retry writes when no space returned closes[t:2436]

git-svn-id: file:///svn/toku/tokudb@18362 c7de825b-a66e-492c-adef-691d508d4ae1

linux/file.c

@@ -6,6 +6,28 @@
 #include <toku_assert.h>
 #include <stdio.h>
 #include <string.h>
+#include <time.h>
+
+static uint64_t get_tnow(void) {
+    struct timeval tv;
+    int r = gettimeofday(&tv, NULL); assert(r == 0);
+    return tv.tv_sec * 1000000ULL + tv.tv_usec;
+}
+
+#define DO_ASSERT_ON_ENOSPC 0
+
+static const int toku_write_enospc_sleep = 1;
+static uint64_t toku_write_enospc_last_report;
+static uint64_t toku_write_enospc_last_time;
+static uint32_t toku_write_enospc_current;
+static uint64_t toku_write_enospc_total;
+
+void
+toku_fs_get_write_info(uint64_t *enospc_last_time, uint64_t *enospc_current, uint64_t *enospc_total) {
+    *enospc_last_time = toku_write_enospc_last_time;
+    *enospc_current = toku_write_enospc_current;
+    *enospc_total = toku_write_enospc_total;
+}
 
 //Print any necessary errors
 //Return whether we should try the write again.
@@ -26,12 +48,44 @@ try_again_after_handling_write_error(int fd, size_t len, ssize_t r_write) {
 	break;
     }
     case ENOSPC: {
+#if DO_ASSERT_ON_ENOSPC
 	char err_msg[sizeof("Failed write of [] bytes to fd=[].") + 20+10]; //64 bit is 20 chars, 32 bit is 10 chars
 	snprintf(err_msg, sizeof(err_msg), "Failed write of [%"PRIu64"] bytes to fd=[%d].", (uint64_t)len, fd);
 	perror(err_msg);
 	fflush(stderr);
 	int out_of_disk_space = 1;
 	assert(!out_of_disk_space); //Give an error message that might be useful if this is the only one that survives.
+#else
+        toku_sync_fetch_and_increment_uint64(&toku_write_enospc_total);
+        toku_sync_fetch_and_increment_uint32(&toku_write_enospc_current);
+
+        uint64_t tnow = get_tnow();
+        toku_write_enospc_last_time = tnow;
+        if (toku_write_enospc_last_report == 0 || tnow - toku_write_enospc_last_report >= 60*1000000) {
+            toku_write_enospc_last_report = tnow;
+
+            const int tstr_length = 26;
+            char tstr[tstr_length];
+            time_t t = time(0);
+            ctime_r(&t, tstr);
+
+            const int MY_MAX_PATH = 256;
+            char fname[MY_MAX_PATH], symname[MY_MAX_PATH];
+            sprintf(fname, "/proc/%d/fd/%d", getpid(), fd);
+            ssize_t n = readlink(fname, symname, MY_MAX_PATH);
+
+            if ((int)n == -1)
+                fprintf(stderr, "%.24s Tokudb No space when writing %"PRIu64" bytes to fd=%d ", tstr, (uint64_t) len, fd);
+            else
+                fprintf(stderr, "%.24s Tokudb No space when writing %"PRIu64" bytes to %*s ", tstr, (uint64_t) len, (int) n, symname); 
+            fprintf(stderr, "retry in %d second%s\n", toku_write_enospc_sleep, toku_write_enospc_sleep > 1 ? "s" : "");
+            fflush(stderr);
+        }
+	sleep(toku_write_enospc_sleep);
+	try_again = 1;
+        toku_sync_fetch_and_decrement_uint32(&toku_write_enospc_current);
+	break;
+#endif
         }
     default:
 	break;
@@ -114,12 +168,6 @@ toku_os_write (int fd, const void *buf, size_t len) {
     return 0;
 }
 
-static uint64_t get_tnow(void) {
-    struct timeval tv;
-    int r = gettimeofday(&tv, NULL); assert(r == 0);
-    return tv.tv_sec * 1000000ULL + tv.tv_usec;
-}
-
 // t_fsync exists for testing purposes only
 static int (*t_fsync)(int) = 0;
 static uint64_t toku_fsync_count;

toku_include/toku_os.h

@@ -67,6 +67,12 @@ int toku_os_initialize_settings(int verbosity)  __attribute__((__visibility__("d
 //
 int toku_os_is_absolute_name(const char* path)  __attribute__((__visibility__("default")));
 
+// Get file system write information
+// *enospc_last_time is the last time ENOSPC was returned by write or pwrite
+// *enospc_current   is the number of threads waiting on space
+// *enospc_total     is the number of times ENOSPC was returned by write or pwrite
+void toku_fs_get_write_info(uint64_t *enospc_last_time, uint64_t *enospc_current, uint64_t *enospc_total);
+
 #if TOKU_WINDOWS
 #include <sys/types.h>
 #include <sys/stat.h>