_START;
                 }
             }
-            close(pipefds[0]);
-            close(pipefds[1]);
+            close(server.rdb_pipe_write);
+            close(server.rdb_pipe_read);
+            zfree(server.rdb_pipe_conns);
+            server.rdb_pipe_conns = NULL;
+            server.rdb_pipe_numconns = 0;
+            server.rdb_pipe_numconns_writing = 0;
             closeChildInfoPipe();
         } else {
             server.stat_fork_time = ustime()-start;
@@ -2574,10 +2452,12 @@ int rdbSaveToSlavesSockets(rdbSaveInfo *rsi) {
             server.rdb_save_time_start = time(NULL);
             server.rdb_child_pid = childpid;
             server.rdb_child_type = RDB_CHILD_TYPE_SOCKET;
+            close(server.rdb_pipe_write); /* close write in parent so that it can detect the close on the child. */
+            if (aeCreateFileEvent(server.el, server.rdb_pipe_read, AE_READABLE, rdbPipeReadHandler,NULL) == AE_ERR) {
+                serverPanic("Unrecoverable error creating server.rdb_pipe_read file event.");
+            }
             updateDictResizePolicy();
         }
-        zfree(clientids);
-        zfree(conns);
         return (childpid == -1) ? C_ERR : C_OK;
     }
     return C_OK; /* Unreached. */
diff --git a/src/replication.c b/src/replication.c
index cbc52cb2e..9f5c19053 100644
--- a/src/replication.c
+++ b/src/replication.c
@@ -884,7 +884,7 @@ void sendBulkToSlave(connection *conn) {
         nwritten = connWrite(conn,slave->replpreamble,sdslen(slave->replpreamble));
         if (nwritten == -1) {
             serverLog(LL_VERBOSE,"Write error sending RDB preamble to replica: %s",
-                strerror(errno));
+                connGetLastError(conn));
             freeClient(slave);
             return;
         }
@@ -911,7 +911,7 @@ void sendBulkToSlave(connection *conn) {
     if ((nwritten = connWrite(conn,buf,buflen)) == -1) {
         if (connGetState(conn) != CONN_STATE_CONNECTED) {
             serverLog(LL_WARNING,"Write error sending DB to replica: %s",
-                strerror(errno));
+                connGetLastError(conn));
             freeClient(slave);
         }
         return;
@@ -926,6 +926,152 @@ void sendBulkToSlave(connection *conn) {
     }
 }
 
