Previous flushing algorithm limited only total number of log blocks to
the minimum of 256K and 4x number of metaslabs in the pool. As result,
system with 1500 disks with 1000 metaslabs each, touching several new
metaslabs each TXG could grow spacemap log to huge size without much
benefits. We've observed one of such systems importing pool for about
45 minutes.
This patch improves the situation from five sides:
- By limiting maximum period for each metaslab to be flushed to 1000
TXGs, that effectively limits maximum number of per-TXG spacemap logs
to load to the same number.
- By making flushing more smooth via accounting number of metaslabs
that were touched after the last flush and actually need another flush,
not just ms_unflushed_txg bump.
- By applying zfs_unflushed_log_block_pct to the number of metaslabs
that were touched after the last flush, not all metaslabs in the pool.
- By aggressively prefetching per-TXG spacemap logs up to 16 TXGs in
advance, making log spacemap load process for wide HDD pool CPU-bound,
accelerating it by many times.
- By reducing zfs_unflushed_log_block_max from 256K to 128K, reducing
single-threaded by nature log processing time from ~10 to ~5 minutes.
As further optimization we could skip bumping ms_unflushed_txg for
metaslabs not touched since the last flush, but that would be an
incompatible change, requiring new pool feature.
Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored-By: iXsystems, Inc.
Closes#12789
Currently, determining which datasets are affected by corruption is
a manual process.
The primary difficulty in reporting the list of affected snapshots is
that since the error was initially found, the snapshot where the error
originally occurred in, may have been deleted. To solve this issue, we
add the ID of the head dataset of the original snapshot which the error
was detected in, to the stored error report. Then any time a filesystem
is deleted, the errors associated with it are deleted as well. Any time
a clone promote occurs, we modify reports associated with the original
head to refer to the new head. The stored error reports are identified
by this head ID, the birth time of the block which the error occurred
in, as well as some information about the error itself are also stored.
Once this information is stored, we can find the set of datasets
affected by an error by walking back the list of snapshots in the given
head until we find one with the appropriate birth txg, and then traverse
through the snapshots of the clone family, terminating a branch if the
block was replaced in a given snapshot. Then we report this information
back to libzfs, and to the zpool status command, where it is displayed
as follows:
pool: test
state: ONLINE
status: One or more devices has experienced an error resulting in data
corruption. Applications may be affected.
action: Restore the file in question if possible. Otherwise restore the
entire pool from backup.
see: https://openzfs.github.io/openzfs-docs/msg/ZFS-8000-8A
scan: scrub repaired 0B in 00:00:00 with 800 errors on Fri Dec 3
08:27:57 2021
config:
NAME STATE READ WRITE CKSUM
test ONLINE 0 0 0
sdb ONLINE 0 0 1.58K
errors: Permanent errors have been detected in the following files:
test@1:/test.0.0
/test/test.0.0
/test/1clone/test.0.0
A new feature flag is introduced to mark the presence of this change, as
well as promotion and backwards compatibility logic. This is an updated
version of #9175. Rebase required fixing the tests, updating the ABI of
libzfs, updating the man pages, fixing bugs, fixing the error returns,
and updating the old on-disk error logs to the new format when
activating the feature.
Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Mark Maybee <mark.maybee@delphix.com>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Co-authored-by: TulsiJain <tulsi.jain@delphix.com>
Signed-off-by: George Amanakis <gamanakis@gmail.com>
Closes#9175Closes#12812
As such, there are no specific synchronous semantics defined for
the xattrs. But for xattr=on, it does log to ZIL and zil_commit() is
done, if sync=always is set on dataset. This provides sync semantics
for xattr=on with sync=always set on dataset.
For the xattr=sa implementation, it doesn't log to ZIL, so, even with
sync=always, xattrs are not guaranteed to be synced before xattr call
returns to caller. So, xattr can be lost if system crash happens, before
txg carrying xattr transaction is synced.
This change adds xattr=sa logging to ZIL on xattr create/remove/update
and xattrs are synced to ZIL (zil_commit() done) for sync=always.
This makes xattr=sa behavior similar to xattr=on.
