X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=engines%2Frbd.c;h=2be9b556634fe8f39f2c9b86047949fdef80ccd4;hp=363be65bea58ded83e37eb7b867283d6aada67f8;hb=f39159a3e578a88459c80d9efd12529399c12550;hpb=fea585d43f975cef74beb3df770168f36ae895b9 diff --git a/engines/rbd.c b/engines/rbd.c index 363be65b..2be9b556 100644 --- a/engines/rbd.c +++ b/engines/rbd.c @@ -13,6 +13,7 @@ struct fio_rbd_iou { struct io_u *io_u; rbd_completion_t completion; int io_seen; + int io_complete; }; struct rbd_data { @@ -20,10 +21,11 @@ struct rbd_data { rados_ioctx_t io_ctx; rbd_image_t image; struct io_u **aio_events; + struct io_u **sort_events; }; struct rbd_options { - struct thread_data *td; + void *pad; char *rbd_name; char *pool_name; char *client_name; @@ -76,67 +78,64 @@ static struct fio_option options[] = { static int _fio_setup_rbd_data(struct thread_data *td, struct rbd_data **rbd_data_ptr) { - struct rbd_data *rbd_data; + struct rbd_data *rbd; if (td->io_ops->data) return 0; - rbd_data = malloc(sizeof(struct rbd_data)); - if (!rbd_data) + rbd = calloc(1, sizeof(struct rbd_data)); + if (!rbd) goto failed; - memset(rbd_data, 0, sizeof(struct rbd_data)); - - rbd_data->aio_events = malloc(td->o.iodepth * sizeof(struct io_u *)); - if (!rbd_data->aio_events) + rbd->aio_events = calloc(td->o.iodepth, sizeof(struct io_u *)); + if (!rbd->aio_events) goto failed; - memset(rbd_data->aio_events, 0, td->o.iodepth * sizeof(struct io_u *)); - - *rbd_data_ptr = rbd_data; + rbd->sort_events = calloc(td->o.iodepth, sizeof(struct io_u *)); + if (!rbd->sort_events) + goto failed; + *rbd_data_ptr = rbd; return 0; failed: - if (rbd_data) - free(rbd_data); + if (rbd) + free(rbd); return 1; } static int _fio_rbd_connect(struct thread_data *td) { - struct rbd_data *rbd_data = td->io_ops->data; + struct rbd_data *rbd = td->io_ops->data; struct rbd_options *o = td->eo; int r; - r = rados_create(&rbd_data->cluster, o->client_name); + r = rados_create(&rbd->cluster, o->client_name); if (r < 0) { log_err("rados_create failed.\n"); goto failed_early; } - r = rados_conf_read_file(rbd_data->cluster, NULL); + r = rados_conf_read_file(rbd->cluster, NULL); if (r < 0) { log_err("rados_conf_read_file failed.\n"); goto failed_early; } - r = rados_connect(rbd_data->cluster); + r = rados_connect(rbd->cluster); if (r < 0) { log_err("rados_connect failed.\n"); goto failed_shutdown; } - r = rados_ioctx_create(rbd_data->cluster, o->pool_name, - &rbd_data->io_ctx); + r = rados_ioctx_create(rbd->cluster, o->pool_name, &rbd->io_ctx); if (r < 0) { log_err("rados_ioctx_create failed.\n"); goto failed_shutdown; } - r = rbd_open(rbd_data->io_ctx, o->rbd_name, &rbd_data->image, - NULL /*snap */ ); + r = rbd_open(rbd->io_ctx, o->rbd_name, &rbd->image, NULL /*snap */ ); if (r < 0) { log_err("rbd_open failed.\n"); goto failed_open; @@ -144,34 +143,34 @@ static int _fio_rbd_connect(struct thread_data *td) return 0; failed_open: - rados_ioctx_destroy(rbd_data->io_ctx); - rbd_data->io_ctx = NULL; + rados_ioctx_destroy(rbd->io_ctx); + rbd->io_ctx = NULL; failed_shutdown: - rados_shutdown(rbd_data->cluster); - rbd_data->cluster = NULL; + rados_shutdown(rbd->cluster); + rbd->cluster = NULL; failed_early: return 1; } -static void _fio_rbd_disconnect(struct rbd_data *rbd_data) +static void _fio_rbd_disconnect(struct rbd_data *rbd) { - if (!rbd_data) + if (!rbd) return; /* shutdown everything */ - if (rbd_data->image) { - rbd_close(rbd_data->image); - rbd_data->image = NULL; + if (rbd->image) { + rbd_close(rbd->image); + rbd->image = NULL; } - if (rbd_data->io_ctx) { - rados_ioctx_destroy(rbd_data->io_ctx); - rbd_data->io_ctx = NULL; + if (rbd->io_ctx) { + rados_ioctx_destroy(rbd->io_ctx); + rbd->io_ctx = NULL; } - if (rbd_data->cluster) { - rados_shutdown(rbd_data->cluster); - rbd_data->cluster = NULL; + if (rbd->cluster) { + rados_shutdown(rbd->cluster); + rbd->cluster = NULL; } } @@ -192,24 +191,25 @@ static void _fio_rbd_finish_aiocb(rbd_completion_t comp, void *data) io_u->resid = io_u->xfer_buflen; } else io_u->error = 0; + + fri->io_complete = 1; } static struct io_u *fio_rbd_event(struct thread_data *td, int event) { - struct rbd_data *rbd_data = td->io_ops->data; + struct rbd_data *rbd = td->io_ops->data; - return rbd_data->aio_events[event]; + return rbd->aio_events[event]; } -static inline int fri_check_complete(struct rbd_data *rbd_data, - struct io_u *io_u, +static inline int fri_check_complete(struct rbd_data *rbd, struct io_u *io_u, unsigned int *events) { struct fio_rbd_iou *fri = io_u->engine_data; - if (rbd_aio_is_complete(fri->completion)) { + if (fri->io_complete) { fri->io_seen = 1; - rbd_data->aio_events[*events] = io_u; + rbd->aio_events[*events] = io_u; (*events)++; rbd_aio_release(fri->completion); @@ -219,32 +219,88 @@ static inline int fri_check_complete(struct rbd_data *rbd_data, return 0; } +static inline int rbd_io_u_seen(struct io_u *io_u) +{ + struct fio_rbd_iou *fri = io_u->engine_data; + + return fri->io_seen; +} + +static void rbd_io_u_wait_complete(struct io_u *io_u) +{ + struct fio_rbd_iou *fri = io_u->engine_data; + + rbd_aio_wait_for_complete(fri->completion); +} + +static int rbd_io_u_cmp(const void *p1, const void *p2) +{ + const struct io_u **a = (const struct io_u **) p1; + const struct io_u **b = (const struct io_u **) p2; + uint64_t at, bt; + + at = utime_since_now(&(*a)->start_time); + bt = utime_since_now(&(*b)->start_time); + + if (at < bt) + return -1; + else if (at == bt) + return 0; + else + return 1; +} + static int rbd_iter_events(struct thread_data *td, unsigned int *events, unsigned int min_evts, int wait) { - struct rbd_data *rbd_data = td->io_ops->data; + struct rbd_data *rbd = td->io_ops->data; unsigned int this_events = 0; struct io_u *io_u; - int i; + int i, sidx; + sidx = 0; io_u_qiter(&td->io_u_all, io_u, i) { - struct fio_rbd_iou *fri = io_u->engine_data; - if (!(io_u->flags & IO_U_F_FLIGHT)) continue; - if (fri->io_seen) + if (rbd_io_u_seen(io_u)) continue; - if (fri_check_complete(rbd_data, io_u, events)) + if (fri_check_complete(rbd, io_u, events)) this_events++; - else if (wait) { - rbd_aio_wait_for_complete(fri->completion); + else if (wait) + rbd->sort_events[sidx++] = io_u; + } + + if (!wait || !sidx) + return this_events; - if (fri_check_complete(rbd_data, io_u, events)) - this_events++; + /* + * Sort events, oldest issue first, then wait on as many as we + * need in order of age. If we have enough events, stop waiting, + * and just check if any of the older ones are done. + */ + if (sidx > 1) + qsort(rbd->sort_events, sidx, sizeof(struct io_u *), rbd_io_u_cmp); + + for (i = 0; i < sidx; i++) { + io_u = rbd->sort_events[i]; + + if (fri_check_complete(rbd, io_u, events)) { + this_events++; + continue; } + + /* + * Stop waiting when we have enough, but continue checking + * all pending IOs if they are complete. + */ if (*events >= min_evts) - break; + continue; + + rbd_io_u_wait_complete(io_u); + + if (fri_check_complete(rbd, io_u, events)) + this_events++; } return this_events; @@ -276,13 +332,14 @@ static int fio_rbd_getevents(struct thread_data *td, unsigned int min, static int fio_rbd_queue(struct thread_data *td, struct io_u *io_u) { - struct rbd_data *rbd_data = td->io_ops->data; + struct