+/* Remove one write handler from the list of connections waiting to be writable
+ * during rdb pipe transfer. */
+void rdbPipeWriteHandlerConnRemoved(struct connection *conn) {
+    if (!connHasWriteHandler(conn))
+        return;
+    connSetWriteHandler(conn, NULL);
+    server.rdb_pipe_numconns_writing--;
+    /* if there are no more writes for now for this conn, or write error: */
+    if (server.rdb_pipe_numconns_writing == 0) {
+        if (aeCreateFileEvent(server.el, server.rdb_pipe_read, AE_READABLE, rdbPipeReadHandler,NULL) == AE_ERR) {
+            serverPanic("Unrecoverable error creating server.rdb_pipe_read file event.");
+        }
+    }
+}
+
+/* Called in diskless master during transfer of data from the rdb pipe, when
+ * the replica becomes writable again. */
+void rdbPipeWriteHandler(struct connection *conn) {
+    serverAssert(server.rdb_pipe_bufflen>0);
+    client *slave = connGetPrivateData(conn);
+    int nwritten;
+    if ((nwritten = connWrite(conn, server.rdb_pipe_buff + slave->repldboff,
+                              server.rdb_pipe_bufflen - slave->repldboff)) == -1)
+    {
+        if (connGetState(conn) == CONN_STATE_CONNECTED)
+            return; /* equivalent to EAGAIN */
+        serverLog(LL_WARNING,"Write error sending DB to replica: %s",
+            connGetLastError(conn));
+        freeClient(slave);
+        return;
+    } else {
+        slave->repldboff += nwritten;
+        server.stat_net_output_bytes += nwritten;
+        if (slave->repldboff < server.rdb_pipe_bufflen)
+            return; /* more data to write.. */
+    }
+    rdbPipeWriteHandlerConnRemoved(conn);
+}
+
+/* When the the pipe serving diskless rdb transfer is drained (write end was
+ * closed), we can clean up all the temporary variables, and cleanup after the
+ * fork child. */
+void RdbPipeCleanup() {
+    close(server.rdb_pipe_read);
+    zfree(server.rdb_pipe_conns);
+    server.rdb_pipe_conns = NULL;
+    server.rdb_pipe_numconns = 0;
+    server.rdb_pipe_numconns_writing = 0;
+    zfree(server.rdb_pipe_buff);
+    server.rdb_pipe_buff = NULL;
+    server.rdb_pipe_bufflen = 0;
+
+    /* Since we're avoiding to detect the child exited as long as the pipe is
+     * not drained, so now is the time to check. */
+    checkChildrenDone();
+}
+
+/* Called in diskless master, when there's data to read from the child's rdb pipe */
+void rdbPipeReadHandler(struct aeEventLoop *eventLoop, int fd, void *clientData, int mask) {
+    UNUSED(mask);
+    UNUSED(clientData);
+    UNUSED(eventLoop);
+    int i;
+    if (!server.rdb_pipe_buff)
+        server.rdb_pipe_buff = zmalloc(PROTO_IOBUF_LEN);
+    serverAssert(server.rdb_pipe_numconns_writing==0);
+
+    while (1) {
+        server.rdb_pipe_bufflen = read(fd, server.rdb_pipe_buff, PROTO_IOBUF_LEN);
+        if (server.rdb_pipe_bufflen < 0) {
+            if (errno == EAGAIN || errno == EWOULDBLOCK)
+                return;
+            serverLog(LL_WARNING,"Diskless rdb transfer, read error sending DB to replicas: %s", strerror(errno));
+            for (i=0; i < server.rdb_pipe_numconns; i++) {
+                connection *conn = server.rdb_pipe_conns[i];
+                if (!conn)
+                    continue;
+                client *slave = connGetPrivateData(conn);
+                freeClient(slave);
+                server.rdb_pipe_conns[i] = NULL;
+            }
+            killRDBChild();
+            return;
+        }
+
+        if (server.rdb_pipe_bufflen == 0) {
+            /* EOF - write end was closed. */
+            int stillUp = 0;
+            aeDeleteFileEvent(server.el, server.rdb_pipe_read, AE_READABLE);
+            for (i=0; i < server.rdb_pipe_numconns; i++)
+            {
+                connection *conn = server.rdb_pipe_conns[i];
+                if (!conn)
+                    continue;
+                stillUp++;
+            }
+            serverLog(LL_WARNING,"Diskless rdb transfer, done reading from pipe, %d replicas still up.", stillUp);
+            RdbPipeCleanup();
+            return;
+        }
+
+        int stillAlive = 0;
+        for (i=0; i < server.rdb_pipe_numconns; i++)
+        {
+            int nwritten;
+            connection *conn = server.rdb_pipe_conns[i];
+            if (!conn)
+                continue;
+
+            client *slave = connGetPrivateData(conn);
+            if ((nwritten = connWrite(conn, server.rdb_pipe_buff, server.rdb_pipe_bufflen)) == -1) {
+                if (connGetState(conn) != CONN_STATE_CONNECTED) {
+                    serverLog(LL_WARNING,"Diskless rdb transfer, write error sending DB to replica: %s",
+                        connGetLastError(conn));
+                    freeClient(slave);
+                    server.rdb_pipe_conns[i] = NULL;
+                    continue;
+                }
+                /* An error and still in connected state, is equivalent to EAGAIN */
+                slave->repldboff = 0;
+            } else {
+                slave->repldboff = nwritten;
+                server.stat_net_output_bytes += nwritten;
+            }
+            /* If we were unable to write all the data to one of the replicas,
+             * setup write handler (and disable pipe read handler, below) */
+            if (nwritten != server.rdb_pipe_bufflen) {
+                server.rdb_pipe_numconns_writing++;
+                connSetWriteHandler(conn, rdbPipeWriteHandler);
+            }
+            stillAlive++;
+        }
+
+        if (stillAlive == 0) {
+            serverLog(LL_WARNING,"Diskless rdb transfer, last replica dropped, killing fork child.");
+            killRDBChild();
+            RdbPipeCleanup();
+        }
+        /*  Remove the pipe read handler if at least one write handler was set. */
+        if (server.rdb_pipe_numconns_writing || stillAlive == 0) {
+            aeDeleteFileEvent(server.el, server.rdb_pipe_read, AE_READABLE);
+            break;
+        }
+    }
+}
+
 /* This function is called at the end of every background saving,
  * or when the replication RDB transfer strategy is modified from
  * disk to socket or the other way around.
diff --git a/src/rio.c b/src/rio.c
index 5e3b4a06e..c8c924380 100644
--- a/src/rio.c
+++ b/src/rio.c
@@ -272,85 +272,67 @@ void rioFreeConn(rio *r, sds *remaining) {
     r->io.conn.buf = NULL;
 }
 
-/* ------------------- File descriptors set implementation ------------------
- * This target is used to write the RDB file to N different replicas via
- * sockets, when the master just streams the data to the replicas without
- * creating an RDB on-disk image (diskless replication option).
+/* ------------------- File descriptor implementation ------------------
+ * This target is used to write the RDB file to pipe, when the master just
+ * streams the data to the replicas without creating an RDB on-disk image
+ * (diskless replication option).
  * It only implements writes. */
 
 /* Returns 1 or 0 for success/failure.
- * The function returns success as long as we are able to correctly write
- * to at least one file descriptor.
  *
  * When buf is NULL and len is 0, the function performs a flush operation
  * if there is some pending buffer, so this function is also used in order
- * to implement rioFdsetFlush(). */
-static size_t rioFdsetWrite(rio *r, const void *buf, size_t len) {
+ * to implement rioFdFlush(). */
+static size_t rioFdWrite(rio *r, const void *buf, size_t len) {
     ssize_t retval;
-    int j;
     unsigned char *p = (unsigned char*) buf;
     int doflush = (buf == NULL && len == 0);
 
