diff options
author | shishir gowda <shishirng@gluster.com> | 2012-02-17 19:21:07 +0530 |
---|---|---|
committer | Vijay Bellur <vijay@gluster.com> | 2012-02-19 04:47:50 -0800 |
commit | 4a65356275cda45f665dfa99d3712f6fb9087aff (patch) | |
tree | 43a58dd347e770073d26c13fd54522e3825d957b /xlators/cluster | |
parent | 6123265556e54891369cc79a72b670a4b23a8a41 (diff) |
cluster/dht: Support for hardlink rebalance when decommissioning
The support for hardlink rebalance is only available for decommissioning
of a node. this can be triggered in two ways
1. remove-brick start
2. if decommission node value is set in vol file, then a normal rebalance
command
The way we handle it is-
if (nlink > 1)
do
* if src file doesnt have linkto xattr
* mark src's linkto to the dst
* else
* perform a link on the dst
* do a look up
* if nlinks = dst.nlinks
* migrate data
* else
* continue crawling
done
Signed-off-by: shishir gowda <shishirng@gluster.com>
Change-Id: If43b5524b872fd1413e9f7aa7f436cb244e30d8d
BUG: 763844
Reviewed-on: http://review.gluster.com/2737
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Amar Tumballi <amarts@redhat.com>
Diffstat (limited to 'xlators/cluster')
-rw-r--r-- | xlators/cluster/dht/src/dht-common.c | 8 | ||||
-rw-r--r-- | xlators/cluster/dht/src/dht-common.h | 13 | ||||
-rw-r--r-- | xlators/cluster/dht/src/dht-rebalance.c | 155 | ||||
-rw-r--r-- | xlators/cluster/dht/src/dht.c | 1 |
4 files changed, 163 insertions, 14 deletions
diff --git a/xlators/cluster/dht/src/dht-common.c b/xlators/cluster/dht/src/dht-common.c index f76dba40f..9f364d10c 100644 --- a/xlators/cluster/dht/src/dht-common.c +++ b/xlators/cluster/dht/src/dht-common.c @@ -2145,7 +2145,7 @@ dht_setxattr (call_frame_t *frame, xlator_t *this, data_t *tmp = NULL; uint32_t dir_spread = 0; char value[4096] = {0,}; - int forced_rebalance = 0; + gf_dht_migrate_data_type_t forced_rebalance = GF_DHT_MIGRATE_DATA; int call_cnt = 0; data_pair_t *trav = NULL; @@ -2214,7 +2214,11 @@ dht_setxattr (call_frame_t *frame, xlator_t *this, (ie, 'target' subvolume given there, etc) */ memcpy (value, tmp->data, tmp->len); if (strcmp (value, "force") == 0) - forced_rebalance = 1; + forced_rebalance = + GF_DHT_MIGRATE_DATA_EVEN_IF_LINK_EXISTS; + + if (conf->decommission_in_progress) + forced_rebalance = GF_DHT_MIGRATE_HARDLINK; local->rebalance.target_node = dht_subvol_get_hashed (this, loc); if (!local->rebalance.target_node) { diff --git a/xlators/cluster/dht/src/dht-common.h b/xlators/cluster/dht/src/dht-common.h index e44c947dd..2ccdea747 100644 --- a/xlators/cluster/dht/src/dht-common.h +++ b/xlators/cluster/dht/src/dht-common.h @@ -245,6 +245,7 @@ struct dht_conf { /* to keep track of nodes which are decomissioned */ xlator_t **decommissioned_bricks; + int decommission_in_progress; /* defrag related */ gf_defrag_info_t *defrag; @@ -262,6 +263,12 @@ struct dht_disk_layout { }; typedef struct dht_disk_layout dht_disk_layout_t; +typedef enum { + GF_DHT_MIGRATE_DATA, + GF_DHT_MIGRATE_DATA_EVEN_IF_LINK_EXISTS, + GF_DHT_MIGRATE_HARDLINK, + GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS +} gf_dht_migrate_data_type_t; #define ENTRY_MISSING(op_ret, op_errno) (op_ret == -1 && op_errno == ENOENT) @@ -655,4 +662,10 @@ gf_defrag_stop (gf_defrag_info_t *defrag, dict_t *output); void* gf_defrag_start (void *this); +int32_t +gf_defrag_handle_hardlink (xlator_t *this, loc_t *loc, dict_t *xattrs, + struct iatt *stbuf); +int +dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to, + int