| Age | Commit message (Collapse) | Author |
|---|
|
6% of physical RAM is info.totalram >> 4 not info.totalram << 4.
Signed-off-by: Chris Webb <chris@arachsys.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
Before 326d7c1, the shrinker used freeram and totalram from a struct
sysinfo (constructed from /proc/meminfo) to target 25% free physical
memory. As well as the slowness of repeatedly reading /proc/meminfo,
this was a problem as freeram rises when the system starts to swap.
We don't want swapping to reduce our estimate of memory pressure.
To work around this, in 326d7c1 the shrinker started to use the total
allocated heap from a glibc-specific interface mallinfo2(), aiming to
shrink such that our heap is less than 80% of physical memory, unless
overall free memory is less than 6% so that becomes the determining factor.
Unfortunately, a sign error in the calculation means this heuristic
never worked. It would shrink aggressively when the process was small,
and not at all when the process grew beyond 80% of physical RAM. Only the
fallback test ensuring the free physical RAM doesn't fall below 6% would
actually kick in under memory pressure. It also breaks portability to
anything other than recent glibc.
Later, in 2440469 the mallinfo2() was replaced with the older mallinfo()
to improve compatibility with older glibc. This is even more problematic:
it's still not portable but also struct mallinfo has (signed) int fields
which overflow for large processes on 32-bit machines with a 3G/1G split.
Rather than trying to use libc-specific debug interfaces and our own heap
to inform the shrinker, use the information about free and total swap
we already have from sysinfo(2) to explicitly compensate for swapping
in our estimate of free physical memory. Target free memory of 6% of
physical RAM adjusted for zero swap use when calculating the pressure
on the shrinker, based on the effective behaviour of 326d7c1 in practice
given the sign error.
As well as fixing portability to non-glibc systems, this loosens the
assumption that we are the only process using significant memory when
setting the shrinker target. It wouldn't be unreasonable to run two
fsck jobs against independent devices on a large RAM machine and want to
balance physical RAM between them.
Signed-off-by: Chris Webb <chris@arachsys.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
Use a single sysinfo(2) call to fill out struct sysinfo instead of
multiple libc sysconf(3) requests, which will only make sysinfo(2) calls
internally anyway. This also enables us to access other struct sysinfo
fields, not just the three filled-out previously.
As we provide our own definition of struct sysinfo in include/linux/mm.h
to match the kernel, which is not guaranteed to align with the definition
libc provides in <sys/sysinfo.h>, use syscall(SYS_sysinfo, ...) directly
instead of the libc wrapper.
Signed-off-by: Chris Webb <chris@arachsys.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
Signed-off-by: Alexander Fougner <fougner89@gmail.com>
|
|
We seem to be hitting a rare crash in the exit path of fsck - when
shutting down the shrinker thread. Disable exiting the shrinker thread
as a workaround.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
flush check more rigorous
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
Reading from /proc/meminfo is really slow
We don't want to start swapping to disk.
Deceptively, memory available goes up when we start to swap to disk making
performance even worse.
To mitigate this:
1. replace reading from meminfo with proper system calls.
2. attempt to lock allocations in physical memory space.
3. check our own allocated memory instead of available memory.
4. still check available memory in the off chance we're trying to play
nice with other apps.
Signed-off-by: Daniel Hill <daniel@gluo.nz>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
the command line
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
|
|
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
If io_submit() returns -EAGAIN, that just means the io context is full
and we need to wait for completions - no need to die.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
userspace
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
Signed-off-by: Torge Matthies <openglfreak@googlemail.com>
|
|
rebalance_work()
|
|
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
into multiple functions
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
bch2_check_discard_freespace_key()
|
|
mode in userspace
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
backpointers in alloc keys
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
bch2_check_extents_to_backpointers()
|
|
A recent libbcachefs update accidentally committed a change that dropped
the O_DIRECT flag - we definitely didn't want to do that.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
unions
|
|
bch2_btree_iter_peek_and_restart_outlined()
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
|
|
btree roots for all alloc btrees
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
- Spin up a background thread to call the shrinkers every 1 second
- Memory allocations will only call reclaim after a failed allocation,
not every single time
This will be a major performance boost on allocation intensive
workloads.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
bch2_get_alloc_in_memory_pos()
|
|
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
|
|
bch2_extent_update()
|
|
fastpath time source
|
|
|
|
This would correspond to GFP_RECLAIM in the kernel - but we don't
distinguish between different types of reclaim here.
This solves a deadlock in the btree node memory allocation path - we
allocate with the btree node cache lock held but without GFP_KERNEL set.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
bch2_btree_delete_range_trans() now uses peek_upto()
|
|
|
|
bch2_btree_delete_range_trans()
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
|
|
|
|
|
|
bch2_btree_delete_range() to bch2_trans_run()
|
|
We were returning sectors instead of bytes.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
printk_ratelimited was behind an #ifdef CONFIG_PRINTK, which we don't
define, so it was a complete noop - oops.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
We're reading device metadat in mostly sequential order - buffered IO
will be faster than O_DIRECT.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
|
|
|
|
journal entries
|
|
When deleting an entry from a heap that was at entry h->used - 1, we'd
end up calling heap_sift() on an entry outside the heap - the entry we
just removed - which would end up re-adding it to the heap and deleting
something we didn't want to delete. Oops...
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
After memory allocation failure, don't rely on /proc/meminfo to figure
out how much memory we should free - instead unconditionally free 1/8th
of each cache.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