-    /* To start we always append to our buffer. If it gets larger than
-     * a given size, we actually write to the sockets. */
-    if (len) {
-        r->io.connset.buf = sdscatlen(r->io.connset.buf,buf,len);
-        len = 0; /* Prevent entering the while below if we don't flush. */
-        if (sdslen(r->io.connset.buf) > PROTO_IOBUF_LEN) doflush = 1;
-    }
-
-    if (doflush) {
-        p = (unsigned char*) r->io.connset.buf;
-        len = sdslen(r->io.connset.buf);
+    /* For small writes, we rather keep the data in user-space buffer, and flush
+     * it only when it grows. however for larger writes, we prefer to flush
+     * any pre-existing buffer, and write the new one directly without reallocs
+     * and memory copying. */
+    if (len > PROTO_IOBUF_LEN) {
+        /* First, flush any pre-existing buffered data. */
+        if (sdslen(r->io.fd.buf)) {
+            if (rioFdWrite(r, NULL, 0) == 0)
+                return 0;
+        }
+        /* Write the new data, keeping 'p' and 'len' from the input. */
+    } else {
+        if (len) {
+            r->io.fd.buf = sdscatlen(r->io.fd.buf,buf,len);
+            if (sdslen(r->io.fd.buf) > PROTO_IOBUF_LEN)
+                doflush = 1;
+            if (!doflush)
+                return 1;
+        }
+        /* Flusing the buffered data. set 'p' and 'len' accordintly. */
+        p = (unsigned char*) r->io.fd.buf;
+        len = sdslen(r->io.fd.buf);
     }
 
-    /* Write in little chunchs so that when there are big writes we
-     * parallelize while the kernel is sending data in background to
-     * the TCP socket. */
-    while(len) {
-        size_t count = len < 1024 ? len : 1024;
-        int broken = 0;
-        for (j = 0; j < r->io.connset.numconns; j++) {
-            if (r->io.connset.state[j] != 0) {
-                /* Skip FDs alraedy in error. */
-                broken++;
-                continue;
-            }
-
-            /* Make sure to write 'count' bytes to the socket regardless
-             * of short writes. */
-            size_t nwritten = 0;
-            while(nwritten != count) {
-                retval = connWrite(r->io.connset.conns[j],p+nwritten,count-nwritten);
-                if (retval <= 0) {
-                    /* With blocking sockets, which is the sole user of this
-                     * rio target, EWOULDBLOCK is returned only because of
-                     * the SO_SNDTIMEO socket option, so we translate the error
-                     * into one more recognizable by the user. */
-                    if (retval == -1 && errno == EWOULDBLOCK) errno = ETIMEDOUT;
-                    break;
-                }
-                nwritten += retval;
-            }
-
-            if (nwritten != count) {
-                /* Mark this FD as broken. */
-                r->io.connset.state[j] = errno;
-                if (r->io.connset.state[j] == 0) r->io.connset.state[j] = EIO;
-            }
+    size_t nwritten = 0;
+    while(nwritten != len) {
+        retval = write(r->io.fd.fd,p+nwritten,len-nwritten);
+        if (retval <= 0) {
+            /* With blocking io, which is the sole user of this
+             * rio target, EWOULDBLOCK is returned only because of
+             * the SO_SNDTIMEO socket option, so we translate the error
+             * into one more recognizable by the user. */
+            if (retval == -1 && errno == EWOULDBLOCK) errno = ETIMEDOUT;
+            return 0; /* error. */
         }
-        if (broken == r->io.connset.numconns) return 0; /* All the FDs in error. */
-        p += count;
-        len -= count;
-        r->io.connset.pos += count;
+        nwritten += retval;
     }
 
-    if (doflush) sdsclear(r->io.connset.buf);
+    r->io.fd.pos += len;
+    sdsclear(r->io.fd.buf);
     return 1;
 }
 
 /* Returns 1 or 0 for success/failure. */
-static size_t rioFdsetRead(rio *r, void *buf, size_t len) {
+static size_t rioFdRead(rio *r, void *buf, size_t len) {
     UNUSED(r);
     UNUSED(buf);
     UNUSED(len);
@@ -358,23 +340,23 @@ static size_t rioFdsetRead(rio *r, void *buf, size_t len) {
 }
 
 /* Returns read/write position in file. */
-static off_t rioFdsetTell(rio *r) {
-    return r->io.connset.pos;
+static off_t rioFdTell(rio *r) {
+    return r->io.fd.pos;
 }
 
 /* Flushes any buffer to target device if applicable. Returns 1 on success
  * and 0 on failures. */
-static int rioFdsetFlush(rio *r) {
+static int rioFdFlush(rio *r) {
     /* Our flush is implemented by the write method, that recognizes a
      * buffer set to NULL with a count of zero as a flush request. */
-    return rioFdsetWrite(r,NULL,0);
+    return rioFdWrite(r,NULL,0);
 }
 
-static const rio rioFdsetIO = {
-    rioFdsetRead,
-    rioFdsetWrite,
-    rioFdsetTell,
-    rioFdsetFlush,
+static const rio rioFdIO = {
+    rioFdRead,
+    rioFdWrite,
+    rioFdTell,
+    rioFdFlush,
     NULL,           /* update_checksum */
     0,              /* current checksum */
     0,              /* flags */
@@ -383,24 +365,16 @@ static const rio rioFdsetIO = {
     { { NULL, 0 } } /* union for io-specific vars */
 };
 
