libaio: commit on EAGAIN

[fio.git] / engines / libaio.c

diff --git a/engines/libaio.c b/engines/libaio.c
index cd10aabfe831c772602cc9523756e27dcc4f310a..7c3a42a48fd99a4819ea020755c8c07a512ac5c7 100644 (file)
--- a/engines/libaio.c
+++ b/engines/libaio.c
@@ -13,6 +13,8 @@
 
 #include "../fio.h"
 
+static int fio_libaio_commit(struct thread_data *td);
+
 struct libaio_data {
        io_context_t aio_ctx;
        struct io_event *aio_events;
@@ -165,8 +167,11 @@ static int fio_libaio_getevents(struct thread_data *td, unsigned int min,
                }
                if (r >= 0)
                        events += r;
-               else if (r == -EAGAIN)
+               else if (r == -EAGAIN) {
+                       fio_libaio_commit(td);
                        usleep(100);
+               } else if (r != -EINTR)
+                       break;
        } while (events < min);
 
        return r < 0 ? r : events;
@@ -234,7 +239,8 @@ static int fio_libaio_commit(struct thread_data *td)
        struct libaio_data *ld = td->io_ops->data;
        struct iocb **iocbs;
        struct io_u **io_us;
-       int ret;
+       struct timeval tv;
+       int ret, wait_start = 0;
 
        if (!ld->queued)
                return 0;
@@ -254,21 +260,46 @@ static int fio_libaio_commit(struct thread_data *td)
                        ld->queued -= ret;
                        ring_inc(ld, &ld->tail, ret);
                        ret = 0;
+                       wait_start = 0;
                } else if (ret == -EINTR || !ret) {
                        if (!ret)
                                io_u_mark_submit(td, ret);
+                       wait_start = 0;
                        continue;
                } else if (ret == -EAGAIN) {
                        /*
                         * If we get EAGAIN, we should break out without
                         * error and let the upper layer reap some
-                        * events for us.
+                        * events for us. If we have no queued IO, we
+                        * must loop here. If we loop for more than 30s,
+                        * just error out, something must be buggy in the
+                        * IO path.
                         */
-                       ret = 0;
+                       if (ld->queued) {
+                               ret = 0;
+                               break;
+                       }
+                       if (!wait_start) {
+                               fio_gettime(&tv, NULL);
+                               wait_start = 0;
+                       } else if (mtime_since_now(&tv) > 30000) {
+                               log_err("fio: aio appears to be stalled, giving up\n");
+                               break;
+                       }
+                       usleep(1);
+                       continue;
+               } else if (ret == -ENOMEM) {
+                       /*
+                        * If we get -ENOMEM, reap events if we can. If
+                        * we cannot, treat it as a fatal event since there's
+                        * nothing we can do about it.
+                        */
+                       if (ld->queued)
+                               ret = 0;
                        break;
                } else
                        break;
-       } while (ld->head != ld->tail);
+       } while (ld->queued);
 
        return ret;
 }