Implementation notes:
The actual logging is fairly straight-forward and does not warrant
additional explanation.
However, it has been 14 years since we last added new TX types
to the ZIL [1], hence this is the first time we do it after the
introduction of zpool features. Therefore, here is an overview of the
feature activation and deactivation workflow:
1. The feature must be enabled. Otherwise, we don't log the new
record type. This ensures compatibility with older software.
2. The feature is activated per-dataset, since the ZIL is per-dataset.
3. If the feature is enabled and dataset is not for zvol, any append to
the ZIL chain will activate the feature for the dataset. Likewise
for starting a new ZIL chain.
4. A dataset that doesn't have a ZIL chain has the feature deactivated.
We ensure (3) by activating on the first zil_commit() after the feature
was enabled. Since activating the features requires waiting for txg
sync, the first zil_commit() after enabling the feature will be slower
than usual. The downside is that this is really a conservative
approximation: even if we never append a 'TX_SETSAXATTR' to the ZIL
chain, we pay the penalty for feature activation. The upside is that the
user is in control of when we pay the penalty, i.e., upon enabling the
feature.
We ensure (4) by hooking into zil_sync(), where ZIL destroy actually
happens.
One more piece on feature activation, since it's spread across
multiple functions:
zil_commit()
zil_process_commit_list()
if lwb == NULL // first zil_commit since zil_open
zil_create()
if no log block pointer in ZIL header:
if feature enabled and not active:
// CASE 1
enable, COALESCE txg wait with dmu_tx that allocated the
log block
else // log block was allocated earlier than this zil_open
if feature enabled and not active:
// CASE 2
enable, EXPLICIT txg wait
else // already have an in-DRAM LWB
if feature enabled and not active:
// this happens when we enable the feature after zil_create
// CASE 3
enable, EXPLICIT txg wait
[1] da6c28aaf6
Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Christian Schwarz <christian.schwarz@nutanix.com>
Reviewed-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Reviewed-by: Ryan Moeller <freqlabs@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Jitendra Patidar <jitendra.patidar@nutanix.com>
Closes#8768Closes#9078
ZFS on Linux originally implemented xattr namespaces in a way that is
incompatible with other operating systems. On illumos, xattrs do not
have namespaces. Every xattr name is visible. FreeBSD has two
universally defined namespaces: EXTATTR_NAMESPACE_USER and
EXTATTR_NAMESPACE_SYSTEM. The system namespace is used for protected
FreeBSD-specific attributes such as MAC labels and pnfs state. These
attributes have the namespace string "freebsd:system:" prefixed to the
name in the encoding scheme used by ZFS. The user namespace is used
for general purpose user attributes and obeys normal access control
mechanisms. These attributes have no namespace string prefixed, so
xattrs written on illumos are accessible in the user namespace on
FreeBSD, and xattrs written to the user namespace on FreeBSD are
accessible by the same name on illumos.
Linux has several xattr namespaces. On Linux, ZFS encodes the
namespace in the xattr name for every namespace, including the user
namespace. As a consequence, an xattr in the user namespace with the
name "foo" is stored by ZFS with the name "user.foo" and therefore
appears on FreeBSD and illumos to have the name "user.foo" rather than
"foo". Conversely, none of the xattrs written on FreeBSD or illumos
are accessible on Linux unless the name happens to be prefixed with one
of the Linux xattr namespaces, in which case the namespace is stripped
from the name. This makes xattrs entirely incompatible between Linux
and other platforms.
We want to make the encoding of user namespace xattrs compatible across
platforms. A critical requirement of this compatibility is for xattrs
from existing pools from FreeBSD and illumos to be accessible by the
same names in the user namespace on Linux. It is also necessary that
existing pools with xattrs written by Linux retain access to those
xattrs by the same names on Linux. Making user namespace xattrs from
Linux accessible by the correct names on other platforms is important.
The handling of other namespaces is not required to be consistent.
Add a fallback mechanism for listing and getting xattrs to treat xattrs
as being in the user namespace if they do not match a known prefix.
Do not allow setting or getting xattrs with a name that is prefixed
with one of the namespace names used by ZFS on supported platforms.