-void rioInitWithConnset(rio *r, connection **conns, int numconns) {
-    int j;
-
-    *r = rioFdsetIO;
-    r->io.connset.conns = zmalloc(sizeof(connection *)*numconns);
-    r->io.connset.state = zmalloc(sizeof(int)*numconns);
-    memcpy(r->io.connset.conns,conns,sizeof(connection *)*numconns);
-    for (j = 0; j < numconns; j++) r->io.connset.state[j] = 0;
-    r->io.connset.numconns = numconns;
-    r->io.connset.pos = 0;
-    r->io.connset.buf = sdsempty();
+void rioInitWithFd(rio *r, int fd) {
+    *r = rioFdIO;
+    r->io.fd.fd = fd;
+    r->io.fd.pos = 0;
+    r->io.fd.buf = sdsempty();
 }
 
 /* release the rio stream. */
-void rioFreeConnset(rio *r) {
-    zfree(r->io.connset.conns);
-    zfree(r->io.connset.state);
-    sdsfree(r->io.connset.buf);
+void rioFreeFd(rio *r) {
+    sdsfree(r->io.fd.buf);
 }
 
 /* ---------------------------- Generic functions ---------------------------- */
diff --git a/src/rio.h b/src/rio.h
index fdde7c20e..9576335e8 100644
--- a/src/rio.h
+++ b/src/rio.h
@@ -77,7 +77,7 @@ struct _rio {
             off_t buffered; /* Bytes written since last fsync. */
             off_t autosync; /* fsync after 'autosync' bytes written. */
         } file;
-        /* Connection object */
+        /* Connection object (used to read from socket) */
         struct {
             connection *conn;   /* Connection */
             off_t pos;    /* pos in buf that was returned */
@@ -85,14 +85,12 @@ struct _rio {
             size_t read_limit;  /* don't allow to buffer/read more than that */
             size_t read_so_far; /* amount of data read from the rio (not buffered) */
         } conn;
-        /* Multiple FDs target (used to write to N sockets). */
+        /* FD target (used to write to pipe). */
         struct {
-            connection **conns;  /* Connections */
-            int *state;         /* Error state of each fd. 0 (if ok) or errno. */
-            int numconns;
+            int fd;       /* File descriptor. */
             off_t pos;
             sds buf;
-        } connset;
+        } fd;
     } io;
 };
 
@@ -161,9 +159,9 @@ static inline void rioClearErrors(rio *r) {
 void rioInitWithFile(rio *r, FILE *fp);
 void rioInitWithBuffer(rio *r, sds s);
 void rioInitWithConn(rio *r, connection *conn, size_t read_limit);
-void rioInitWithConnset(rio *r, connection **conns, int numconns);
+void rioInitWithFd(rio *r, int fd);
 
-void rioFreeConnset(rio *r);
+void rioFreeFd(rio *r);
 void rioFreeConn(rio *r, sds* out_remainingBufferedData);
 
 size_t rioWriteBulkCount(rio *r, char prefix, long count);
diff --git a/src/server.c b/src/server.c
index 930a01723..f05ce3151 100644
--- a/src/server.c
+++ b/src/server.c
@@ -1746,6 +1746,48 @@ void updateCachedTime(void) {
     server.daylight_active = tm.tm_isdst;
 }
 
