"SORT by nosort" (skip sorting) respect sorted set ordering.

[redis.git] / src / syncio.c

diff --git a/src/syncio.c b/src/syncio.c
index 81eeda08f9be27232aea66b061163f71d8790499..0c202c9e242a47f730a2f7fa95aa64ae24354d2f 100644 (file)
--- a/src/syncio.c
+++ b/src/syncio.c
@@ -1,9 +1,4 @@
-/* Synchronous socket I/O operations, with timeouts.
- * Redis performs most of the I/O in a nonblocking way, with the exception
- * of the SYNC command where the slave does it in a blocking way, and
- * the MIGRATE command that must be blocking in order to be atomic from the
- * point of view of the two instances (one migrating the key and one receiving
- * the key). This is why need the following blocking I/O functions.
+/* Synchronous socket and file I/O operations useful across the core.
  *
  * Copyright (c) 2009-2010, Salvatore Sanfilippo <antirez at gmail dot com>
  * All rights reserved.
@@ -35,48 +30,99 @@
 
 #include "redis.h"
 
-int syncWrite(int fd, char *ptr, ssize_t size, int timeout) {
+/* ----------------- Blocking sockets I/O with timeouts --------------------- */
+
+/* Redis performs most of the I/O in a nonblocking way, with the exception
+ * of the SYNC command where the slave does it in a blocking way, and
+ * the MIGRATE command that must be blocking in order to be atomic from the
+ * point of view of the two instances (one migrating the key and one receiving
+ * the key). This is why need the following blocking I/O functions.
+ *
+ * All the functions take the timeout in milliseconds. */
+
+#define REDIS_SYNCIO_RESOLUTION 10 /* Resolution in milliseconds */
+
+/* Write the specified payload to 'fd'. If writing the whole payload will be
+ * done within 'timeout' milliseconds the operation succeeds and 'size' is
+ * returned. Otherwise the operation fails, -1 is returned, and an unspecified
+ * partial write could be performed against the file descriptor. */
+ssize_t syncWrite(int fd, char *ptr, ssize_t size, long long timeout) {
     ssize_t nwritten, ret = size;
-    time_t start = time(NULL);
+    long long start = mstime();
+    long long remaining = timeout;
 
-    timeout++;
-    while(size) {
-        if (aeWait(fd,AE_WRITABLE,1000) & AE_WRITABLE) {
-            nwritten = write(fd,ptr,size);
-            if (nwritten == -1) return -1;
+    while(1) {
+        long long wait = (remaining > REDIS_SYNCIO_RESOLUTION) ?
+                          remaining : REDIS_SYNCIO_RESOLUTION;
+        long long elapsed;
+
+        /* Optimistically try to write before checking if the file descriptor
+         * is actually writable. At worst we get EAGAIN. */
+        nwritten = write(fd,ptr,size);
+        if (nwritten == -1) {
+            if (errno != EAGAIN) return -1;
+        } else {
             ptr += nwritten;
             size -= nwritten;
         }
-        if ((time(NULL)-start) > timeout) {
+        if (size == 0) return ret;
+
+        /* Wait */
+        aeWait(fd,AE_WRITABLE,wait);
+        elapsed = mstime() - start;
+        if (elapsed >= timeout) {
             errno = ETIMEDOUT;
             return -1;
         }
+        remaining = timeout - elapsed;
     }
-    return ret;
 }
 
-int syncRead(int fd, char *ptr, ssize_t size, int timeout) {
+/* Read the specified amount of bytes from 'fd'. If all the bytes are read
+ * within 'timeout' milliseconds the operation succeed and 'size' is returned.
+ * Otherwise the operation fails, -1 is returned, and an unspecified amount of
+ * data could be read from the file descriptor. */
+ssize_t syncRead(int fd, char *ptr, ssize_t size, long long timeout) {
     ssize_t nread, totread = 0;
-    time_t start = time(NULL);
+    long long start = mstime();
+    long long remaining = timeout;
 
-    timeout++;
-    while(size) {
-        if (aeWait(fd,AE_READABLE,1000) & AE_READABLE) {
-            nread = read(fd,ptr,size);
-            if (nread <= 0) return -1;
+    if (size == 0) return 0;
+    while(1) {
+        long long wait = (remaining > REDIS_SYNCIO_RESOLUTION) ?
+                          remaining : REDIS_SYNCIO_RESOLUTION;
+        long long elapsed;
+
+        /* Optimistically try to read before checking if the file descriptor
+         * is actually readable. At worst we get EAGAIN. */
+        nread = read(fd,ptr,size);
+        if (nread == 0) return -1; /* short read. */
+        if (nread == -1) {
+            if (errno != EAGAIN) return -1;
+        } else {
             ptr += nread;
             size -= nread;
             totread += nread;
         }
-        if ((time(NULL)-start) > timeout) {
+        if (size == 0) return totread;
+
+        /* Wait */
+        aeWait(fd,AE_READABLE,wait);
+        elapsed = mstime() - start;
+        if (elapsed >= timeout) {
             errno = ETIMEDOUT;
             return -1;
         }
+        remaining = timeout - elapsed;
     }
-    return totread;
 }
 
-int syncReadLine(int fd, char *ptr, ssize_t size, int timeout) {
+/* Read a line making sure that every char will not require more than 'timeout'
+ * milliseconds to be read.
+ * 
+ * On success the number of bytes read is returned, otherwise -1.
+ * On success the string is always correctly terminated with a 0 byte. */
+ssize_t syncReadLine(int fd, char *ptr, ssize_t size, long long timeout) {
     ssize_t nread = 0;
 
     size--;