diff options
Diffstat (limited to 'xlators/cluster/dht/src/dht-rebalance.c')
| -rw-r--r-- | xlators/cluster/dht/src/dht-rebalance.c | 706 |
1 files changed, 65 insertions, 641 deletions
diff --git a/xlators/cluster/dht/src/dht-rebalance.c b/xlators/cluster/dht/src/dht-rebalance.c index d850eef62ab..8ba8082bd86 100644 --- a/xlators/cluster/dht/src/dht-rebalance.c +++ b/xlators/cluster/dht/src/dht-rebalance.c @@ -14,7 +14,6 @@ #include <signal.h> #include <glusterfs/events.h> #include "glusterfs/compat-errno.h" // for ENODATA on BSD -#include <string.h> #define GF_DISK_SECTOR_SIZE 512 #define DHT_REBALANCE_PID 4242 /* Change it if required */ @@ -610,26 +609,23 @@ __dht_rebalance_create_dst_file(xlator_t *this, xlator_t *to, xlator_t *from, goto out; } - if (!!dht_is_tier_xlator(this)) { - xdata = dict_new(); - if (!xdata) { - *fop_errno = ENOMEM; - ret = -1; - gf_msg(this->name, GF_LOG_ERROR, ENOMEM, - DHT_MSG_MIGRATE_FILE_FAILED, "%s: dict_new failed)", - loc->path); - goto out; - } + xdata = dict_new(); + if (!xdata) { + *fop_errno = ENOMEM; + ret = -1; + gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MIGRATE_FILE_FAILED, + "%s: dict_new failed)", loc->path); + goto out; + } - ret = dict_set_int32(xdata, GF_CLEAN_WRITE_PROTECTION, 1); - if (ret) { - *fop_errno = ENOMEM; - ret = -1; - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, - "%s: failed to set dictionary value: key = %s ", loc->path, - GF_CLEAN_WRITE_PROTECTION); - goto out; - } + ret = dict_set_int32_sizen(xdata, GF_CLEAN_WRITE_PROTECTION, 1); + if (ret) { + *fop_errno = ENOMEM; + ret = -1; + gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, + "%s: failed to set dictionary value: key = %s ", loc->path, + GF_CLEAN_WRITE_PROTECTION); + goto out; } ret = syncop_lookup(to, loc, &new_stbuf, NULL, xdata, NULL); @@ -1097,7 +1093,7 @@ __dht_rebalance_migrate_data(xlator_t *this, gf_defrag_info_t *defrag, break; } - if (!conf->force_migration && !dht_is_tier_xlator(this)) { + if (!conf->force_migration) { if (!xdata) { xdata = dict_new(); if (!xdata) { @@ -1537,21 +1533,6 @@ dht_migrate_file(xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to, goto out; } - /* If defrag is NULL, it should be assumed that migration is triggered - * from client using the trusted.distribute.migrate-data virtual xattr - */ - defrag = conf->defrag; - - /* migration of files from clients is restricted to non-tiered clients - * for now */ - if (!defrag && dht_is_tier_xlator(this)) { - ret = ENOTSUP; - goto out; - } - - if (defrag && defrag->tier_conf.is_tier) - log_level = GF_LOG_TRACE; - gf_log(this->name, log_level, "%s: attempting to move from %s to %s", loc->path, from->name, to->name); @@ -2301,14 +2282,12 @@ out: } } - if (!dht_is_tier_xlator(this)) { - lk_ret = syncop_removexattr(to, loc, GF_PROTECT_FROM_EXTERNAL_WRITES, - NULL, NULL); - if (lk_ret && (lk_ret != -ENODATA) && (lk_ret != -ENOATTR)) { - gf_msg(this->name, GF_LOG_WARNING, -lk_ret, 0, - "%s: removexattr failed key %s", loc->path, - GF_PROTECT_FROM_EXTERNAL_WRITES); - } + lk_ret = syncop_removexattr(to, loc, GF_PROTECT_FROM_EXTERNAL_WRITES, NULL, + NULL); + if (lk_ret && (lk_ret != -ENODATA) && (lk_ret != -ENOATTR)) { + gf_msg(this->name, GF_LOG_WARNING, -lk_ret, 0, + "%s: removexattr failed key %s", loc->path, + GF_PROTECT_FROM_EXTERNAL_WRITES); } if (dict) @@ -2895,8 +2874,7 @@ gf_defrag_migrate_single_file(void *opaque) if (defrag->stats == _gf_true) { gettimeofday(&end, NULL); - elapsed = (end.tv_sec - start.tv_sec) * 1e6 + - (end.tv_usec - start.tv_usec); + elapsed = gf_tvdiff(&start, &end); gf_log(this->name, GF_LOG_INFO, "Migration of " "file:%s size:%" PRIu64 @@ -3075,7 +3053,7 @@ int static gf_defrag_get_entry(xlator_t *this, int i, dht_conf_t *conf, gf_defrag_info_t *defrag, fd_t *fd, dict_t *migrate_data, struct dir_dfmeta *dir_dfmeta, dict_t *xattr_req, - int *should_commit_hash, int *perrno) + int *perrno) { int ret = 0; char is_linkfile = 0; @@ -3279,7 +3257,7 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, int dfc_index = 0; int throttle_up = 0; struct dir_dfmeta *dir_dfmeta = NULL; - int should_commit_hash = 1; + xlator_t *old_THIS = NULL; gf_log(this->name, GF_LOG_INFO, "migrate data called on %s", loc->path); gettimeofday(&dir_start, NULL); @@ -3292,6 +3270,9 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, goto out; } + old_THIS = THIS; + THIS = this; + dir_dfmeta = GF_CALLOC(1, sizeof(*dir_dfmeta), gf_common_mt_pointer); if (!dir_dfmeta) { gf_log(this->name, GF_LOG_ERROR, "dir_dfmeta is NULL"); @@ -3456,7 +3437,7 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, ret = gf_defrag_get_entry(this, dfc_index, &container, loc, conf, defrag, dir_dfmeta->lfd[dfc_index], migrate_data, dir_dfmeta, xattr_req, - &should_commit_hash, perrno); + perrno); if (defrag->defrag_status == GF_DEFRAG_STATUS_STOPPED) { goto out; @@ -3500,24 +3481,19 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, } gettimeofday(&end, NULL); - elapsed = (end.tv_sec - dir_start.tv_sec) * 1e6 + - (end.tv_usec - dir_start.tv_usec); + elapsed = gf_tvdiff(&dir_start, &end); gf_log(this->name, GF_LOG_INFO, "Migration operation on dir %s took " "%.2f secs", loc->path, elapsed / 1e6); ret = 0; out: - + THIS = old_THIS; gf_defrag_free_dir_dfmeta(dir_dfmeta, local_subvols_cnt); if (xattr_req) dict_unref(xattr_req); - if (ret == 0 && should_commit_hash == 0) { - ret = 2; - } - /* It does not matter if it errored out - this number is * used to calculate rebalance estimated time to complete. * No locking required as dirs are processed by a single thread. @@ -3525,6 +3501,7 @@ out: defrag->num_dirs_processed++; return ret; } + int gf_defrag_settle_hash(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, dict_t *fix_layout) @@ -3539,7 +3516,6 @@ gf_defrag_settle_hash(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, * rebalance is complete. */ if (defrag->cmd == GF_DEFRAG_CMD_START_LAYOUT_FIX || - defrag->cmd == GF_DEFRAG_CMD_START_DETACH_TIER || defrag->cmd == GF_DEFRAG_CMD_DETACH_START) { return 0; } @@ -3585,114 +3561,6 @@ gf_defrag_settle_hash(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, return 0; } -/* Function for doing a named lookup on file inodes during an attach tier - * So that a hardlink lookup heal i.e gfid to parent gfid lookup heal - * happens on pre-existing data. This is required so that the ctr database has - * hardlinks of all the exisitng file in the volume. CTR xlator on the - * brick/server side does db update/insert of the hardlink on a namelookup. - * Currently the namedlookup is done synchronous to the fixlayout that is - * triggered by attach tier. This is not performant, adding more time to - * fixlayout. The performant approach is record the hardlinks on a compressed - * datastore and then do the namelookup asynchronously later, giving the ctr db - * eventual consistency - * */ -int -gf_fix_layout_tier_attach_lookup(xlator_t *this, loc_t *parent_loc, - gf_dirent_t *file_dentry) -{ - int ret = -1; - dict_t *lookup_xdata = NULL; - dht_conf_t *conf = NULL; - loc_t file_loc = { - 0, - }; - struct iatt iatt = { - 0, - }; - - GF_VALIDATE_OR_GOTO("tier", this, out); - - GF_VALIDATE_OR_GOTO(this->name, parent_loc, out); - - GF_VALIDATE_OR_GOTO(this->name, file_dentry, out); - - GF_VALIDATE_OR_GOTO(this->name, this->private, out); - - if (!parent_loc->inode) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR, - "%s/%s parent is NULL", parent_loc->path, file_dentry->d_name); - goto out; - } - - conf = this->private; - - loc_wipe(&file_loc); - - if (gf_uuid_is_null(file_dentry->d_stat.ia_gfid)) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR, - "%s/%s gfid not present", parent_loc->path, file_dentry->d_name); - goto out; - } - - gf_uuid_copy(file_loc.gfid, file_dentry->d_stat.ia_gfid); - - if (gf_uuid_is_null(parent_loc->gfid)) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR, - "%s/%s" - " gfid not present", - parent_loc->path, file_dentry->d_name); - goto out; - } - - gf_uuid_copy(file_loc.pargfid, parent_loc->gfid); - - ret = dht_build_child_loc(this, &file_loc, parent_loc, file_dentry->d_name); - if (ret) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR, - "Child loc build failed"); - ret = -1; - goto out; - } - - lookup_xdata = dict_new(); - if (!lookup_xdata) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR, - "Failed creating lookup dict for %s", file_dentry->d_name); - goto out; - } - - ret = dict_set_int32(lookup_xdata, CTR_ATTACH_TIER_LOOKUP, 1); - if (ret) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR, - "Failed to set lookup flag"); - goto out; - } - - gf_uuid_copy(file_loc.parent->gfid, parent_loc->gfid); - - /* Sending lookup to cold tier only */ - ret = syncop_lookup(conf->subvolumes[0], &file_loc, &iatt, NULL, - lookup_xdata, NULL); - if (ret) { - /* If the file does not exist on the cold tier than it must */ - /* have been discovered on the hot tier. This is not an error. */ - gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_LOG_TIER_STATUS, - "%s lookup to cold tier on attach heal failed", file_loc.path); - goto out; - } - - ret = 0; - -out: - - loc_wipe(&file_loc); - - if (lookup_xdata) - dict_unref(lookup_xdata); - - return ret; -} - int gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, dict_t *fix_layout, dict_t *migrate_data) @@ -3712,7 +3580,6 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, }; inode_t *linked_inode = NULL, *inode = NULL; dht_conf_t *conf = NULL; - int should_commit_hash = 1; int perrno = 0; conf = this->private; @@ -3815,16 +3682,6 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, if (!strcmp(entry->d_name, ".") || !strcmp(entry->d_name, "..")) continue; if (!IA_ISDIR(entry->d_stat.ia_type)) { - /* If its a fix layout during the attach - * tier operation do lookups on files - * on cold subvolume so that there is a - * CTR DB Lookup Heal triggered on existing - * data. - * */ - if (defrag->cmd == GF_DEFRAG_CMD_START_TIER) { - gf_fix_layout_tier_attach_lookup(this, loc, entry); - } - continue; } loc_wipe(&entry_loc); @@ -3841,8 +3698,6 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, goto out; } else { - should_commit_hash = 0; - continue; } } @@ -3905,7 +3760,6 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, ret = -1; goto out; } else { - should_commit_hash = 0; continue; } } @@ -3923,7 +3777,7 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, goto out; } - if (ret && ret != 2) { + if (ret) { gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LAYOUT_FIX_FAILED, "Fix layout failed for %s", entry_loc.path); @@ -3990,11 +3844,10 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, } } - if ((defrag->cmd != GF_DEFRAG_CMD_START_TIER) && - (defrag->cmd != GF_DEFRAG_CMD_START_LAYOUT_FIX)) { + if (defrag->cmd != GF_DEFRAG_CMD_START_LAYOUT_FIX) { ret = gf_defrag_process_dir(this, defrag, loc, migrate_data, &perrno); - if (ret && (ret != 2)) { + if (ret) { if (perrno == ENOENT || perrno == ESTALE) { ret = 0; goto out; @@ -4010,18 +3863,13 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, if (conf->decommission_in_progress) { goto out; } - - should_commit_hash = 0; } - } else if (ret == 2) { - should_commit_hash = 0; } } gf_msg_trace(this->name, 0, "fix layout called on %s", loc->path); - if (should_commit_hash && - gf_defrag_settle_hash(this, defrag, loc, fix_layout) != 0) { + if (gf_defrag_settle_hash(this, defrag, loc, fix_layout) != 0) { defrag->total_failures++; gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SETTLE_HASH_FAILED, @@ -4045,372 +3893,6 @@ out: if (fd) fd_unref(fd); - if (ret == 0 && should_commit_hash == 0) { - ret = 2; - } - - return ret; -} - -int -gf_defrag_fix_layout_puredist(xlator_t *this, gf_defrag_info_t *defrag, - loc_t *loc, dict_t *fix_layout, - dict_t *migrate_data) -{ - int ret = -1; - loc_t entry_loc = { - 0, - }; - fd_t *fd = NULL; - inode_t *linked_inode = NULL, *inode = NULL; - dht_conf_t *conf = NULL; - int should_commit_hash = 1; - int perrno = 0; - /* absolute brick path length */ - int brick_len = 0; - /* dir path length (relative to gluster mount) */ - int dir_len = 0; - /* absolute dir path length */ - int total_len = 0; - struct dirent *entry = NULL; - struct dirent scratch[2] = {{ - 0, - }}; - DIR *dirp = NULL; - int full_entry_length = 0; - int entry_len = 0; - char full_entry_path[4096] = { - 0, - }; - char full_dir_path[4096] = { - 0, - }; - ssize_t size = 0; - uuid_t tmp_gfid; - struct stat tmpbuf = { - 0, - }; - struct iatt iatt = { - 0, - }; - - struct stat lstatbuf = { - 0, - }; - struct iatt stbuf = { - 0, - }; - - conf = this->private; - if (!conf) { - ret = -1; - goto out; - } - - /* - * Since the primary target for the following lookup is to figure out if the - * entry still exists, going to do a direct stat call rather than going - * through the whole gluster stack. There are some benefits of doing gluster - * lookup, but this is redundant since we have done already one gluster - * lookup in the parent function. - * - * Randomly selecting the first local subvol to read, since it is expected - * that the directory structure is present in all the subvols identically - */ - - brick_len = strlen(defrag->local_brick_paths[0]); - /* discarding the first "/" */ - dir_len = strlen(loc->path) - 1; - /* Extra two: one for "/" at the end and one more for '\0'*/ - total_len = brick_len + dir_len + 2; - - snprintf(full_dir_path, total_len, "%s%s/", defrag->local_brick_paths[0], - loc->path + 1); - - ret = sys_lstat(full_dir_path, &tmpbuf); - if (ret == -1) { - gf_log(this->name, GF_LOG_ERROR, - "[absolutepath %s] directory " - "not found, path %s error %d", - full_dir_path, loc->path, errno); - goto out; - } - - dirp = sys_opendir(full_dir_path); - if (!dirp) { - ret = -1; - gf_msg(this->name, GF_LOG_ERROR, errno, 0, "failed to open dir : %s", - loc->path); - if (conf->decommission_subvols_cnt) { - defrag->total_failures++; - } - goto out; - } - - while ((entry = sys_readdir(dirp, scratch)) != NULL) { - if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) { - ret = 1; - goto out; - } - if (!strcmp(entry->d_name, ".") || !strcmp(entry->d_name, "..") || - !strcmp(entry->d_name, ".glusterfs")) - continue; - - /* TODO: Need to add a check for _DIRENT_HAVE_D_TYPE flag to fall back - to stat in case d_type is not defined */ - if (entry->d_type != DT_DIR) { - continue; - } - - entry_len = strlen(entry->d_name); - full_entry_length = total_len + entry_len + 1; /* one more for "/"*/ - - snprintf(full_entry_path, full_entry_length, "%s%s/", full_dir_path, - entry->d_name); - - size = sys_lgetxattr(full_entry_path, GFID_XATTR_KEY, tmp_gfid, 16); - if (size != 16) { - gf_log(this->name, GF_LOG_ERROR, "gfid not found, path %s", - full_entry_path); - continue; - } - - loc_wipe(&entry_loc); - - ret = dht_build_child_loc(this, &entry_loc, loc, entry->d_name); - if (ret) { - gf_log(this->name, GF_LOG_ERROR, - "Child loc" - " build failed for entry: %s", - entry->d_name); - - if (conf->decommission_in_progress) { - defrag->defrag_status = GF_DEFRAG_STATUS_FAILED; - - goto out; - } else { - should_commit_hash = 0; - - continue; - } - } - - if (gf_uuid_is_null(tmp_gfid)) { - gf_log(this->name, GF_LOG_ERROR, - "%s/%s" - " gfid not present", - loc->path, entry->d_name); - continue; - } - - gf_uuid_copy(entry_loc.gfid, tmp_gfid); - - /*In case the gfid stored in the inode by inode_link - *and the gfid obtained in the lookup differs, then - *client3_3_lookup_cbk will return ESTALE and proper - *error will be captured. - */ - memset(&lstatbuf, 0, sizeof(struct stat)); - ret = sys_lstat(full_entry_path, &lstatbuf); - if (ret == -1) { - gf_msg(this->name, GF_LOG_ERROR, errno, 0, "lstat failed for %s", - entry->d_name); - } - - memset(&stbuf, 0, sizeof(struct iatt)); - iatt_from_stat(&stbuf, &lstatbuf); - gf_uuid_copy(stbuf.ia_gfid, entry_loc.gfid); - linked_inode = inode_link(entry_loc.inode, loc->inode, entry->d_name, - &stbuf); - - inode = entry_loc.inode; - entry_loc.inode = linked_inode; - inode_unref(inode); - - if (gf_uuid_is_null(loc->gfid)) { - gf_log(this->name, GF_LOG_ERROR, - "%s/%s" - " gfid not present", - loc->path, entry->d_name); - continue; - } - - gf_uuid_copy(entry_loc.pargfid, loc->gfid); - - ret = syncop_lookup(this, &entry_loc, &iatt, NULL, NULL, NULL); - if (ret) { - if (-ret == ENOENT || -ret == ESTALE) { - gf_msg(this->name, GF_LOG_INFO, -ret, DHT_MSG_DIR_LOOKUP_FAILED, - "Dir:%s renamed or removed. " - "Skipping", - loc->path); - ret = 0; - if (conf->decommission_subvols_cnt) { - defrag->total_failures++; - } - continue; - } else { - gf_msg(this->name, GF_LOG_ERROR, -ret, - DHT_MSG_DIR_LOOKUP_FAILED, "lookup failed for:%s", - entry_loc.path); - - defrag->total_failures++; - - if (conf->decommission_in_progress) { - defrag->defrag_status = GF_DEFRAG_STATUS_FAILED; - ret = -1; - goto out; - } else { - should_commit_hash = 0; - continue; - } - } - } - - /* A return value of 2 means, either process_dir or - * lookup of a dir failed. Hence, don't commit hash - * for the current directory*/ - - ret = gf_defrag_fix_layout_puredist(this, defrag, &entry_loc, - fix_layout, migrate_data); - - if (defrag->defrag_status == GF_DEFRAG_STATUS_STOPPED) { - goto out; - } - - if (ret && ret != 2) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LAYOUT_FIX_FAILED, - "Fix layout failed for %s", entry_loc.path); - - defrag->total_failures++; - - if (conf->decommission_in_progress) { - defrag->defrag_status = GF_DEFRAG_STATUS_FAILED; - - goto out; - } else { - /* Let's not commit-hash if - * gf_defrag_fix_layout failed*/ - continue; - } - } - } - - ret = sys_closedir(dirp); - if (ret) { - gf_msg_debug(this->name, 0, - "Failed to close dir %s. Reason :" - " %s", - full_dir_path, strerror(errno)); - ret = 0; - } - - dirp = NULL; - - /* A directory layout is fixed only after its subdirs are healed to - * any newly added bricks. If the layout is fixed before subdirs are - * healed, the newly added brick will get a non-null layout. - * Any subdirs which hash to that layout will no longer show up - * in a directory listing until they are healed. - */ - - ret = syncop_setxattr(this, loc, fix_layout, 0, NULL, NULL); - - /* In case of a race where the directory is deleted just before - * layout setxattr, the errors are updated in the layout structure. - * We can use this information to make a decision whether the directory - * is deleted entirely. - */ - if (ret == 0) { - ret = dht_dir_layout_error_check(this, loc->inode); - ret = -ret; - } - - if (ret) { - if (-ret == ENOENT || -ret == ESTALE) { - gf_msg(this->name, GF_LOG_INFO, -ret, DHT_MSG_LAYOUT_FIX_FAILED, - "Setxattr failed. Dir %s " - "renamed or removed", - loc->path); - if (conf->decommission_subvols_cnt) { - defrag->total_failures++; - } - ret = 0; - goto out; - } else { - gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_LAYOUT_FIX_FAILED, - "Setxattr failed for %s", loc->path); - - defrag->total_failures++; - - if (conf->decommission_in_progress) { - defrag->defrag_status = GF_DEFRAG_STATUS_FAILED; - ret = -1; - goto out; - } - } - } - - if ((defrag->cmd != GF_DEFRAG_CMD_START_TIER) && - (defrag->cmd != GF_DEFRAG_CMD_START_LAYOUT_FIX)) { - ret = gf_defrag_process_dir(this, defrag, loc, migrate_data, &perrno); - - if (ret && (ret != 2)) { - if (perrno == ENOENT || perrno == ESTALE) { - ret = 0; - goto out; - } else { - defrag->total_failures++; - - gf_msg(this->name, GF_LOG_ERROR, 0, - DHT_MSG_DEFRAG_PROCESS_DIR_FAILED, - "gf_defrag_process_dir failed for " - "directory: %s", - loc->path); - - if (conf->decommission_in_progress) { - goto out; - } - - should_commit_hash = 0; - } - } else if (ret == 2) { - should_commit_hash = 0; - } - } - - gf_msg_trace(this->name, 0, "fix layout called on %s", loc->path); - - if (should_commit_hash && - gf_defrag_settle_hash(this, defrag, loc, fix_layout) != 0) { - defrag->total_failures++; - - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SETTLE_HASH_FAILED, - "Settle hash failed for %s", loc->path); - - ret = -1; - - if (conf->decommission_in_progress) { - defrag->defrag_status = GF_DEFRAG_STATUS_FAILED; - goto out; - } - } - - ret = 0; -out: - loc_wipe(&entry_loc); - - if (fd) - fd_unref(fd); - - if (ret == 0 && should_commit_hash == 0) { - ret = 2; - } - - if (dirp) { - sys_closedir(dirp); - } - return ret; } @@ -4419,31 +3901,26 @@ dht_init_local_subvols_and_nodeuuids(xlator_t *this, dht_conf_t *conf, loc_t *loc) { dict_t *dict = NULL; - gf_defrag_info_t *defrag = NULL; uuid_t *uuid_ptr = NULL; int ret = -1; int i = 0; int j = 0; - defrag = conf->defrag; - - if (defrag->cmd != GF_DEFRAG_CMD_START_TIER) { - /* Find local subvolumes */ - ret = syncop_getxattr(this, loc, &dict, GF_REBAL_FIND_LOCAL_SUBVOL, - NULL, NULL); - if (ret && (ret != -ENODATA)) { - gf_msg(this->name, GF_LOG_ERROR, -ret, 0, - "local " - "subvolume determination failed with error: %d", - -ret); - ret = -1; - goto out; - } - - if (!ret) - goto out; + /* Find local subvolumes */ + ret = syncop_getxattr(this, loc, &dict, GF_REBAL_FIND_LOCAL_SUBVOL, NULL, + NULL); + if (ret && (ret != -ENODATA)) { + gf_msg(this->name, GF_LOG_ERROR, -ret, 0, + "local " + "subvolume determination failed with error: %d", + -ret); + ret = -1; + goto out; } + if (!ret) + goto out; + ret = syncop_getxattr(this, loc, &dict, GF_REBAL_OLD_FIND_LOCAL_SUBVOL, NULL, NULL); if (ret) { @@ -4534,9 +4011,6 @@ dht_file_counter_thread(void *args) struct timespec time_to_wait = { 0, }; - struct timeval now = { - 0, - }; uint64_t tmp_size = 0; if (!args) @@ -4546,9 +4020,8 @@ dht_file_counter_thread(void *args) dht_build_root_loc(defrag->root_inode, &root_loc); while (defrag->defrag_status == GF_DEFRAG_STATUS_STARTED) { - gettimeofday(&now, NULL); - time_to_wait.tv_sec = now.tv_sec + 600; - time_to_wait.tv_nsec = 0; + timespec_now(&time_to_wait); + time_to_wait.tv_sec += 600; pthread_mutex_lock(&defrag->fc_mutex); pthread_cond_timedwait(&defrag->fc_wakeup_cond, &defrag->fc_mutex, @@ -4621,7 +4094,7 @@ gf_defrag_estimates_init(xlator_t *this, loc_t *loc, pthread_t *filecnt_thread) goto out; } - ret = gf_thread_create(filecnt_thread, NULL, &dht_file_counter_thread, + ret = gf_thread_create(filecnt_thread, NULL, dht_file_counter_thread, (void *)defrag, "dhtfcnt"); if (ret) { @@ -4678,7 +4151,7 @@ gf_defrag_parallel_migration_init(xlator_t *this, gf_defrag_info_t *defrag, /*Spawn Threads Here*/ while (index < thread_spawn_count) { - ret = gf_thread_create(&(tid[index]), NULL, &gf_defrag_task, + ret = gf_thread_create(&(tid[index]), NULL, gf_defrag_task, (void *)defrag, "dhtmig%d", (index + 1) & 0x3ff); if (ret != 0) { gf_msg("DHT", GF_LOG_ERROR, ret, 0, "Thread[%d] creation failed. ", @@ -4768,7 +4241,6 @@ gf_defrag_start_crawl(void *data) pthread_t *tid = NULL; pthread_t filecnt_thread; gf_boolean_t fc_thread_started = _gf_false; - int i = 0; this = data; if (!this) @@ -4786,7 +4258,8 @@ gf_defrag_start_crawl(void *data) if (!defrag) goto exit; - gettimeofday(&defrag->start_time, NULL); + defrag->start_time = gf_time(); + dht_build_root_inode(this, &defrag->root_inode); if (!defrag->root_inode) goto out; @@ -4903,12 +4376,6 @@ gf_defrag_start_crawl(void *data) goto out; } - ret = dht_get_brick_paths(this, conf, &loc); - if (ret) { - gf_log(this->name, GF_LOG_WARNING, "could not get brick path"); - ret = 0; - } - /* Initialise the structures required for parallel migration */ ret = gf_defrag_parallel_migration_init(this, defrag, &tid, &thread_index); @@ -4926,27 +4393,14 @@ gf_defrag_start_crawl(void *data) } } - /* TODO: Need to introduce a flag to safely operate in the old way */ - if (defrag->operate_dist && defrag->is_pure_distribute) { - ret = gf_defrag_fix_layout_puredist(this, defrag, &loc, fix_layout, - migrate_data); - if (ret && ret != 2) { - defrag->total_failures++; - ret = -1; - goto out; - } - } else { - ret = gf_defrag_fix_layout(this, defrag, &loc, fix_layout, - migrate_data); - if (ret && ret != 2) { - defrag->total_failures++; - ret = -1; - goto out; - } + ret = gf_defrag_fix_layout(this, defrag, &loc, fix_layout, migrate_data); + if (ret) { + defrag->total_failures++; + ret = -1; + goto out; } - if (ret != 2 && - gf_defrag_settle_hash(this, defrag, &loc, fix_layout) != 0) { + if (gf_defrag_settle_hash(this, defrag, &loc, fix_layout) != 0) { defrag->total_failures++; ret = -1; goto out; @@ -4988,14 +4442,6 @@ out: } UNLOCK(&defrag->lock); - for (i = 0; i < conf->local_subvols_cnt; i++) { - if (defrag->local_brick_paths[i]) { - GF_FREE(defrag->local_brick_paths[i]); - } - } - - GF_FREE(defrag->local_brick_paths); - GF_FREE(defrag); conf->defrag = NULL; @@ -5069,9 +4515,6 @@ gf_defrag_get_estimates_based_on_size(dht_conf_t *conf) uint64_t total_processed = 0; uint64_t tmp_count = 0; uint64_t time_to_complete = 0; - struct timeval now = { - 0, - }; double elapsed = 0; defrag = conf->defrag; @@ -5079,8 +4522,7 @@ gf_defrag_get_estimates_based_on_size(dht_conf_t *conf) if (!g_totalsize) goto out; - gettimeofday(&now, NULL); - elapsed = now.tv_sec - defrag->start_time.tv_sec; + elapsed = gf_time() - defrag->start_time; /* Don't calculate the estimates for the first 10 minutes. * It is unlikely to be accurate and estimates are not required @@ -5130,13 +4572,8 @@ gf_defrag_status_get(dht_conf_t *conf, dict_t *dict) uint64_t lookup = 0; uint64_t failures = 0; uint64_t skipped = 0; - uint64_t promoted = 0; - uint64_t demoted = 0; char *status = ""; double elapsed = 0; - struct timeval end = { - 0, - }; uint64_t time_to_complete = 0; uint64_t time_left = 0; gf_defrag_info_t *defrag = conf->defrag; @@ -5153,17 +4590,12 @@ gf_defrag_status_get(dht_conf_t *conf, dict_t *dict) lookup = defrag->num_files_lookedup; failures = defrag->total_failures; skipped = defrag->skipped; - promoted = defrag->total_files_promoted; - demoted = defrag->total_files_demoted; - gettimeofday(&end, NULL); - - elapsed = end.tv_sec - defrag->start_time.tv_sec; + elapsed = gf_time() - defrag->start_time; /* The rebalance is still in progress */ - if ((defrag->cmd != GF_DEFRAG_CMD_START_TIER) && - (defrag->defrag_status == GF_DEFRAG_STATUS_STARTED)) { + if (defrag->defrag_status == GF_DEFRAG_STATUS_STARTED) { time_to_complete = gf_defrag_get_estimates_based_on_size(conf); if (time_to_complete && (time_to_complete > elapsed)) @@ -5178,14 +4610,6 @@ gf_defrag_status_get(dht_conf_t *conf, dict_t *dict) if (!dict) goto log; - ret = dict_set_uint64(dict, "promoted", promoted); - if (ret) - gf_log(THIS->name, GF_LOG_WARNING, "failed to set promoted count"); - - ret = dict_set_uint64(dict, "demoted", demoted); - if (ret) - gf_log(THIS->name, GF_LOG_WARNING, "failed to set demoted count"); - ret = dict_set_uint64(dict, "files", files); if (ret) gf_log(THIS->name, GF_LOG_WARNING, "failed to set file count"); |