+void checkChildrenDone(void) {
+    int statloc;
+    pid_t pid;
+
+    /* If we have a diskless rdb child (note that we support only one concurrent
+     * child), we want to avoid collecting it's exit status and acting on it
+     * as long as we didn't finish to drain the pipe, since then we're at risk
+     * of starting a new fork and a new pipe before we're done with the previous
+     * one. */
+    if (server.rdb_child_pid != -1 && server.rdb_pipe_conns)
+        return;
+
+    if ((pid = wait3(&statloc,WNOHANG,NULL)) != 0) {
+        int exitcode = WEXITSTATUS(statloc);
+        int bysignal = 0;
+
+        if (WIFSIGNALED(statloc)) bysignal = WTERMSIG(statloc);
+
+        if (pid == -1) {
+            serverLog(LL_WARNING,"wait3() returned an error: %s. "
+                "rdb_child_pid = %d, aof_child_pid = %d",
+                strerror(errno),
+                (int) server.rdb_child_pid,
+                (int) server.aof_child_pid);
+        } else if (pid == server.rdb_child_pid) {
+            backgroundSaveDoneHandler(exitcode,bysignal);
+            if (!bysignal && exitcode == 0) receiveChildInfo();
+        } else if (pid == server.aof_child_pid) {
+            backgroundRewriteDoneHandler(exitcode,bysignal);
+            if (!bysignal && exitcode == 0) receiveChildInfo();
+        } else {
+            if (!ldbRemoveChild(pid)) {
+                serverLog(LL_WARNING,
+                    "Warning, detected child with unmatched pid: %ld",
+                    (long)pid);
+            }
+        }
+        updateDictResizePolicy();
+        closeChildInfoPipe();
+    }
+}
+
 /* This is our timer interrupt, called server.hz times per second.
  * Here is where we do a number of things that need to be done asynchronously.
  * For instance:
@@ -1898,37 +1940,7 @@ int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientData) {
     if (server.rdb_child_pid != -1 || server.aof_child_pid != -1 ||
         ldbPendingChildren())
     {
-        int statloc;
-        pid_t pid;
-
-        if ((pid = wait3(&statloc,WNOHANG,NULL)) != 0) {
-            int exitcode = WEXITSTATUS(statloc);
-            int bysignal = 0;
-
-            if (WIFSIGNALED(statloc)) bysignal = WTERMSIG(statloc);
-
-            if (pid == -1) {
-                serverLog(LL_WARNING,"wait3() returned an error: %s. "
-                    "rdb_child_pid = %d, aof_child_pid = %d",
-                    strerror(errno),
-                    (int) server.rdb_child_pid,
-                    (int) server.aof_child_pid);
-            } else if (pid == server.rdb_child_pid) {
-                backgroundSaveDoneHandler(exitcode,bysignal);
-                if (!bysignal && exitcode == 0) receiveChildInfo();
-            } else if (pid == server.aof_child_pid) {
-                backgroundRewriteDoneHandler(exitcode,bysignal);
-                if (!bysignal && exitcode == 0) receiveChildInfo();
-            } else {
-                if (!ldbRemoveChild(pid)) {
-                    serverLog(LL_WARNING,
-                        "Warning, detected child with unmatched pid: %ld",
-                        (long)pid);
-                }
-            }
-            updateDictResizePolicy();
-            closeChildInfoPipe();
-        }
+        checkChildrenDone();
     } else {
         /* If there is not a background saving/rewrite in progress check if
          * we have to save/rewrite now. */
@@ -2081,6 +2093,11 @@ void beforeSleep(struct aeEventLoop *eventLoop) {
     /* Handle writes with pending output buffers. */
     handleClientsWithPendingWritesUsingThreads();
 
+    /* TODO: How do i handle write barriers flag */
+    tlsProcessPendingData();
+    /* If tls already has pending unread data don't sleep at all. */
+    aeDontWait(server.el, tlsHasPendingData());
+
     /* Close clients that need to be closed asynchronous */
     freeClientsInAsyncFreeQueue();
 
@@ -2280,6 +2297,7 @@ void initServerConfig(void) {
     server.aof_rewrite_incremental_fsync = CONFIG_DEFAULT_AOF_REWRITE_INCREMENTAL_FSYNC;
     server.rdb_save_incremental_fsync = CONFIG_DEFAULT_RDB_SAVE_INCREMENTAL_FSYNC;
     server.rdb_key_save_delay = CONFIG_DEFAULT_RDB_KEY_SAVE_DELAY;
+    server.key_load_delay = CONFIG_DEFAULT_KEY_LOAD_DELAY;
     server.aof_load_truncated = CONFIG_DEFAULT_AOF_LOAD_TRUNCATED;
     server.aof_use_rdb_preamble = CONFIG_DEFAULT_AOF_USE_RDB_PREAMBLE;
     server.pidfile = NULL;
@@ -2813,6 +2831,11 @@ void initServer(void) {
     server.rdb_child_pid = -1;
     server.aof_child_pid = -1;
     server.rdb_child_type = RDB_CHILD_TYPE_NONE;
+    server.rdb_pipe_conns = NULL;
+    server.rdb_pipe_numconns = 0;
+    server.rdb_pipe_numconns_writing = 0;
+    server.rdb_pipe_buff = NULL;
+    server.rdb_pipe_bufflen = 0;
     server.rdb_bgsave_scheduled = 0;
     server.child_info_pipe[0] = -1;
     server.child_info_pipe[1] = -1;
diff --git a/src/server.h b/src/server.h
index 1b41c9ac5..89f219033 100644
--- a/src/server.h
+++ b/src/server.h
@@ -135,6 +135,7 @@ typedef long long mstime_t; /* millisecond time type. */
 #define CONFIG_DEFAULT_REPL_DISKLESS_SYNC 0
 #define CONFIG_DEFAULT_REPL_DISKLESS_SYNC_DELAY 5
 #define CONFIG_DEFAULT_RDB_KEY_SAVE_DELAY 0
+#define CONFIG_DEFAULT_KEY_LOAD_DELAY 0
 #define CONFIG_DEFAULT_SLAVE_SERVE_STALE_DATA 1
 #define CONFIG_DEFAULT_SLAVE_READ_ONLY 1
 #define CONFIG_DEFAULT_SLAVE_IGNORE_MAXMEMORY 1
@@ -1236,10 +1237,17 @@ struct redisServer {
     int rdb_child_type;             /* Type of save by active child. */
     int lastbgsave_status;          /* C_OK or C_ERR */
     int stop_writes_on_bgsave_err;  /* Don't allow writes if can't BGSAVE */
-    int rdb_pipe_write_result_to_parent; /* RDB pipes used to return the state */
-    int rdb_pipe_read_result_from_child; /* of each slave in diskless SYNC. */
+    int rdb_pipe_write;             /* RDB pipes used to transfer the rdb */
+    int rdb_pipe_read;              /* data to the parent process in diskless repl. */
+    connection **rdb_pipe_conns;    /* Connections which are currently the */
+    int rdb_pipe_numconns;          /* target of diskless rdb fork child. */
+    int rdb_pipe_numconns_writing;  /* Number of rdb conns with pending writes. */
+    char *rdb_pipe_buff;            /* In diskless replication, this buffer holds data */
+    int rdb_pipe_bufflen;           /* that was read from the the rdb pipe. */
     int rdb_key_save_delay;         /* Delay in microseconds between keys while
                                      * writing the RDB. (for testings) */
+    int key_load_delay;             /* Delay in microseconds between keys while
+                                     * loading aof or rdb. (for testings) */
     /* Pipe and data structures for child -> parent info sharing. */
     int child_info_pipe[2];         /* Pipe used to write the child_info_data. */
     struct {
@@ -1779,6 +1787,8 @@ void clearReplicationId2(void);
 void chopReplicationBacklog(void);
 void replicationCacheMasterUsingMyself(void);
 void feedReplicationBacklog(void *ptr, size_t len);
+void rdbPipeReadHandler(struct aeEventLoop *eventLoop, int fd, void *clientData, int mask);
+void rdbPipeWriteHandlerConnRemoved(struct connection *conn);
 
