diff options
author | Amar Tumballi <amar@gluster.com> | 2011-07-01 04:41:54 +0000 |
---|---|---|
committer | Anand Avati <avati@gluster.com> | 2011-07-14 01:01:44 -0700 |
commit | 8d18bc22d1cfc8a481bc8d20da5843c65b567226 (patch) | |
tree | 680b1a4884b382aef8a8789a6dc6623797965fc4 /xlators/cluster/dht/src/dht-rebalance.c | |
parent | 3f76fe277ec0a826bf3052134ad877ee65ce9e79 (diff) |
distribute: bring in feature to do rebalance from inside
do the 'rebalance' on a file with 'setxattr()' with
'distribute.migrate-data' key. the valid 'value' is "force" or
anything else.
current implementation of 'rebalance' is same as 'glusterd-rebalance',
and hence comes with limitation of not rebalancing hardlinks, non-regular
files, and files with open-fds.
Signed-off-by: Amar Tumballi <amar@gluster.com>
Signed-off-by: Anand Avati <avati@gluster.com>
BUG: 3070 (After 'rebalance', the gfid of the file gets changed)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=3070
Diffstat (limited to 'xlators/cluster/dht/src/dht-rebalance.c')
-rw-r--r-- | xlators/cluster/dht/src/dht-rebalance.c | 518 |
1 files changed, 518 insertions, 0 deletions
diff --git a/xlators/cluster/dht/src/dht-rebalance.c b/xlators/cluster/dht/src/dht-rebalance.c new file mode 100644 index 00000000000..6dea40f0b57 --- /dev/null +++ b/xlators/cluster/dht/src/dht-rebalance.c @@ -0,0 +1,518 @@ +/* + Copyright (c) 2011 Gluster, Inc. <http://www.gluster.com> + This file is part of GlusterFS. + + GlusterFS is free software; you can redistribute it and/or modify + it under the terms of the GNU Affero General Public License as published + by the Free Software Foundation; either version 3 of the License, + or (at your option) any later version. + + GlusterFS is distributed in the hope that it will be useful, but + WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Affero General Public License for more details. + + You should have received a copy of the GNU Affero General Public License + along with this program. If not, see + <http://www.gnu.org/licenses/>. +*/ + + +#ifndef _CONFIG_H +#define _CONFIG_H +#include "config.h" +#endif + +#include "dht-common.h" + +#define GF_DISK_SECTOR_SIZE 512 +#define DHT_REBALANCE_PID 4242 /* Change it if required */ +#define DHT_REBALANCE_BLKSIZE (128 * 1024) +#define DHT_MIGRATE_EVEN_IF_LINK_EXISTS 1 + +static int +dht_write_with_holes (xlator_t *to, fd_t *fd, struct iovec *vec, int count, + int32_t size, off_t offset, struct iobref *iobref) +{ + int i = 0; + int ret = -1; + int start_idx = 0; + int tmp_offset = 0; + int write_needed = 0; + int buf_len = 0; + int size_pending = 0; + char *buf = NULL; + + /* loop through each vector */ + for (i = 0; i < count; i++) { + buf = vec[i].iov_base; + buf_len = vec[i].iov_len; + + for (start_idx = 0; (start_idx + GF_DISK_SECTOR_SIZE) <= buf_len; + start_idx += GF_DISK_SECTOR_SIZE) { + + if (mem_0filled (buf + start_idx, GF_DISK_SECTOR_SIZE) != 0) { + write_needed = 1; + continue; + } + + if (write_needed) { + ret = syncop_write (to, fd, (buf + tmp_offset), + (start_idx - tmp_offset), + (offset + tmp_offset), + iobref); + if (ret < 0) + goto out; + + write_needed = 0; + } + tmp_offset = start_idx + GF_DISK_SECTOR_SIZE; + } + + if ((start_idx < buf_len) || write_needed) { + /* This means, last chunk is not yet written.. write it */ + ret = syncop_write (to, fd, (buf + tmp_offset), + (buf_len - tmp_offset), + (offset + tmp_offset), iobref); + if (ret < 0) + goto out; + } + + size_pending = (size - buf_len); + if (!size_pending) + break; + } + + /* do it regardless of all the above cases as we had to 'write' the + given number of bytes */ + ret = syncop_ftruncate (to, fd, offset + size); + if (ret) { + gf_log (THIS->name, GF_LOG_WARNING, + "failed to perform truncate on %s", to->name); + goto out; + } + + ret = size; +out: + return ret; + +} + +static inline int +__is_file_migratable (xlator_t *this, loc_t *loc, dict_t *rsp_dict, + struct iatt *stbuf) +{ + int ret = -1; + int open_fd_count = 0; + + if (!IA_ISREG (stbuf->ia_type)) { + gf_log (this->name, GF_LOG_WARNING, + "%s: migrate-file called on non-regular entry (0%o)", + loc->path, stbuf->ia_type); + ret = -1; + goto out; + } + + if (stbuf->ia_nlink > 1) { + /* TODO : support migrating hardlinks */ + gf_log (this->name, GF_LOG_WARNING, "%s: file has hardlinks", + loc->path); + ret = -ENOTSUP; + goto out; + } + + ret = dict_get_int32 (rsp_dict, GLUSTERFS_OPEN_FD_COUNT, &open_fd_count); + if (!ret && (open_fd_count > 0)) { + /* TODO: support migration of files with open fds */ + gf_log (this->name, GF_LOG_WARNING, + "%s: file has open fds, not attempting migration", + loc->path); + goto out; + } + ret = 0; + +out: + return ret; +} + +static inline int +__dht_rebalance_create_dst_file (xlator_t *to, loc_t *loc, struct iatt *stbuf, + dict_t *dict, fd_t **dst_fd, int *need_unlink) +{ + xlator_t *this = NULL; + int ret = -1; + mode_t mode = 0; + fd_t *fd = NULL; + struct iatt new_stbuf = {0,}; + + this = THIS; + + ret = dict_set_static_bin (dict, "gfid-req", stbuf->ia_gfid, 16); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, + "%s: failed to set gfid in dict for create", loc->path); + goto out; + } + + fd = fd_create (loc->inode, DHT_REBALANCE_PID); + if (!fd) { + gf_log (this->name, GF_LOG_ERROR, + "%s: fd create failed (destination)", loc->path); + ret = -1; + goto out; + } + + ret = syncop_lookup (to, loc, NULL, &new_stbuf, NULL, NULL); + if (ret) { + gf_log (this->name, GF_LOG_DEBUG, "failed to lookup %s on %s", + loc->path, to->name); + + mode = st_mode_from_ia (stbuf->ia_prot, stbuf->ia_type); + ret = syncop_create (to, loc, O_WRONLY, mode, fd, dict); + if (ret < 0) { + gf_log (this->name, GF_LOG_ERROR, + "failed to create %s on %s", loc->path, to->name); + goto out; + } + + *need_unlink = 1; + goto done; + } + + /* File exits in the destination, just do the open if gfid matches */ + if (uuid_compare (stbuf->ia_gfid, new_stbuf.ia_gfid) != 0) { + gf_log (this->name, GF_LOG_ERROR, + "file %s exits in %s with different gfid", + loc->path, to->name); + fd_unref (fd); + goto out; + } + + ret = syncop_open (to, loc, O_WRONLY, fd); + if (ret < 0) { + gf_log (this->name, GF_LOG_ERROR, + "failed to open file %s on %s", + loc->path, to->name); + fd_unref (fd); + goto out; + } +done: + if (dst_fd) + *dst_fd = fd; + + /* success */ + ret = 0; + +out: + return ret; +} + +static inline int +__dht_check_free_space (xlator_t *to, xlator_t *from, loc_t *loc, + struct iatt *stbuf) +{ + struct statvfs src_statfs = {0,}; + struct statvfs dst_statfs = {0,}; + int ret = -1; + xlator_t *this = NULL; + + this = THIS; + + ret = syncop_statfs (from, loc, &src_statfs); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, + "failed to get statfs of %s on %s", + loc->path, from->name); + goto