flag); #endif/* _DHT_H */ diff --git a/xlators/cluster/dht/src/dht-rebalance.c b/xlators/cluster/dht/src/dht-rebalance.c index 46fc8773e..ecf664bb7 100644 --- a/xlators/cluster/dht/src/dht-rebalance.c +++ b/xlators/cluster/dht/src/dht-rebalance.c @@ -28,7 +28,6 @@ #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, @@ -99,8 +98,114 @@ out: } +int32_t +gf_defrag_handle_hardlink (xlator_t *this, loc_t *loc, dict_t *xattrs, + struct iatt *stbuf) +{ + int32_t ret = -1; + xlator_t *cached_subvol = NULL; + xlator_t *hashed_subvol = NULL; + xlator_t *linkto_subvol = NULL; + data_t *data = NULL; + struct iatt iatt = {0,}; + int32_t op_errno = 0; + + GF_VALIDATE_OR_GOTO ("defrag", loc, out); + GF_VALIDATE_OR_GOTO ("defrag", loc->name, out); + GF_VALIDATE_OR_GOTO ("defrag", stbuf, out); + GF_VALIDATE_OR_GOTO ("defrag", this, out); + GF_VALIDATE_OR_GOTO ("defrag", xattrs, out); + + if (uuid_is_null (loc->pargfid)) { + gf_log ("", GF_LOG_ERROR, "loc->pargfid is NULL for " + "%s", loc->path); + goto out; + } + + if (uuid_is_null (loc->gfid)) { + gf_log ("", GF_LOG_ERROR, "loc->gfid is NULL for " + "%s", loc->path); + goto out; + } + + cached_subvol = dht_subvol_get_cached (this, loc->inode); + if (!cached_subvol) { + gf_log (this->name, GF_LOG_ERROR, "Failed to get cached subvol" + " for %s on %s", loc->name, this->name); + goto out; + } + + hashed_subvol = dht_subvol_get_hashed (this, loc); + if (!hashed_subvol) { + gf_log (this->name, GF_LOG_ERROR, "Failed to get hashed subvol" + " for %s on %s", loc->name, this->name); + goto out; + } + + gf_log (this->name, GF_LOG_INFO, "Attempting to migrate hardlink %s " + "with gfid %s from %s -> %s", loc->name, uuid_utoa (loc->gfid), + cached_subvol->name, hashed_subvol->name); + data = dict_get (xattrs, DHT_LINKFILE_KEY); + /* set linkto on cached -> hashed if not present, else link it */ + if (!data) { + ret = dict_set_str (xattrs, DHT_LINKFILE_KEY, + hashed_subvol->name); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, "Failed to set " + "linkto xattr in dict for %s", loc->name); + goto out; + } + + ret = syncop_setxattr (cached_subvol, loc, xattrs, 0); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, "Linkto setxattr " + "failed %s -> %s (%s)", cached_subvol->name, + loc->name, strerror (errno)); + goto out; + } + goto out; + } else { + linkto_subvol = dht_linkfile_subvol (this, NULL, NULL, xattrs); + if (!linkto_subvol) { + gf_log (this->name, GF_LOG_ERROR, "Failed to get " + "linkto subvol for %s", loc->name); + } else { + hashed_subvol = linkto_subvol; + } + + ret = syncop_link (hashed_subvol, loc, loc); + if (ret) { + op_errno = errno; + gf_log (this->name, GF_LOG_ERROR, "link of %s -> %s" + " failed on subvol %s (%s)", loc->name, + uuid_utoa(loc->gfid), + hashed_subvol->name, strerror (op_errno)); + if (op_errno != EEXIST) + goto out; + } + } + ret = syncop_lookup (hashed_subvol, loc, NULL, &iatt, NULL, NULL); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, "Failed lookup %s on %s (%s)" + , loc->name, hashed_subvol->name, strerror (errno)); + goto out; + } + + if (iatt.ia_nlink == stbuf->ia_nlink) { + ret = dht_migrate_file (this, loc, cached_subvol, hashed_subvol, + GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS); + if (ret) + goto out; + } + ret = 0; +out: + return ret; +} + + static inline int -__is_file_migratable (xlator_t *this, loc_t *loc, struct iatt *stbuf) +__is_file_migratable (xlator_t *this, loc_t *loc, + struct