 /* Generic persistence functions */
 void startLoadingFile(FILE* fp, char* filename);
@@ -1944,6 +1954,7 @@ unsigned int LRU_CLOCK(void);
 const char *evictPolicyToString(void);
 struct redisMemOverhead *getMemoryOverheadData(void);
 void freeMemoryOverheadData(struct redisMemOverhead *mh);
+void checkChildrenDone(void);
 
 #define RESTART_SERVER_NONE 0
 #define RESTART_SERVER_GRACEFULLY (1<<0)     /* Do proper shutdown. */
diff --git a/src/tls.c b/src/tls.c
index dabb2ee0f..c939ebf55 100644
--- a/src/tls.c
+++ b/src/tls.c
@@ -30,6 +30,7 @@
 
 #include "server.h"
 #include "connhelpers.h"
+#include "adlist.h"
 
 #ifdef USE_OPENSSL
 
@@ -41,6 +42,10 @@ extern ConnectionType CT_Socket;
 
 SSL_CTX *redis_tls_ctx;
 
+/* list of connections with pending data already read from the socket, but not
+ * served to the reader yet. */
+static list *pending_list = NULL;
+
 void tlsInit(void) {
     ERR_load_crypto_strings();
     SSL_load_error_strings();
@@ -49,6 +54,8 @@ void tlsInit(void) {
     if (!RAND_poll()) {
         serverLog(LL_WARNING, "OpenSSL: Failed to seed random number generator.");
     }
+
+    pending_list = listCreate();
 }
 
 int tlsConfigureServer(void) {
@@ -188,6 +195,7 @@ typedef struct tls_connection {
     int flags;
     SSL *ssl;
     char *ssl_error;
+    listNode *pending_list_node;
 } tls_connection;
 
 connection *connCreateTLS(void) {
@@ -288,11 +296,7 @@ void updateSSLEvent(tls_connection *conn) {
         aeDeleteFileEvent(server.el, conn->c.fd, AE_WRITABLE);
 }
 
-
-static void tlsEventHandler(struct aeEventLoop *el, int fd, void *clientData, int mask) {
-    UNUSED(el);
-    UNUSED(fd);
-    tls_connection *conn = clientData;
+static void tlsHandleEvent(tls_connection *conn, int mask) {
     int ret;
 
     TLSCONN_DEBUG("tlsEventHandler(): fd=%d, state=%d, mask=%d, r=%d, w=%d, flags=%d",
@@ -369,6 +373,15 @@ static void tlsEventHandler(struct aeEventLoop *el, int fd, void *clientData, in
 
             if ((mask & AE_READABLE) && conn->c.read_handler) {
                 if (!callHandler((connection *) conn, conn->c.read_handler)) return;
+                if (SSL_has_pending(conn->ssl)) {
+                    if (!conn->pending_list_node) {
+                        listAddNodeTail(pending_list, conn);
+                        conn->pending_list_node = listLast(pending_list);
+                    }
+                } else if (conn->pending_list_node) {
+                    listDelNode(pending_list, conn->pending_list_node);
+                    conn->pending_list_node = NULL;
+                }
             }
 
             if ((mask & AE_WRITABLE) && conn->c.write_handler) {
@@ -382,6 +395,13 @@ static void tlsEventHandler(struct aeEventLoop *el, int fd, void *clientData, in
     updateSSLEvent(conn);
 }
 
+static void tlsEventHandler(struct aeEventLoop *el, int fd, void *clientData, int mask) {
+    UNUSED(el);
+    UNUSED(fd);
+    tls_connection *conn = clientData;
+    tlsHandleEvent(conn, mask);
+}
+
 static void connTLSClose(connection *conn_) {
     tls_connection *conn = (tls_connection *) conn_;
 
@@ -395,6 +415,11 @@ static void connTLSClose(connection *conn_) {
         conn->ssl_error = NULL;
     }
 
