aboutsummaryrefslogtreecommitdiffstats
path: root/uts/common/fs/zfs/dmu_object.c
diff options
context:
space:
mode:
authorAndriy Gapon <avg@FreeBSD.org>2017-04-14 17:13:49 +0000
committerAndriy Gapon <avg@FreeBSD.org>2017-04-14 17:13:49 +0000
commit40e6e931e08db158d51eeeb2c377e793a2785dbb (patch)
tree3a8f3cf248353b8a6a6d9b230da9db30257942bb /uts/common/fs/zfs/dmu_object.c
parent9bad09e9a8f414144e53a860942e691f80b48ed8 (diff)
downloadsrc-40e6e931e08db158d51eeeb2c377e793a2785dbb.tar.gz
src-40e6e931e08db158d51eeeb2c377e793a2785dbb.zip
7430 Backfill metadnode more intelligently
illumos/illumos-gate@af346df58864e8fe897b1ff1a3a4c12f9294391b https://github.com/illumos/illumos-gate/commit/af346df58864e8fe897b1ff1a3a4c12f9294391b https://www.illumos.org/issues/7430 Description and patch from brought over from the following ZoL commit: https:// github.com/zfsonlinux/zfs/commit/68cbd56e182ab949f58d004778d463aeb3f595c6 Only attempt to backfill lower metadnode object numbers if at least 4096 objects have been freed since the last rescan, and at most once per transaction group. This avoids a pathology in dmu_object_alloc() that caused O(N^2) behavior for create-heavy workloads and substantially improves object creation rates. As summarized by @mahrens in #4636: "Normally, the object allocator simply checks to see if the next object is available. The slow calls happened when dmu_object_alloc() checks to see if it can backfill lower object numbers. This happens every time we move on to a new L1 indirect block (i.e. every 32 * 128 = 4096 objects). When re-checking lower object numbers, we use the on-disk fill count (blkptr_t:blk_fill) to quickly skip over indirect blocks that don’t have enough free dnodes (defined as an L2 with at least 393,216 of 524,288 dnodes free). Therefore, we may find that a block of dnodes has a low (or zero) fill count, and yet we can’t allocate any of its dnodes, because they've been allocated in memory but not yet written to disk. In this case we have to hold each of the dnodes and then notice that it has been allocated in memory. The end result is that allocating N objects in the same TXG can require CPU usage proportional to N^2." Add a tunable dmu_rescan_dnode_threshold to define the number of objects that must be freed before a rescan is performed. Don't bother to export this as a module option because testing doesn't show a compelling reason to change it. The vast majority of the performance gain comes from limit the rescan to at most once per TXG. Reviewed by: Alek Pinchuk <alek@nexenta.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Approved by: Gordon Ross <gordon.w.ross@gmail.com> Author: Ned Bass <bass6@llnl.gov>
Notes
Notes: svn path=/vendor-sys/illumos/dist/; revision=316868
Diffstat (limited to 'uts/common/fs/zfs/dmu_object.c')
-rw-r--r--uts/common/fs/zfs/dmu_object.c31
1 files changed, 20 insertions, 11 deletions
diff --git a/uts/common/fs/zfs/dmu_object.c b/uts/common/fs/zfs/dmu_object.c
index 2c9802f51eff..3e4171ce6cc6 100644
--- a/uts/common/fs/zfs/dmu_object.c
+++ b/uts/common/fs/zfs/dmu_object.c
@@ -36,20 +36,22 @@ dmu_object_alloc(objset_t *os, dmu_object_type_t ot, int blocksize,
dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
{
uint64_t object;
- uint64_t L2_dnode_count = DNODES_PER_BLOCK <<
+ uint64_t L1_dnode_count = DNODES_PER_BLOCK <<
(DMU_META_DNODE(os)->dn_indblkshift - SPA_BLKPTRSHIFT);
dnode_t *dn = NULL;
- int restarted = B_FALSE;
mutex_enter(&os->os_obj_lock);
for (;;) {
object = os->os_obj_next;
/*
- * Each time we polish off an L2 bp worth of dnodes
- * (2^13 objects), move to another L2 bp that's still
- * reasonably sparse (at most 1/4 full). Look from the
- * beginning once, but after that keep looking from here.
- * If we can't find one, just keep going from here.
+ * Each time we polish off a L1 bp worth of dnodes (2^12
+ * objects), move to another L1 bp that's still reasonably
+ * sparse (at most 1/4 full). Look from the beginning at most
+ * once per txg, but after that keep looking from here.
+ * os_scan_dnodes is set during txg sync if enough objects
+ * have been freed since the previous rescan to justify
+ * backfilling again. If we can't find a suitable block, just
+ * keep going from here.
*
* Note that dmu_traverse depends on the behavior that we use
* multiple blocks of the dnode object before going back to
@@ -57,12 +59,19 @@ dmu_object_alloc(objset_t *os, dmu_object_type_t ot, int blocksize,
* that property or find another solution to the issues
* described in traverse_visitbp.
*/
- if (P2PHASE(object, L2_dnode_count) == 0) {
- uint64_t offset = restarted ? object << DNODE_SHIFT : 0;
- int error = dnode_next_offset(DMU_META_DNODE(os),
+
+ if (P2PHASE(object, L1_dnode_count) == 0) {
+ uint64_t offset;
+ int error;
+ if (os->os_rescan_dnodes) {
+ offset = 0;
+ os->os_rescan_dnodes = B_FALSE;
+ } else {
+ offset = object << DNODE_SHIFT;
+ }
+ error = dnode_next_offset(DMU_META_DNODE(os),
DNODE_FIND_HOLE,
&offset, 2, DNODES_PER_BLOCK >> 2, 0);
- restarted = B_TRUE;
if (error == 0)
object = offset >> DNODE_SHIFT;
}