c/mirror_zfs
1
0
Fork 0
mirror of https://git.proxmox.com/git/mirror_zfs.git synced 2024-12-26 11:19:32 +03:00
Code Issues Packages Projects Releases Wiki Activity

Fix stalled txg with repeated noop scans

Browse Source 
Currently, the DSL scan code figures out when it should suspend
processing and allow a txg to continue by calling the function
dsl_scan_check_suspend(). Unfortunately, this function only
allows the scan to suspend at a level 0 block. In the event that
the system is scanning a bunch of empty snapshots or a resilver
is running with a high enough scn_cur_min_txg, the scan will
stop processing each dataset at the root level, deciding it
has nothing left to do. This means that the check_suspend
function is never called and the txg remains stuck until a
dataset is found that has data to scan.

This patch fixes the problem by allowing scans to suspend at
the root level of the objset. For backwards compatibility, we
use the bookmark <objsetid, 0, 0, 0> when we suspend here so
that older versions of the code will work as intended.

Reviewed-by: Matt Ahrens <matt@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes #9300
This commit is contained in:
Tom Caputi 2019-09-11 14:16:48 -04:00 committed by Brian Behlendorf
parent 5b51c15861
commit 5815f7ac30
1 changed files with 12 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
module/zfs/dsl_scan.c
View File

@ -1272,8 +1272,8 @@ dsl_scan_check_suspend(dsl_scan_t *scn, const zbookmark_phys_t *zb)
if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark)) if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark))
return (B_FALSE); /* we're resuming */ return (B_FALSE); /* we're resuming */
/* We only know how to resume from level-0 blocks. */ /* We only know how to resume from level-0 and objset blocks. */
if (zb && zb->zb_level != 0) if (zb && (zb->zb_level != 0 && zb->zb_level != ZB_ROOT_LEVEL))
return (B_FALSE); return (B_FALSE);
/* /*
@ -1304,7 +1304,16 @@ dsl_scan_check_suspend(dsl_scan_t *scn, const zbookmark_phys_t *zb)
NSEC2SEC(sync_time_ns) >= zfs_txg_timeout)) || NSEC2SEC(sync_time_ns) >= zfs_txg_timeout)) ||
spa_shutting_down(scn->scn_dp->dp_spa) || spa_shutting_down(scn->scn_dp->dp_spa) ||
(zfs_scan_strict_mem_lim && dsl_scan_should_clear(scn))) { (zfs_scan_strict_mem_lim && dsl_scan_should_clear(scn))) {
if (zb) { if (zb && zb->zb_level == ZB_ROOT_LEVEL) {
dprintf("suspending at first available bookmark "
"%llx/%llx/%llx/%llx\n",
(longlong_t)zb->zb_objset,
(longlong_t)zb->zb_object,
(longlong_t)zb->zb_level,
(longlong_t)zb->zb_blkid);
SET_BOOKMARK(&scn->scn_phys.scn_bookmark,
zb->zb_objset, 0, 0, 0);
} else if (zb != NULL) {
dprintf("suspending at bookmark %llx/%llx/%llx/%llx\n", dprintf("suspending at bookmark %llx/%llx/%llx/%llx\n",
(longlong_t)zb->zb_objset, (longlong_t)zb->zb_objset,
(longlong_t)zb->zb_object, (longlong_t)zb->zb_object,