+    if (conn->pending_list_node) {
+        listDelNode(pending_list, conn->pending_list_node);
+        conn->pending_list_node = NULL;
+    }
+
     CT_Socket.close(conn_);
 }
 
@@ -610,17 +635,6 @@ exit:
     return nread;
 }
 
-/* TODO: This is probably not the right thing to do, but as we handle proxying from child
- * processes we'll probably not need any shutdown mechanism anyway so this is just a
- * place holder for now.
- */
-static int connTLSShutdown(connection *conn_, int how) {
-    UNUSED(how);
-    tls_connection *conn = (tls_connection *) conn_;
-
-    return SSL_shutdown(conn->ssl);
-}
-
 ConnectionType CT_TLS = {
     .ae_handler = tlsEventHandler,
     .accept = connTLSAccept,
@@ -635,9 +649,25 @@ ConnectionType CT_TLS = {
     .sync_write = connTLSSyncWrite,
     .sync_read = connTLSSyncRead,
     .sync_readline = connTLSSyncReadLine,
-    .shutdown = connTLSShutdown
 };
 
+int tlsHasPendingData() {
+    if (!pending_list)
+        return 0;
+    return listLength(pending_list) > 0;
+}
+
+void tlsProcessPendingData() {
+    listIter li;
+    listNode *ln;
+
+    listRewind(pending_list,&li);
+    while((ln = listNext(&li))) {
+        tls_connection *conn = listNodeValue(ln);
+        tlsHandleEvent(conn, AE_READABLE);
+    }
+}
+
 #else   /* USE_OPENSSL */
 
 void tlsInit(void) {
@@ -666,4 +696,11 @@ connection *connCreateAcceptedTLS(int fd, int require_auth) {
     return NULL;
 }
 
+int tlsHasPendingData() {
+    return 0;
+}
+
+void tlsProcessPendingData() {
+}
+
 #endif
diff --git a/tests/integration/replication.tcl b/tests/integration/replication.tcl
index 43bc684b4..192137e87 100644
--- a/tests/integration/replication.tcl
+++ b/tests/integration/replication.tcl
@@ -29,9 +29,6 @@ start_server {tags {"repl"}} {
         $slave slaveof $master_host $master_port
 
         test {Slave enters handshake} {
-            if {$::tls} {
-                fail "TLS with repl-diskless-sync not supported yet."
-            }
             wait_for_condition 50 1000 {
                 [string match *handshake* [$slave role]]
             } else {
@@ -187,10 +184,6 @@ start_server {tags {"repl"}} {
 }
 
 foreach mdl {no yes} {
-    if {$::tls && $mdl eq "yes"} {
-        puts "** Skipping test: TLS with repl-diskless-sync not supported yet."
-        continue
-    }
     foreach sdl {disabled swapdb} {
         start_server {tags {"repl"}} {
             set master [srv 0 client]
@@ -327,9 +320,6 @@ start_server {tags {"repl"}} {
 }
 
 test {slave fails full sync and diskless load swapdb recoveres it} {
-    if {$::tls} {
-        fail ""
-    }
     start_server {tags {"repl"}} {
         set slave [srv 0 client]
         set slave_host [srv 0 host]
@@ -397,10 +387,6 @@ test {slave fails full sync and diskless load swapdb recoveres it} {
 }
 
 test {diskless loading short read} {
-    if {$::tls} {
-        fail "TLS with repl-diskless-sync not supported yet."
-    }
-
     start_server {tags {"repl"}} {
         set replica [srv 0 client]
         set replica_host [srv 0 host]
@@ -480,3 +466,159 @@ test {diskless loading short read} {
     }
 }
 
