c/mirror_zfs
1
0
Fork 0
mirror of https://git.proxmox.com/git/mirror_zfs.git synced 2026-05-22 02:27:36 +03:00
Code Issues Packages Projects Releases Wiki Activity

More adaptive ARC eviction

Browse Source 
Traditionally ARC adaptation was limited to MRU/MFU distribution.  But
for years people with metadata-centric workload demanded mechanisms to
also manage data/metadata distribution, that in original ZFS was just
a FIFO.  As result ZFS effectively got separate states for data and
metadata, minimum and maximum metadata limits etc, but it all required
manual tuning, was not adaptive and in its heart remained a bad FIFO.

This change removes most of existing eviction logic, rewriting it from
scratch.  This makes MRU/MFU adaptation individual for data and meta-
data, same as the distribution between data and metadata themselves.
Since most of required states separation was already done, it only
required to make arcs_size state field specific per data/metadata.

The adaptation logic is still based on previous concept of ghost hits,
just now it balances ARC capacity between 4 states: MRU data, MRU
metadata, MFU data and MFU metadata.  To simplify arc_c changes instead
of arc_p measured in bytes, this code uses 3 variable arc_meta, arc_pd
and arc_pm, representing ARC balance between metadata and data, MRU and
MFU for data, and MRU and MFU for metadata respectively as 32-bit fixed
point fractions.  Since we care about the math result only when need to
evict, this moves all the logic from arc_adapt() to arc_evict(), that
reduces per-block overhead, since per-block operations are limited to
stats collection, now moved from arc_adapt() to arc_access() and using
cheaper wmsums.  This also allows to remove ugly ARC_HDR_DO_ADAPT flag
from many places.

This change also removes number of metadata specific tunables, part of
which were actually not functioning correctly, since not all metadata
are equal and some (like L2ARC headers) are not really evictable.
Instead it introduced single opaque knob zfs_arc_meta_balance, tuning
ARC's reaction on ghost hits, allowing administrator give more or less
preference to metadata without setting strict limits.

