Fix: only print event errors every 1048576 hits
authorMathieu Desnoyers <mathieu.desnoyers@efficios.com>
Fri, 24 Feb 2012 01:11:17 +0000 (20:11 -0500)
committerMathieu Desnoyers <mathieu.desnoyers@efficios.com>
Fri, 24 Feb 2012 01:11:17 +0000 (20:11 -0500)
Never spam the console, even in debug mode, since it would make debug
information very much useless.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
libringbuffer/ring_buffer_frontend.c

index 9d89c6fa68a7117b06b1701f17d33b2b3762e308..0996ecefc0313ff5ef8b955ce7dffd3a6a793849 100644 (file)
@@ -58,6 +58,9 @@
 #define max(a, b)      ((a) > (b) ? (a) : (b))
 #endif
 
+/* Print DBG() messages about events lost only every 1048576 hits */
+#define DBG_PRINT_NR_LOST      (1UL << 20)
+
 /*
  * Use POSIX SHM: shm_open(3) and shm_unlink(3).
  * close(2) to close the fd returned by shm_open.
@@ -1336,12 +1339,19 @@ int lib_ring_buffer_try_reserve_slow(struct lttng_ust_lib_ring_buffer *buf,
                                 - subbuf_trunc((unsigned long)
                                     uatomic_read(&buf->consumed), chan)
                                >= chan->backend.buf_size)) {
+                               unsigned long nr_lost;
+
                                /*
                                 * We do not overwrite non consumed buffers
                                 * and we are full : record is lost.
                                 */
+                               nr_lost = v_read(config, &buf->records_lost_full);
                                v_inc(config, &buf->records_lost_full);
-                               DBG("Record lost: buffer is full\n");
+                               if ((nr_lost & (DBG_PRINT_NR_LOST - 1)) == 0) {
+                                       DBG("%lu or more records lost in (%s:%d) (buffer full)\n",
+                                               nr_lost + 1, chan->backend.name,
+                                               buf->backend.cpu);
+                               }
                                return -ENOBUFS;
                        } else {
                                /*
@@ -1352,14 +1362,21 @@ int lib_ring_buffer_try_reserve_slow(struct lttng_ust_lib_ring_buffer *buf,
                                 */
                        }
                } else {
+                       unsigned long nr_lost;
+
                        /*
                         * Next subbuffer reserve offset does not match the
                         * commit offset. Drop record in producer-consumer and
                         * overwrite mode. Caused by either a writer OOPS or too
                         * many nested writes over a reserve/commit pair.
                         */
+                       nr_lost = v_read(config, &buf->records_lost_wrap);
                        v_inc(config, &buf->records_lost_wrap);
-                       DBG("Record lost: buffer wrap-around\n");
+                       if ((nr_lost & (DBG_PRINT_NR_LOST - 1)) == 0) {
+                               DBG("%lu or more records lost in (%s:%d) (wrap-around)\n",
+                                       nr_lost + 1, chan->backend.name,
+                                       buf->backend.cpu);
+                       }
                        return -EIO;
                }
                offsets->size =
@@ -1373,12 +1390,20 @@ int lib_ring_buffer_try_reserve_slow(struct lttng_ust_lib_ring_buffer *buf,
                        + ctx->data_size;
                if (caa_unlikely(subbuf_offset(offsets->begin, chan)
                             + offsets->size > chan->backend.subbuf_size)) {
+                       unsigned long nr_lost;
+
                        /*
                         * Record too big for subbuffers, report error, don't
                         * complete the sub-buffer switch.
                         */
+                       nr_lost = v_read(config, &buf->records_lost_big);
                        v_inc(config, &buf->records_lost_big);
-                       DBG("Record lost: record size (%zu bytes) is too large for buffer\n", offsets->size);
+                       if ((nr_lost & (DBG_PRINT_NR_LOST - 1)) == 0) {
+                               DBG("%lu or more records lost in (%s:%d) record size "
+                                       " of %zu bytes is too large for buffer\n",
+                                       nr_lost + 1, chan->backend.name,
+                                       buf->backend.cpu, offsets->size);
+                       }
                        return -ENOSPC;
                } else {
                        /*
This page took 0.02938 seconds and 4 git commands to generate.