iatt *stbuf, dict_t *xattrs, int flags) { int ret = -1; @@ -111,11 +216,25 @@ __is_file_migratable (xlator_t *this, loc_t *loc, struct iatt *stbuf) goto out; } + if (flags == GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS) { + ret = 0; + 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; + /* support for decomission */ + if (flags == GF_DHT_MIGRATE_HARDLINK) { + ret = gf_defrag_handle_hardlink (this, loc, + xattrs, stbuf); + if (ret) { + gf_log (this->name, GF_LOG_WARNING, + "%s: failed to migrate file with link", + loc->path); + } + } else { + gf_log (this->name, GF_LOG_WARNING, + "%s: file has hardlinks", loc->path); + } + ret = ENOTSUP; goto out; } @@ -504,6 +623,7 @@ dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to, fd_t *dst_fd = NULL; dict_t *dict = NULL; dict_t *xattr = NULL; + dict_t *xattr_rsp = NULL; int file_has_holes = 0; gf_log (this->name, GF_LOG_INFO, "%s: attempting to move from %s to %s", @@ -513,19 +633,29 @@ dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to, if (!dict) goto out; + ret = dict_set_int32 (dict, DHT_LINKFILE_KEY, 256); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, + "%s: failed to set 'linkto' key in dict", loc->path); + goto out; + } + /* Phase 1 - Data migration is in progress from now on */ - ret = syncop_lookup (from, loc, NULL, &stbuf, NULL, NULL); + ret = syncop_lookup (from, loc, dict, &stbuf, &xattr_rsp, NULL); if (ret) { gf_log (this->name, GF_LOG_ERROR, "%s: lookup failed on %s (%s)", loc->path, from->name, strerror (errno)); goto out; } + /* we no more require this key */ + dict_del (dict, DHT_LINKFILE_KEY); + /* preserve source mode, so set the same to the destination */ src_ia_prot = stbuf.ia_prot; /* Check if file can be migrated */ - ret = __is_file_migratable (this, loc, &stbuf); + ret = __is_file_migratable (this, loc, &stbuf, xattr_rsp, flag); if (ret) goto out; @@ -543,7 +673,7 @@ dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to, goto out; /* Should happen on all files when 'force' option is not given */ - if (flag != DHT_MIGRATE_EVEN_IF_LINK_EXISTS) { + if (flag == GF_DHT_MIGRATE_DATA) { ret = __dht_check_free_space (to, from, loc, &stbuf); if (ret) { goto out; @@ -714,6 +844,8 @@ out: if (xattr) dict_unref (xattr); + if (xattr_rsp) + dict_unref (xattr_rsp); if (dst_fd) syncop_close (dst_fd); @@ -909,6 +1041,8 @@ gf_defrag_migrate_data (xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, struct iatt iatt = {0,}; int32_t op_errno = 0; + gf_log (this->name, GF_LOG_INFO, "migate data called on %s", + loc->path); fd = fd_create (loc->inode, defrag->pid); if (!fd) { gf_log (this->name, GF_LOG_ERROR, "Failed to create fd"); @@ -950,8 +1084,6 @@ gf_defrag_migrate_data (xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, continue; defrag->num_files_lookedup++; - if (entry->d_stat.ia_nlink > 1) - continue; loc_wipe (&entry_loc); ret =dht_build_child_loc (this, &entry_loc, loc, @@ -1223,7 +1355,6 @@ gf_defrag_start_crawl (void *data) defrag = conf->defrag; if (!defrag) goto out; - dht_build_root_inode (this, &defrag->root_inode); if (!defrag->root_inode) goto out; diff --git a/xlators/cluster/dht/src/dht.c b/xlators/cluster/dht/src/dht.c index 816bf868e..bb6f8c099 100644 --- a/xlators/cluster/dht/src/dht.c +++ b/xlators/cluster/dht/src/dht.c @@ -284,6 +284,7 @@ dht_parse_decommissioned_bricks (xlator_t *this, dht_conf_t *conf, } ret = 0; + conf->decommission_in_progress = 1; out: if (dup_brick) GF_FREE (dup_brick); |