431 lines
16 KiB
Text
431 lines
16 KiB
Text
Say you've got a big slow raid 6, and an X-25E or three. Wouldn't it be
|
|
nice if you could use them as cache... Hence bcache.
|
|
|
|
Wiki and git repositories are at:
|
|
http://bcache.evilpiepirate.org
|
|
http://evilpiepirate.org/git/linux-bcache.git
|
|
http://evilpiepirate.org/git/bcache-tools.git
|
|
|
|
It's designed around the performance characteristics of SSDs - it only allocates
|
|
in erase block sized buckets, and it uses a hybrid btree/log to track cached
|
|
extants (which can be anywhere from a single sector to the bucket size). It's
|
|
designed to avoid random writes at all costs; it fills up an erase block
|
|
sequentially, then issues a discard before reusing it.
|
|
|
|
Both writethrough and writeback caching are supported. Writeback defaults to
|
|
off, but can be switched on and off arbitrarily at runtime. Bcache goes to
|
|
great lengths to protect your data - it reliably handles unclean shutdown. (It
|
|
doesn't even have a notion of a clean shutdown; bcache simply doesn't return
|
|
writes as completed until they're on stable storage).
|
|
|
|
Writeback caching can use most of the cache for buffering writes - writing
|
|
dirty data to the backing device is always done sequentially, scanning from the
|
|
start to the end of the index.
|
|
|
|
Since random IO is what SSDs excel at, there generally won't be much benefit
|
|
to caching large sequential IO. Bcache detects sequential IO and skips it;
|
|
it also keeps a rolling average of the IO sizes per task, and as long as the
|
|
average is above the cutoff it will skip all IO from that task - instead of
|
|
caching the first 512k after every seek. Backups and large file copies should
|
|
thus entirely bypass the cache.
|
|
|
|
In the event of a data IO error on the flash it will try to recover by reading
|
|
from disk or invalidating cache entries. For unrecoverable errors (meta data
|
|
or dirty data), caching is automatically disabled; if dirty data was present
|
|
in the cache it first disables writeback caching and waits for all dirty data
|
|
to be flushed.
|
|
|
|
Getting started:
|
|
You'll need make-bcache from the bcache-tools repository. Both the cache device
|
|
and backing device must be formatted before use.
|
|
make-bcache -B /dev/sdb
|
|
make-bcache -C /dev/sdc
|
|
|
|
make-bcache has the ability to format multiple devices at the same time - if
|
|
you format your backing devices and cache device at the same time, you won't
|
|
have to manually attach:
|
|
make-bcache -B /dev/sda /dev/sdb -C /dev/sdc
|
|
|
|
To make bcache devices known to the kernel, echo them to /sys/fs/bcache/register:
|
|
|
|
echo /dev/sdb > /sys/fs/bcache/register
|
|
echo /dev/sdc > /sys/fs/bcache/register
|
|
|
|
To register your bcache devices automatically, you could add something like
|
|
this to an init script:
|
|
|
|
echo /dev/sd* > /sys/fs/bcache/register_quiet
|
|
|
|
It'll look for bcache superblocks and ignore everything that doesn't have one.
|
|
|
|
Registering the backing device makes the bcache show up in /dev; you can now
|
|
format it and use it as normal. But the first time using a new bcache device,
|
|
it'll be running in passthrough mode until you attach it to a cache. See the
|
|
section on attaching.
|
|
|
|
The devices show up at /dev/bcacheN, and can be controlled via sysfs from
|
|
/sys/block/bcacheN/bcache:
|
|
|
|
mkfs.ext4 /dev/bcache0
|
|
mount /dev/bcache0 /mnt
|
|
|
|
Cache devices are managed as sets; multiple caches per set isn't supported yet
|
|
but will allow for mirroring of metadata and dirty data in the future. Your new
|
|
cache set shows up as /sys/fs/bcache/<UUID>
|
|
|
|
ATTACHING:
|
|
|
|
After your cache device and backing device are registered, the backing device
|
|
must be attached to your cache set to enable caching. Attaching a backing
|
|
device to a cache set is done thusly, with the UUID of the cache set in
|
|
/sys/fs/bcache:
|
|
|
|
echo <UUID> > /sys/block/bcache0/bcache/attach
|
|
|
|
This only has to be done once. The next time you reboot, just reregister all
|
|
your bcache devices. If a backing device has data in a cache somewhere, the
|
|
/dev/bcache# device won't be created until the cache shows up - particularly
|
|
important if you have writeback caching turned on.
