summary |
shortlog |
log |
commit | commitdiff |
tree
raw |
patch |
inline | side by side (from parent 1:
7e26413)
When MDTs are not balanced, QOS code tries to keep subdirectory
creation local to the same MDT when it is deep in the directory
tree, to avoid creating too many remote directories, but the
existing weight to stay on the parent MDT until 50% of other MDTs
is too radical, and causes mkdirs to be "stuck" on the same MDT.
* remove "lq_threshold_rr" from above calculation because the check
in ltd_qos_is_usable() handles this, so use only "dir_depth".
* the factor is changed to "16 / (dir_depth + 10)", then it's less
likely to stick to the parent MDT for top levels, while more
likely to stay on the parent MDT for low levels:
depth=0 -> 160%, depth=4 -> 114%, depth=6 -> 100%,
depth=8 -> 88%, depth=12 -> 72%
* rename lli_depth to lli_dir_depth to make usage more clear.
Signed-off-by: Lai Siyao <lai.siyao@whamcloud.com>
Change-Id: Iec6b77919b630d4baee6d54bee7bdb8ca9fb8574
Reviewed-on: https://review.whamcloud.com/45544
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Hongchao Zhang <hongchao@whamcloud.com>
if (IS_ERR(op_data))
RETURN(PTR_ERR(op_data));
if (IS_ERR(op_data))
RETURN(PTR_ERR(op_data));
- op_data->op_dir_depth = ll_i2info(parent)->lli_depth;
+ op_data->op_dir_depth = ll_i2info(parent)->lli_dir_depth;
if (ll_sbi_has_encrypt(sbi) &&
(IS_ENCRYPTED(parent) ||
if (ll_sbi_has_encrypt(sbi) &&
(IS_ENCRYPTED(parent) ||
* -- I am the owner of dir statahead. */
pid_t lli_opendir_pid;
/* directory depth to ROOT */
* -- I am the owner of dir statahead. */
pid_t lli_opendir_pid;
/* directory depth to ROOT */
- unsigned short lli_depth;
+ unsigned short lli_dir_depth;
/* stat will try to access statahead entries or start
* statahead if this flag is set, and this flag will be
* set upon dir open, and cleared when dir is closed,
/* stat will try to access statahead entries or start
* statahead if this flag is set, and this flag will be
* set upon dir open, and cleared when dir is closed,
return;
lli = ll_i2info(inode);
return;
lli = ll_i2info(inode);
- lli->lli_depth = ll_i2info(dir)->lli_depth + 1;
- CDEBUG(D_INODE, DFID" depth %hu\n", PFID(&lli->lli_fid), lli->lli_depth);
+ lli->lli_dir_depth = ll_i2info(dir)->lli_dir_depth + 1;
+ CDEBUG(D_INODE, DFID" depth %hu\n",
+ PFID(&lli->lli_fid), lli->lli_dir_depth);
}
void ll_truncate_inode_pages_final(struct inode *inode)
}
void ll_truncate_inode_pages_final(struct inode *inode)
struct ll_inode_info *lli = ll_i2info(dir);
struct lmv_stripe_md *lsm;
struct ll_inode_info *lli = ll_i2info(dir);
struct lmv_stripe_md *lsm;
- op_data->op_dir_depth = lli->lli_depth;
+ op_data->op_dir_depth = lli->lli_dir_depth;
/* parent directory is striped */
if (unlikely(lli->lli_lsm_md))
/* parent directory is striped */
if (unlikely(lli->lli_lsm_md))
if (lsm->lsm_md_max_inherit != LMV_INHERIT_NONE &&
(lsm->lsm_md_max_inherit == LMV_INHERIT_UNLIMITED ||
if (lsm->lsm_md_max_inherit != LMV_INHERIT_NONE &&
(lsm->lsm_md_max_inherit == LMV_INHERIT_UNLIMITED ||
- lsm->lsm_md_max_inherit >= lli->lli_depth)) {
+ lsm->lsm_md_max_inherit >= lli->lli_dir_depth)) {
op_data->op_flags |= MF_QOS_MKDIR;
if (lsm->lsm_md_max_inherit_rr != LMV_INHERIT_RR_NONE &&
(lsm->lsm_md_max_inherit_rr == LMV_INHERIT_RR_UNLIMITED ||
op_data->op_flags |= MF_QOS_MKDIR;
if (lsm->lsm_md_max_inherit_rr != LMV_INHERIT_RR_NONE &&
(lsm->lsm_md_max_inherit_rr == LMV_INHERIT_RR_UNLIMITED ||
- lsm->lsm_md_max_inherit_rr >= lli->lli_depth))
+ lsm->lsm_md_max_inherit_rr >= lli->lli_dir_depth))
op_data->op_flags |= MF_RR_MKDIR;
CDEBUG(D_INODE, DFID" requests qos mkdir %#x\n",
PFID(&lli->lli_fid), op_data->op_flags);
op_data->op_flags |= MF_RR_MKDIR;
CDEBUG(D_INODE, DFID" requests qos mkdir %#x\n",
PFID(&lli->lli_fid), op_data->op_flags);
/* if current MDT has above-average space, within range of the QOS
* threshold, stay on the same MDT to avoid creating needless remote
/* if current MDT has above-average space, within range of the QOS
* threshold, stay on the same MDT to avoid creating needless remote
- * MDT directories. It's more likely for low level directories.
+ * MDT directories. It's more likely for low level directories
+ * "16 / (dir_depth + 10)" is the factor to make it more unlikely for
+ * top level directories, while more likely for low levels.
- rand = total_avail * (256 - lmv->lmv_qos.lq_threshold_rr) /
- (total_usable * 256 * (1 + dir_depth / 4));
+ rand = total_avail * 16 / (total_usable * (dir_depth + 10));
if (cur && cur->ltd_qos.ltq_avail >= rand) {
tgt = cur;
GOTO(unlock, tgt);
if (cur && cur->ltd_qos.ltq_avail >= rand) {
tgt = cur;
GOTO(unlock, tgt);