Some of old code parts like arc_evict_meta() are just removed, because
since introduction of ABD ARC they really make no sense: only headers
referenced by small number of buffers are not evictable, and they are
really not evictable no matter what this code do.  Instead just call
arc_prune_async() if too much metadata appear not evictable.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Allan Jude <allan@klarasystems.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes #14359
This commit is contained in:
Alexander Motin
2023-03-08 14:17:23 -05:00
committed by GitHub
parent 8d9752569b
commit a8d83e2a24
10 changed files with 474 additions and 781 deletions
Download Patch File Download Diff File Expand all files Collapse all files
module/os/freebsd/zfs/arc_os.c
+1 -1
View File
@@ -159,7 +159,7 @@ arc_prune_task(void *arg)
/*
* Notify registered consumers they must drop holds on a portion of the ARC
* buffered they reference. This provides a mechanism to ensure the ARC can
* honor the arc_meta_limit and reclaim otherwise pinned ARC buffers. This
* honor the metadata limit and reclaim otherwise pinned ARC buffers. This
* is analogous to dnlc_reduce_cache() but more generic.
*
* This operation is performed asynchronously so it may be safely called
module/os/freebsd/zfs/sysctl_os.c
+51 -26
View File
@@ -359,89 +359,114 @@ SYSCTL_INT(_vfs_zfs, OID_AUTO, l2arc_norw,
"No reads during writes (LEGACY)");
/* END CSTYLED */
static int
param_get_arc_state_size(SYSCTL_HANDLER_ARGS)
{
arc_state_t *state = (arc_state_t *)arg1;
int64_t val;
val = zfs_refcount_count(&state->arcs_size[ARC_BUFC_DATA]) +
zfs_refcount_count(&state->arcs_size[ARC_BUFC_METADATA]);
return (sysctl_handle_64(oidp, &val, 0, req));
}
extern arc_state_t ARC_anon;
/* BEGIN CSTYLED */
SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, anon_size, CTLFLAG_RD,
&ARC_anon.arcs_size.rc_count, 0, "size of anonymous state");
SYSCTL_PROC(_vfs_zfs, OID_AUTO, anon_size,
CTLTYPE_S64 | CTLFLAG_RD | CTLFLAG_MPSAFE,
&ARC_anon, 0, param_get_arc_state_size, "Q",
"size of anonymous state");
SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, anon_metadata_esize, CTLFLAG_RD,
&ARC_anon.arcs_esize[ARC_BUFC_METADATA].rc_count, 0,
"size of metadata in anonymous state");
"size of evictable metadata in anonymous state");
SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, anon_data_esize, CTLFLAG_RD,
&ARC_anon.arcs_esize[ARC_BUFC_DATA].rc_count, 0,
"size of data in anonymous state");
"size of evictable data in anonymous state");
/* END CSTYLED */
extern arc_state_t ARC_mru;
/* BEGIN CSTYLED */
SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_size, CTLFLAG_RD,
&ARC_mru.arcs_size.rc_count, 0, "size of mru state");
SYSCTL_PROC(_vfs_zfs, OID_AUTO, mru_size,
CTLTYPE_S64 | CTLFLAG_RD | CTLFLAG_MPSAFE,
&ARC_mru, 0, param_get_arc_state_size, "Q",
"size of mru state");
SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_metadata_esize, CTLFLAG_RD,
&ARC_mru.arcs_esize[ARC_BUFC_METADATA].rc_count, 0,
"size of metadata in mru state");
"size of evictable metadata in mru state");
SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_data_esize, CTLFLAG_RD,
&ARC_mru.arcs_esize[ARC_BUFC_DATA].rc_count, 0,
"size of data in mru state");
"size of evictable data in mru state");
/* END CSTYLED */
extern arc_state_t ARC_mru_ghost;
/* BEGIN CSTYLED */
SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_ghost_size, CTLFLAG_RD,
&ARC_mru_ghost.arcs_size.rc_count, 0, "size of mru ghost state");
SYSCTL_PROC(_vfs_zfs, OID_AUTO, mru_ghost_size,
CTLTYPE_S64 | CTLFLAG_RD | CTLFLAG_MPSAFE,
&ARC_mru_ghost, 0, param_get_arc_state_size, "Q",
"size of mru ghost state");
SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_ghost_metadata_esize, CTLFLAG_RD,
&ARC_mru_ghost.arcs_esize[ARC_BUFC_METADATA].rc_count, 0,
"size of metadata in mru ghost state");
"size of evictable metadata in mru ghost state");
SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_ghost_data_esize, CTLFLAG_RD,
&ARC_mru_ghost.arcs_esize[ARC_BUFC_DATA].rc_count, 0,
"size of data in mru ghost state");
"size of evictable data in mru ghost state");
/* END CSTYLED */
extern arc_state_t ARC_mfu;
/* BEGIN CSTYLED */
SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_size, CTLFLAG_RD,
&ARC_mfu.arcs_size.rc_count, 0, "size of mfu state");
SYSCTL_PROC(_vfs_zfs, OID_AUTO, mfu_size,
CTLTYPE_S64 | CTLFLAG_RD | CTLFLAG_MPSAFE,
&ARC_mfu, 0, param_get_arc_state_size, "Q",
"size of mfu state");
SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_metadata_esize, CTLFLAG_RD,
&ARC_mfu.arcs_esize[ARC_BUFC_METADATA].rc_count, 0,
"size of metadata in mfu state");
"size of evictable metadata in mfu state");
SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_data_esize, CTLFLAG_RD,
&ARC_mfu.arcs_esize[ARC_BUFC_DATA].rc_count, 0,
"size of data in mfu state");
"size of evictable data in mfu state");
/* END CSTYLED */
extern arc_state_t ARC_mfu_ghost;
/* BEGIN CSTYLED */
SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_ghost_size, CTLFLAG_RD,
&ARC_mfu_ghost.arcs_size.rc_count, 0, "size of mfu ghost state");
SYSCTL_PROC(_vfs_zfs, OID_AUTO, mfu_ghost_size,
CTLTYPE_S64 | CTLFLAG_RD | CTLFLAG_MPSAFE,
&ARC_mfu_ghost, 0, param_get_arc_state_size, "Q",
"size of mfu ghost state");
SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_ghost_metadata_esize, CTLFLAG_RD,
&ARC_mfu_ghost.arcs_esize[ARC_BUFC_METADATA].rc_count, 0,
"size of metadata in mfu ghost state");
"size of evictable metadata in mfu ghost state");
SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_ghost_data_esize, CTLFLAG_RD,
&ARC_mfu_ghost.arcs_esize[ARC_BUFC_DATA].rc_count, 0,
"size of data in mfu ghost state");
"size of evictable data in mfu ghost state");
/* END CSTYLED */
extern arc_state_t ARC_uncached;
/* BEGIN CSTYLED */
SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, uncached_size, CTLFLAG_RD,
&ARC_uncached.arcs_size.rc_count, 0, "size of uncached state");
SYSCTL_PROC(_vfs_zfs, OID_AUTO, uncached_size,
CTLTYPE_S64 | CTLFLAG_RD | CTLFLAG_MPSAFE,
&ARC_uncached, 0, param_get_arc_state_size, "Q",
"size of uncached state");
SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, uncached_metadata_esize, CTLFLAG_RD,
&ARC_uncached.arcs_esize[ARC_BUFC_METADATA].rc_count, 0,
"size of metadata in uncached state");
"size of evictable metadata in uncached state");
SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, uncached_data_esize, CTLFLAG_RD,
&ARC_uncached.arcs_esize[ARC_BUFC_DATA].rc_count, 0,
"size of data in uncached state");
"size of evictable data in uncached state");
/* END CSTYLED */
extern arc_state_t ARC_l2c_only;
/* BEGIN CSTYLED */
SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, l2c_only_size, CTLFLAG_RD,
&ARC_l2c_only.arcs_size.rc_count, 0, "size of mru state");
SYSCTL_PROC(_vfs_zfs, OID_AUTO, l2c_only_size,
CTLTYPE_S64 | CTLFLAG_RD | CTLFLAG_MPSAFE,
&ARC_l2c_only, 0, param_get_arc_state_size, "Q",
"size of l2c_only state");
/* END CSTYLED */
/* dbuf.c */
module/os/linux/zfs/arc_os.c
+1 -1
View File
@@ -504,7 +504,7 @@ arc_prune_task(void *ptr)
/*
* Notify registered consumers they must drop holds on a portion of the ARC
* buffered they reference. This provides a mechanism to ensure the ARC can
* honor the arc_meta_limit and reclaim otherwise pinned ARC buffers. This
* honor the metadata limit and reclaim otherwise pinned ARC buffers. This
* is analogous to dnlc_reduce_cache() but more generic.
*
* This operation is performed asynchronously so it may be safely called