rbd_data *rbd = td->io_ops->data; struct fio_rbd_iou *fri = io_u->engine_data; int r = -1; fio_ro_check(td, io_u); fri->io_seen = 0; + fri->io_complete = 0; r = rbd_aio_create_completion(fri, _fio_rbd_finish_aiocb, &fri->completion); @@ -292,32 +349,30 @@ static int fio_rbd_queue(struct thread_data *td, struct io_u *io_u) } if (io_u->ddir == DDIR_WRITE) { - r = rbd_aio_write(rbd_data->image, io_u->offset, - io_u->xfer_buflen, io_u->xfer_buf, - fri->completion); + r = rbd_aio_write(rbd->image, io_u->offset, io_u->xfer_buflen, + io_u->xfer_buf, fri->completion); if (r < 0) { log_err("rbd_aio_write failed.\n"); goto failed_comp; } } else if (io_u->ddir == DDIR_READ) { - r = rbd_aio_read(rbd_data->image, io_u->offset, - io_u->xfer_buflen, io_u->xfer_buf, - fri->completion); + r = rbd_aio_read(rbd->image, io_u->offset, io_u->xfer_buflen, + io_u->xfer_buf, fri->completion); if (r < 0) { log_err("rbd_aio_read failed.\n"); goto failed_comp; } } else if (io_u->ddir == DDIR_TRIM) { - r = rbd_aio_discard(rbd_data->image, io_u->offset, - io_u->xfer_buflen, fri->completion); + r = rbd_aio_discard(rbd->image, io_u->offset, + io_u->xfer_buflen, fri->completion); if (r < 0) { log_err("rbd_aio_discard failed.\n"); goto failed_comp; } } else if (io_u->ddir == DDIR_SYNC) { - r = rbd_aio_flush(rbd_data->image, fri->completion); + r = rbd_aio_flush(rbd->image, fri->completion); if (r < 0) { log_err("rbd_flush failed.\n"); goto failed_comp; @@ -355,35 +410,35 @@ failed: static void fio_rbd_cleanup(struct thread_data *td) { - struct rbd_data *rbd_data = td->io_ops->data; + struct rbd_data *rbd = td->io_ops->data; - if (rbd_data) { - _fio_rbd_disconnect(rbd_data); - free(rbd_data->aio_events); - free(rbd_data); + if (rbd) { + _fio_rbd_disconnect(rbd); + free(rbd->aio_events); + free(rbd->sort_events); + free(rbd); } - } static int fio_rbd_setup(struct thread_data *td) { - int r = 0; rbd_image_info_t info; struct fio_file *f; - struct rbd_data *rbd_data = NULL; + struct rbd_data *rbd = NULL; int major, minor, extra; + int r; /* log version of librbd. No cluster connection required. */ rbd_version(&major, &minor, &extra); log_info("rbd engine: RBD version: %d.%d.%d\n", major, minor, extra); /* allocate engine specific structure to deal with librbd. */ - r = _fio_setup_rbd_data(td, &rbd_data); + r = _fio_setup_rbd_data(td, &rbd); if (r) { log_err("fio_setup_rbd_data failed.\n"); goto cleanup; } - td->io_ops->data = rbd_data; + td->io_ops->data = rbd; /* librbd does not allow us to run first in the main thread and later * in a fork child. It needs to be the same process context all the @@ -402,7 +457,7 @@ static int fio_rbd_setup(struct thread_data *td) } /* get size of the RADOS block device */ - r = rbd_stat(rbd_data->image, &info, sizeof(info)); + r = rbd_stat(rbd->image, &info, sizeof(info)); if (r < 0) { log_err("rbd_status failed.\n"); goto disconnect; @@ -424,11 +479,11 @@ static int fio_rbd_setup(struct thread_data *td) /* disconnect, then we were only connected to determine * the size of the RBD. */ - _fio_rbd_disconnect(rbd_data); + _fio_rbd_disconnect(rbd); return 0; disconnect: - _fio_rbd_disconnect(rbd_data); + _fio_rbd_disconnect(rbd); cleanup: fio_rbd_cleanup(td); return r; @@ -442,9 +497,9 @@ static int fio_rbd_open(struct thread_data *td, struct fio_file *f) static int fio_rbd_invalidate(struct thread_data *td, struct fio_file *f) { #if defined(CONFIG_RBD_INVAL) - struct rbd_data *rbd_data = td->io_ops->data; + struct rbd_data *rbd = td->io_ops->data; - return rbd_invalidate_cache(rbd_data->image); + return rbd_invalidate_cache(rbd->image); #else return 0; #endif