|
|
|
|
If you're booting up and your cache device is gone and never coming back, you
|
|
can force run the backing device:
|
|
|
|
echo 1 > /sys/block/sdb/bcache/running
|
|
|
|
(You need to use /sys/block/sdb (or whatever your backing device is called), not
|
|
/sys/block/bcache0, because bcache0 doesn't exist yet. If you're using a
|
|
partition, the bcache directory would be at /sys/block/sdb/sdb2/bcache)
|
|
|
|
The backing device will still use that cache set if it shows up in the future,
|
|
but all the cached data will be invalidated. If there was dirty data in the
|
|
cache, don't expect the filesystem to be recoverable - you will have massive
|
|
filesystem corruption, though ext4's fsck does work miracles.
|
|
|
|
ERROR HANDLING:
|
|
|
|
Bcache tries to transparently handle IO errors to/from the cache device without
|
|
affecting normal operation; if it sees too many errors (the threshold is
|
|
configurable, and defaults to 0) it shuts down the cache device and switches all
|
|
the backing devices to passthrough mode.
|
|
|
|
- For reads from the cache, if they error we just retry the read from the
|
|
backing device.
|
|
|
|
- For writethrough writes, if the write to the cache errors we just switch to
|
|
invalidating the data at that lba in the cache (i.e. the same thing we do for
|
|
a write that bypasses the cache)
|
|
|
|
- For writeback writes, we currently pass that error back up to the
|
|
filesystem/userspace. This could be improved - we could retry it as a write
|
|
that skips the cache so we don't have to error the write.
|
|
|
|
- When we detach, we first try to flush any dirty data (if we were running in
|
|
writeback mode). It currently doesn't do anything intelligent if it fails to
|
|
read some of the dirty data, though.
|
|
|
|
TROUBLESHOOTING PERFORMANCE:
|
|
|
|
Bcache has a bunch of config options and tunables. The defaults are intended to
|
|
be reasonable for typical desktop and server workloads, but they're not what you
|
|
want for getting the best possible numbers when benchmarking.
|
|
|
|
- Bad write performance
|
|
|
|
If write performance is not what you expected, you probably wanted to be
|
|
running in writeback mode, which isn't the default (not due to a lack of
|
|
maturity, but simply because in writeback mode you'll lose data if something
|
|
happens to your SSD)
|
|
|
|
# echo writeback > /sys/block/bcache0/cache_mode
|
|
|
|
- Bad performance, or traffic not going to the SSD that you'd expect
|
|
|
|
By default, bcache doesn't cache everything. It tries to skip sequential IO -
|
|
because you really want to be caching the random IO, and if you copy a 10
|
|
gigabyte file you probably don't want that pushing 10 gigabytes of randomly
|
|
accessed data out of your cache.
|
|
|
|
But if you want to benchmark reads from cache, and you start out with fio
|
|
writing an 8 gigabyte test file - so you want to disable that.
|
|
|
|
# echo 0 > /sys/block/bcache0/bcache/sequential_cutoff
|
|
|
|
To set it back to the default (4 mb), do
|
|
|
|
# echo 4M > /sys/block/bcache0/bcache/sequential_cutoff
|
|
|
|
- Traffic's still going to the spindle/still getting cache misses
|
|
|
|
In the real world, SSDs don't always keep up with disks - particularly with
|
|
slower SSDs, many disks being cached by one SSD, or mostly sequential IO. So
|
|
you want to avoid being bottlenecked by the SSD and having it slow everything
|
|
down.
|
|
|
|
To avoid that bcache tracks latency to the cache device, and gradually
|
|
throttles traffic if the latency exceeds a threshold (it does this by
|
|
cranking down the sequential bypass).
|
|
|
|
You can disable this if you need to by setting the thresholds to 0:
|
|
|
|
# echo 0 > /sys/fs/bcache/<cache set>/congested_read_threshold_us
|
|
# echo 0 > /sys/fs/bcache/<cache set>/congested_write_threshold_us
|
|
|
|
The default is 2000 us (2 milliseconds) for reads, and 20000 for writes.
|
|
|
|
- Still getting cache misses, of the same data
|
|
|
|
One last issue that sometimes trips people up is actually an old bug, due to
|
|
the way cache coherency is handled for cache misses. If a btree node is full,
|
|
a cache miss won't be able to insert a key for the new data and the data
|
|
won't be written to the cache.
|
|
|
|
In practice this isn't an issue because as soon as a write comes along it'll
|
|
cause the btree node to be split, and you need almost no write traffic for
|
|
this to not show up enough to be noticable (especially since bcache's btree
|
|
nodes are huge and index large regions of the device). But when you're
|
|
benchmarking, if you're trying to warm the cache by reading a bunch of data
|
|
and there's no other traffic - that can be a problem.