+# get current stime and utime metrics for a thread (since it's creation)
+proc get_cpu_metrics { statfile } {
+    if { [ catch {
+        set fid   [ open $statfile r ]
+        set data  [ read $fid 1024 ]
+        ::close $fid
+        set data  [ split $data ]
+
+        ;## number of jiffies it has been scheduled...
+        set utime [ lindex $data 13 ]
+        set stime [ lindex $data 14 ]
+    } err ] } {
+        error "assertion:can't parse /proc: $err"
+    }
+    set mstime [clock milliseconds]
+    return [ list $mstime $utime $stime ]
+}
+
+# compute %utime and %stime of a thread between two measurements
+proc compute_cpu_usage {start end} {
+    set clock_ticks [exec getconf CLK_TCK]
+    # convert ms time to jiffies and calc delta
+    set dtime [ expr { ([lindex $end 0] - [lindex $start 0]) * double($clock_ticks) / 1000 } ]
+    set utime [ expr { [lindex $end 1] - [lindex $start 1] } ]
+    set stime [ expr { [lindex $end 2] - [lindex $start 2] } ]
+    set pucpu  [ expr { ($utime / $dtime) * 100 } ]
+    set pscpu  [ expr { ($stime / $dtime) * 100 } ]
+    return [ list $pucpu $pscpu ]
+}
+
+
+# test diskless rdb pipe with multiple replicas, which may drop half way
+start_server {tags {"repl"}} {
+    set master [srv 0 client]
+    $master config set repl-diskless-sync yes
+    $master config set repl-diskless-sync-delay 1
+    set master_host [srv 0 host]
+    set master_port [srv 0 port]
+    # put enough data in the db that the rdb file will be bigger than the socket buffers
+    # and since we'll have key-load-delay of 100, 10000 keys will take at least 1 second
+    # we also need the replica to process requests during transfer (which it does only once in 2mb)
+    $master debug populate 10000 test 10000
+    $master config set rdbcompression no
+    foreach all_drop {no slow fast all} {
+        test "diskless $all_drop replicas drop during rdb pipe" {
+            set replicas {}
+            set replicas_alive {}
+            # start one replica that will read the rdb fast, and one that will be slow
+            start_server {} {
+                lappend replicas [srv 0 client]
+                lappend replicas_alive [srv 0 client]
+                start_server {} {
+                    lappend replicas [srv 0 client]
+                    lappend replicas_alive [srv 0 client]
+
+                    # start replication
+                    # it's enough for just one replica to be slow, and have it's write handler enabled
+                    # so that the whole rdb generation process is bound to that
+                    [lindex $replicas 0] config set repl-diskless-load swapdb
+                    [lindex $replicas 0] config set key-load-delay 100
+                    [lindex $replicas 0] replicaof $master_host $master_port
+                    [lindex $replicas 1] replicaof $master_host $master_port
+
+                    # wait for the replicas to start reading the rdb
+                    # using the log file since the replica only responds to INFO once in 2mb
+                    wait_for_log_message -1 "*Loading DB in memory*" 8 800 10
+
+                    set master_statfile [format "/proc/%s/stat" [srv -2 pid]]
+                    set master_start_metrics [get_cpu_metrics $master_statfile]
+                    set start_time [clock seconds]
+
+                    # wait a while so that the pipe socket writer will be
+                    # blocked on write (since replica 0 is slow to read from the socket)
+                    after 500
+
+                    # add some command to be present in the command stream after the rdb.
+                    $master incr $all_drop
+
+                    # disconnect replicas depending on the current test
+                    if {$all_drop == "all" || $all_drop == "fast"} {
+                        exec kill [srv 0 pid]
+                        set replicas_alive [lreplace $replicas_alive 1 1]
+                    }
+                    if {$all_drop == "all" || $all_drop == "slow"} {
+                        exec kill [srv -1 pid]
+                        set replicas_alive [lreplace $replicas_alive 0 0]
+                    }
+
+                    # wait for rdb child to exit
+                    wait_for_condition 500 100 {
+                        [s -2 rdb_bgsave_in_progress] == 0
+                    } else {
+                        fail "rdb child didn't terminate"
+                    }
+
+                    # make sure we got what we were aiming for, by looking for the message in the log file
+                    if {$all_drop == "all"} {
+                        wait_for_log_message -2 "*Diskless rdb transfer, last replica dropped, killing fork child*" 12 1 1
+                    }
+                    if {$all_drop == "no"} {
+                        wait_for_log_message -2 "*Diskless rdb transfer, done reading from pipe, 2 replicas still up*" 12 1 1
+                    }
+                    if {$all_drop == "slow" || $all_drop == "fast"} {
+                        wait_for_log_message -2 "*Diskless rdb transfer, done reading from pipe, 1 replicas still up*" 12 1 1
+                    }
+
+                    # make sure we don't have a busy loop going thought epoll_wait
+                    set master_end_metrics [get_cpu_metrics $master_statfile]
+                    set time_elapsed [expr {[clock seconds]-$start_time}]
+                    set master_cpu [compute_cpu_usage $master_start_metrics $master_end_metrics]
+                    set master_utime [lindex $master_cpu 0]
+                    set master_stime [lindex $master_cpu 1]
+                    if {$::verbose} {
+                        puts "elapsed: $time_elapsed"
+                        puts "master utime: $master_utime"
+                        puts "master stime: $master_stime"
+                    }
+                    if {$all_drop == "all" || $all_drop == "slow"} {
+                        assert {$master_utime < 30}
+                        assert {$master_stime < 30}
+                    }
+                    if {$all_drop == "none" || $all_drop == "fast"} {
+                        assert {$master_utime < 15}
+                        assert {$master_stime < 15}
+                    }
+
+                    # verify the data integrity
+                    foreach replica $replicas_alive {
+                        # Wait that replicas acknowledge they are online so
+                        # we are sure that DBSIZE and DEBUG DIGEST will not
+                        # fail because of timing issues.
+                        wait_for_condition 50 100 {
+                            [lindex [$replica role] 3] eq {connected}
+                        } else {
+                            fail "replicas still not connected after some time"
+                        }
+
+                        # Make sure that replicas and master have same
+                        # number of keys
+                        wait_for_condition 50 100 {
+                            [$master dbsize] == [$replica dbsize]
+                        } else {
+                            fail "Different number of keys between master and replicas after too long time."
+                        }
+
+                        # Check digests
+                        set digest [$master debug digest]
+                        set digest0 [$replica debug digest]
+                        assert {$digest ne 0000000000000000000000000000000000000000}
+                        assert {$digest eq $digest0}
+                    }
+                }
+            }
+        }
+    }
+}