out; + } + + ret = syncop_statfs (to, loc, &dst_statfs); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, + "failed to get statfs of %s on %s", + loc->path, to->name); + goto out; + } + if (((dst_statfs.f_bavail * + dst_statfs.f_bsize) / GF_DISK_SECTOR_SIZE) > + (((src_statfs.f_bavail * src_statfs.f_bsize) / + GF_DISK_SECTOR_SIZE) - stbuf->ia_blocks)) { + gf_log (this->name, GF_LOG_WARNING, + "data movement attempted from node (%s) with" + " higher disk space to a node (%s) with " + "lesser disk space (%s)", from->name, + to->name, loc->path); + + ret = -1; + goto out; + } + + ret = 0; +out: + return ret; +} + +static inline int +__dht_rebalane_migrate_data (xlator_t *from, xlator_t *to, fd_t *src, fd_t *dst, + int hole_exists) +{ + int ret = -1; + int count = 0; + off_t offset = 0; + struct iovec *vector = NULL; + struct iobref *iobref = NULL; + + while (1) { + ret = syncop_readv (from, src, DHT_REBALANCE_BLKSIZE, + offset, &vector, &count, &iobref); + if (!ret || (ret < 0)) { + break; + } + + if (hole_exists) + ret = dht_write_with_holes (to, dst, vector, count, + ret, offset, iobref); + else + ret = syncop_writev (to, dst, vector, count, + offset, iobref); + if (ret < 0) { + break; + } + offset += ret; + + if (vector) + GF_FREE (vector); + if (iobref) + iobref_unref (iobref); + iobref = NULL; + vector = NULL; + } + if (iobref) + iobref_unref (iobref); + if (vector) + GF_FREE (vector); + + if (ret >= 0) + ret = 0; + + return ret; +} + +int +dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to, + int flag) +{ + int ret = -1; + struct iatt new_stbuf = {0,}; + struct iatt stbuf = {0,}; + fd_t *src_fd = NULL; + fd_t *dst_fd = NULL; + dict_t *dict = NULL; + dict_t *xattr = NULL; + dict_t *rsp_dict = NULL; + int file_has_holes = 0; + int need_unlink = 0; + + gf_log (this->name, GF_LOG_INFO, "%s: attempting to move from %s to %s", + loc->path, from->name, to->name); + + dict = dict_new (); + if (!dict) + goto out; + + ret = dict_set_int32 (dict, GLUSTERFS_OPEN_FD_COUNT, 4); + if (ret) + gf_log (this->name, GF_LOG_WARNING, + "%s: failed to set fd-count key in dict, may attempt " + "migration of file which has open fds", loc->path); + + ret = syncop_lookup (from, loc, dict, &stbuf, &rsp_dict, NULL); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, "failed to lookup %s on %s", + loc->path, from->name); + goto out; + } + + /* Check if file can be migrated */ + ret = __is_file_migratable (this, loc, rsp_dict, &stbuf); + if (ret) + goto out; + + /* create the destination */ + ret = __dht_rebalance_create_dst_file (to, loc, &stbuf, dict, &dst_fd, + &need_unlink); + if (ret) + goto out; + + /* Should happen on all files when 'force' option is not given */ + if (flag != DHT_MIGRATE_EVEN_IF_LINK_EXISTS) { + ret = __dht_check_free_space (to, from, loc, &stbuf); + if (ret) + goto out; + } + + /* Try to preserve 'holes' while migrating data */ + if (stbuf.ia_size > (stbuf.ia_blocks * GF_DISK_SECTOR_SIZE)) + file_has_holes = 1; + + src_fd = fd_create (loc->inode, DHT_REBALANCE_PID); + if (!src_fd) { + gf_log (this->name, GF_LOG_ERROR, + "%s: fd create failed (source)", loc->path); + ret = -1; + goto out; + } + + ret = syncop_open (from, loc, O_RDONLY, src_fd); + if (ret == -1) { + gf_log (this->name, GF_LOG_ERROR, + "failed to open file %s on %s", + loc->path, from->name); + goto out; + } + + /* All I/O happens in this function */ + ret = __dht_rebalane_migrate_data (from, to, src_fd, dst_fd, + file_has_holes); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, "%s: failed to migrate data", + loc->path); + goto out; + } + + ret = syncop_lookup (from, loc, NULL, &new_stbuf, NULL, NULL); + if (ret < 0) { + /* Failed to get the stat info */ + gf_log (this->name, GF_LOG_ERROR, + "failed to lookup file %s on %s", + loc->path, from->name); + need_unlink = 0; + goto out; + } + + /* No need to rebalance, if there is some + activity on source file */ + if (new_stbuf.ia_mtime != stbuf.ia_mtime) { + gf_log (this->name, GF_LOG_WARNING, + "%s: ignoring destination file as source has " + "undergone some changes while migration was happening", + loc->path); + ret = -1; + goto out; + } + + ret = syncop_setattr (to, loc, &new_stbuf, + (GF_SET_ATTR_UID | GF_SET_ATTR_GID | + GF_SET_ATTR_MODE | GF_SET_ATTR_ATIME | + GF_SET_ATTR_MTIME), NULL, NULL); + if (ret) { + gf_log (this->name, GF_LOG_WARNING, + "%s: failed to perform setattr on %s", + loc->path, to->name); + } + + ret = syncop_listxattr (from, loc, &xattr); + if (ret == -1) + gf_log (this->name, GF_LOG_WARNING, + "%s: failed to get xattr from %s", loc->path, from->name); + + ret = syncop_setxattr (to, loc, xattr, 0); + if (ret == -1) + gf_log (this->name, GF_LOG_WARNING, + "%s: failed to set xattr on %s", loc->path, to->name); + + /* rebalance complete */ + syncop_close (dst_fd); + syncop_close (src_fd); + syncop_unlink (from, loc); + need_unlink = 0; + + gf_log (this->name, GF_LOG_INFO, + "completed migration of %s from subvolume %s to %s", + loc->path, from->name, to->name); + + ret = 0; +out: + if (dict) + dict_unref (dict); + + if (ret) { + if (dst_fd) + syncop_close (dst_fd); + if (src_fd) + syncop_close (src_fd); + if (need_unlink) + syncop_unlink (to, loc); + } + + return ret; +} + +static int +rebalance_task (void *data) +{ + int ret = -1; + xlator_t *this = NULL; + dht_local_t *local = NULL; + call_frame_t *frame = NULL; + + frame = data; + this = THIS; + + local = frame->local; + + /* This function is 'synchrounous', hence if it returns, + we are done with the task */ + ret = dht_migrate_file (THIS, &local->loc, local->from_subvol, + local->to_subvol, local->flags); + + return ret; +} + +static int +rebalance_task_completion (int op_ret, call_frame_t *sync_frame, void *data) +{ + int ret = -1; + uint64_t layout_int = 0; + dht_layout_t *layout = 0; + xlator_t *this = NULL; + dht_local_t *local = NULL; + + this = THIS; + local = sync_frame->local; + + if (!op_ret) { + /* Make sure we have valid 'layout' in inode ctx + after the operation */ + ret = inode_ctx_del (local->loc.inode, this, &layout_int); + if (!ret && layout_int) { + layout = (dht_layout_t *)(long)layout_int; + dht_layout_unref (this, layout); + } + + ret = dht_layout_preset (this, local->to_subvol, + local->loc.inode); + if (ret) + gf_log (this->name, GF_LOG_WARNING, + "%s: failed to set inode ctx", local->loc.path); + } + + /* if success, errno is not checked, + if ret is -1, then let errno be 'ENOTSUP' */ + DHT_STACK_UNWIND (setxattr, sync_frame, op_ret, ENOTSUP); + return 0; +} + +int +dht_start_rebalance_task (xlator_t *this, call_frame_t *frame) +{ + int ret = -1; + dht_conf_t *conf = NULL; + + conf = this->private; + + ret = synctask_new (conf->env, rebalance_task, + rebalance_task_completion, + frame, frame); + return ret; +} |