|
|
|
|
Solution: warm the cache by doing writes, or use the testing branch (there's
|
|
a fix for the issue there).
|
|
|
|
SYSFS - BACKING DEVICE:
|
|
|
|
attach
|
|
Echo the UUID of a cache set to this file to enable caching.
|
|
|
|
cache_mode
|
|
Can be one of either writethrough, writeback, writearound or none.
|
|
|
|
clear_stats
|
|
Writing to this file resets the running total stats (not the day/hour/5 minute
|
|
decaying versions).
|
|
|
|
detach
|
|
Write to this file to detach from a cache set. If there is dirty data in the
|
|
cache, it will be flushed first.
|
|
|
|
dirty_data
|
|
Amount of dirty data for this backing device in the cache. Continuously
|
|
updated unlike the cache set's version, but may be slightly off.
|
|
|
|
label
|
|
Name of underlying device.
|
|
|
|
readahead
|
|
Size of readahead that should be performed. Defaults to 0. If set to e.g.
|
|
1M, it will round cache miss reads up to that size, but without overlapping
|
|
existing cache entries.
|
|
|
|
running
|
|
1 if bcache is running (i.e. whether the /dev/bcache device exists, whether
|
|
it's in passthrough mode or caching).
|
|
|
|
sequential_cutoff
|
|
A sequential IO will bypass the cache once it passes this threshhold; the
|
|
most recent 128 IOs are tracked so sequential IO can be detected even when
|
|
it isn't all done at once.
|
|
|
|
sequential_merge
|
|
If non zero, bcache keeps a list of the last 128 requests submitted to compare
|
|
against all new requests to determine which new requests are sequential
|
|
continuations of previous requests for the purpose of determining sequential
|
|
cutoff. This is necessary if the sequential cutoff value is greater than the
|
|
maximum acceptable sequential size for any single request.
|
|
|
|
state
|
|
The backing device can be in one of four different states:
|
|
|
|
no cache: Has never been attached to a cache set.
|
|
|
|
clean: Part of a cache set, and there is no cached dirty data.
|
|
|
|
dirty: Part of a cache set, and there is cached dirty data.
|
|
|
|
inconsistent: The backing device was forcibly run by the user when there was
|
|
dirty data cached but the cache set was unavailable; whatever data was on the
|
|
backing device has likely been corrupted.
|
|
|
|
stop
|
|
Write to this file to shut down the bcache device and close the backing
|
|
device.
|
|
|
|
writeback_delay
|
|
When dirty data is written to the cache and it previously did not contain
|
|
any, waits some number of seconds before initiating writeback. Defaults to
|
|
30.
|
|
|
|
writeback_percent
|
|
If nonzero, bcache tries to keep around this percentage of the cache dirty by
|
|
throttling background writeback and using a PD controller to smoothly adjust
|
|
the rate.
|
|
|
|
writeback_rate
|
|
Rate in sectors per second - if writeback_percent is nonzero, background
|
|
writeback is throttled to this rate. Continuously adjusted by bcache but may
|
|
also be set by the user.
|
|
|
|
writeback_running
|
|
If off, writeback of dirty data will not take place at all. Dirty data will
|
|
still be added to the cache until it is mostly full; only meant for
|
|
benchmarking. Defaults to on.
|
|
|
|
SYSFS - BACKING DEVICE STATS:
|
|
|
|
There are directories with these numbers for a running total, as well as
|
|
versions that decay over the past day, hour and 5 minutes; they're also
|
|
aggregated in the cache set directory as well.
|
|
|
|
bypassed
|
|
Amount of IO (both reads and writes) that has bypassed the cache
|
|
|
|
cache_hits
|
|
cache_misses
|
|
cache_hit_ratio
|
|
Hits and misses are counted per individual IO as bcache sees them; a
|
|
partial hit is counted as a miss.
|
|
|
|
cache_bypass_hits
|
|
cache_bypass_misses
|
|
Hits and misses for IO that is intended to skip the cache are still counted,
|
|
but broken out here.
|
|
|
|
cache_miss_collisions
|
|
Counts instances where data was going to be inserted into the cache from a
|
|
cache miss, but raced with a write and data was already present (usually 0
|
|
since the synchronization for cache misses was rewritten)
|
|
|
|
cache_readaheads
|
|
Count of times readahead occured.
|
|
|
|
SYSFS - CACHE SET:
|
|
|
|
average_key_size
|
|
Average data per key in the btree.