Allow choosing between legacy illumos and FreeBSD compatibility and
legacy Linux compatibility with a new tunable. This facilitates
replication and migration of pools between hosts with different
compatibility needs.
The tunable controls whether or not to prefix the namespace to the
name. If the xattr is already present with the alternate prefix,
remove it so only the new version persists. By default the platform's
existing convention is used.
Reviewed-by: Christian Schwarz <christian.schwarz@nutanix.com>
Reviewed-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Ryan Moeller <ryan@iXsystems.com>
Closes#11919
On FreeBSD vnode reclamation is single-threaded, protected by single
global lock. Linux seems to be able to use a thread per mount point,
but at this time it creates more harm than good.
Reduce number of threads to 1, adding tunable in case somebody wants
to try more.
Reviewed-by: Ryan Moeller <ryan@ixsystems.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Chris Dunlop <chris@onthe.net.au>
Reviewed-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Closes#12896
Issue #9966
Strict hole reporting was previously disabled by default as a
performance optimization. However, this has lead to confusion
over the expected behavior and a variety of workarounds being
adopted by consumers of ZFS. Change the default behavior to
always report holes and force the TXG sync.
Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#12746
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Amanakis <gamanakis@gmail.com>
Signed-off-by: Max Zettlmeißl <max@zettlmeissl.de>
Closes#12784
Special allocation class or dedup vdevs may have roughly the same
performance as L2ARC vdevs. Introduce a new tunable to exclude those
buffers from being cacheable on L2ARC.
Reviewed-by: Don Brady <don.brady@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: George Amanakis <gamanakis@gmail.com>
Closes#11761Closes#12285
When using lseek(2) to report data/holes memory mapped regions of
the file were ignored. This could result in incorrect results.
To handle this zfs_holey_common() was updated to asynchronously
writeback any dirty mmap(2) regions prior to reporting holes.
Additionally, while not strictly required, the dn_struct_rwlock is
now held over the dirty check to prevent the dnode structure from
changing. This ensures that a clean dnode can't be dirtied before
the data/hole is located. The range lock is now also taken to
ensure the call cannot race with zfs_write().
Furthermore, the code was refactored to provide a dnode_is_dirty()
helper function which checks the dnode for any dirty records to
determine its dirtiness.
Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Rich Ercolani <rincebrain@gmail.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #11900Closes#12724
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Ryan Moeller <ryan@ixsystems.com>
Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Signed-off-by: Gordon Bergling <gbergling@googlemail.com>
Closes#12464
* Add Module Parameters Regarding Log Size Limit
zfs_wrlog_data_max
The upper limit of TX_WRITE log data. Once it is reached,
write operation is blocked, until log data is cleared out
after txg sync. It only counts TX_WRITE log with WR_COPIED
or WR_NEED_COPY.
Reviewed-by: Prakash Surya <prakash.surya@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: jxdking <lostking2008@hotmail.com>
Closes#12284
arc_evict_hdr() returns number of evicted bytes in scope of specific
state. For ghost states it does not mean the amount of really freed
memory, but the logical buffer size. It is correct for the eviction
process, but not for waking up threads waiting for ARC size reduction,
as added in "Revise ARC shrinker algorithm" commit, causing premature
wakeups while ARC is still overflowed, allowing even bigger overflow,
plus processing overhead when next allocation will also get blocked,
probably also for too short time.
To fix that make arc_evict_hdr() also return the amount of really
freed memory, which for the ghost states is only the header, and use
it to update arc_evict_count instead. Originally I was thinking to
not return it at all, since arc_get_data_impl() does not account for
the headers, but decided that some slow allocation progress is better
than long waits, reaching on my tests up to 100ms.
To reduce negative latency effects of long time periods when reclaim
thread can free little real memory, start reclamation process earlier,
before we actually reached the overflow threshold, when we have to
throttle new allocations. We can also do it without taking global
arc_evict_lock, reducing the contention.
Reviewed-by: George Wilson <gwilson@delphix.com>
Reviewed-by: Allan Jude <allan@klarasystems.com>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored-By: iXsystems, Inc.
Closes#12279