|
|
|
|
bdev<0..n>
|
|
Symlink to each of the attached backing devices.
|
|
|
|
block_size
|
|
Block size of the cache devices.
|
|
|
|
btree_cache_size
|
|
Amount of memory currently used by the btree cache
|
|
|
|
bucket_size
|
|
Size of buckets
|
|
|
|
cache<0..n>
|
|
Symlink to each of the cache devices comprising this cache set.
|
|
|
|
cache_available_percent
|
|
Percentage of cache device free.
|
|
|
|
clear_stats
|
|
Clears the statistics associated with this cache
|
|
|
|
dirty_data
|
|
Amount of dirty data is in the cache (updated when garbage collection runs).
|
|
|
|
flash_vol_create
|
|
Echoing a size to this file (in human readable units, k/M/G) creates a thinly
|
|
provisioned volume backed by the cache set.
|
|
|
|
io_error_halflife
|
|
io_error_limit
|
|
These determines how many errors we accept before disabling the cache.
|
|
Each error is decayed by the half life (in # ios). If the decaying count
|
|
reaches io_error_limit dirty data is written out and the cache is disabled.
|
|
|
|
journal_delay_ms
|
|
Journal writes will delay for up to this many milliseconds, unless a cache
|
|
flush happens sooner. Defaults to 100.
|
|
|
|
root_usage_percent
|
|
Percentage of the root btree node in use. If this gets too high the node
|
|
will split, increasing the tree depth.
|
|
|
|
stop
|
|
Write to this file to shut down the cache set - waits until all attached
|
|
backing devices have been shut down.
|
|
|
|
tree_depth
|
|
Depth of the btree (A single node btree has depth 0).
|
|
|
|
unregister
|
|
Detaches all backing devices and closes the cache devices; if dirty data is
|
|
present it will disable writeback caching and wait for it to be flushed.
|
|
|
|
SYSFS - CACHE SET INTERNAL:
|
|
|
|
This directory also exposes timings for a number of internal operations, with
|
|
separate files for average duration, average frequency, last occurence and max
|
|
duration: garbage collection, btree read, btree node sorts and btree splits.
|
|
|
|
active_journal_entries
|
|
Number of journal entries that are newer than the index.
|
|
|
|
btree_nodes
|
|
Total nodes in the btree.
|
|
|
|
btree_used_percent
|
|
Average fraction of btree in use.
|
|
|
|
bset_tree_stats
|
|
Statistics about the auxiliary search trees
|
|
|
|
btree_cache_max_chain
|
|
Longest chain in the btree node cache's hash table
|
|
|
|
cache_read_races
|
|
Counts instances where while data was being read from the cache, the bucket
|
|
was reused and invalidated - i.e. where the pointer was stale after the read
|
|
completed. When this occurs the data is reread from the backing device.
|
|
|
|
trigger_gc
|
|
Writing to this file forces garbage collection to run.
|
|
|
|
SYSFS - CACHE DEVICE:
|
|
|
|
block_size
|
|
Minimum granularity of writes - should match hardware sector size.
|
|
|
|
btree_written
|
|
Sum of all btree writes, in (kilo/mega/giga) bytes
|
|
|
|
bucket_size
|
|
Size of buckets
|
|
|
|
cache_replacement_policy
|
|
One of either lru, fifo or random.
|
|
|
|
discard
|
|
Boolean; if on a discard/TRIM will be issued to each bucket before it is
|
|
reused. Defaults to off, since SATA TRIM is an unqueued command (and thus
|
|
slow).
|
|
|
|
freelist_percent
|
|
Size of the freelist as a percentage of nbuckets. Can be written to to
|
|
increase the number of buckets kept on the freelist, which lets you
|
|
artificially reduce the size of the cache at runtime. Mostly for testing
|
|
purposes (i.e. testing how different size caches affect your hit rate), but
|
|
since buckets are discarded when they move on to the freelist will also make
|
|
the SSD's garbage collection easier by effectively giving it more reserved
|
|
space.
|
|
|
|
io_errors
|
|
Number of errors that have occured, decayed by io_error_halflife.
|
|
|
|
metadata_written
|
|
Sum of all non data writes (btree writes and all other metadata).
|
|
|
|
nbuckets
|
|
Total buckets in this cache
|
|
|
|
priority_stats
|
|
Statistics about how recently data in the cache has been accessed. This can
|
|
reveal your working set size.
|
|
|
|
written
|
|
Sum of all data that has been written to the cache; comparison with
|
|
btree_written gives the amount of write inflation in bcache.
|