During RocksDB testing with MySQL, we found cases
where blobfs would try to update and sync the length
xattr on the underlying blob while an existing update
was already in progress. This was primarily driven
by RocksDB performing appends and syncs on the log
file from multiple writer threads.
The simplest way to fix this is to just process one
sync_request at a time. There could be a tiny bit
of inefficiency here if multiple threads are appending
and syncing a file in parallel - we can look at some
additional optimizations if we find a case where that
is noticeable.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Reported-by: Changpeng Liu <changpeng.liu@intel.com>
Tested-by: Changpeng Liu <changpeng.liu@intel.com>
Change-Id: I7ab7814494d365bae8716efd0b828337286cc7b7
Reviewed-on: https://review.gerrithub.io/369490
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
If a file is extended and then the new region is immediately
read without being written, there as a math error in the
buffer cache.
Change-Id: Ibd2bbe98c734f98df43eada799ed62de4081964b
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/367265
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
The math works out so that index is always in range here, but add an
assert anyway to catch potential mistakes in the future and to placate
scan-build.
Change-Id: I09ad37dde56bc315543b3c142d91ca9a49e9b32c
Signed-off-by: Daniel Verkamp <daniel.verkamp@intel.com>
Reviewed-on: https://review.gerrithub.io/364530
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Synchronous blobfs channels may allocate requests from
the synchronous thread, but free the request from the
async thread - especially for code that is shared between
sync and async modes. So add a lock to the request list
for sync channels only.
Note this only affects metadata path (i.e. sync, delete, open)
and not I/O path.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I932c500807c2f459c697fab2ffd91a88b88b0c87
Reviewed-on: https://review.gerrithub.io/362964
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Daniel Verkamp <daniel.verkamp@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
The file's lock does not need to be held across
calls to this function as the file is not dereferenced
at all.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Idf0a52f2528521f12e45963bb0ab1f414b37380f
Reviewed-on: https://review.gerrithub.io/363138
Reviewed-by: Daniel Verkamp <daniel.verkamp@intel.com>
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
- rename spdk_malloc_socket to spdk_dma_malloc_socket
- rename spdk_malloc to spdk_dma_malloc
- rename spdk_zmalloc to spdk_dma_zmalloc
- rename spdk_realloc to spdk_dma_realloc
- rename spdk_free to spdk_dma_free
Change-Id: I52a11b7a4243281f9c56f503e826fd7c4a1fd883
Signed-off-by: John Meneghini <johnm@netapp.com>
Reviewed-on: https://review.gerrithub.io/362604
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
This wasn't used anywhere and we currently believe there
are superior software-only techniques for controlling
quality of service.
Change-Id: Icdadd5870ed0629b338c307d2619bbc242c3e7a3
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/362065
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Keep a global counter of the number of opened
blobfs instances. Allocate the cache pool when the
first instance is opened, and free the cache pool
when the last instance closes.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I99db2b587b738badcd2a54c9f5dc6ec8370ce22c
Reviewed-on: https://review.gerrithub.io/362606
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
This is no longer used anywhere. For the places where we previously
used it, we've since found alternate solutions that do not
require it.
Change-Id: I738a80b95ef50348ce1c14969a3812b0a625b3fd
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/362064
Tested-by: <sys_sgsw@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
This isn't strictly necessary, but follow the
pattern for sync and I/O devices.
Change-Id: I34b1bd6f9da06008e419f55ee9810022f5fed3d5
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/361674
Tested-by: <sys_sgsw@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
This allows us to no longer use the unique=true flag as well
as assign sync channels different queue depths.
Change-Id: I0b3129a3139bf39d745c2a4dc1d9f402c0a8faa8
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/361673
Tested-by: <sys_sgsw@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
This allows us to have a different queue depth
on I/O channels than on metadata/sync channels
Change-Id: I30dfc95af550082efffec4b12ed6c4cd4b8c11cf
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/361672
Tested-by: <sys_sgsw@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
This temporarily increases the size of the sync channel
to 512 (like all of the other channels). This will be
fixed in a later patch.
Change-Id: I679f1cb7c9f7afdad0acf9a3a26f9632f60cc510
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/361670
Tested-by: <sys_sgsw@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
The channel memory isn't allocated by these
libraries, so they can't free it.
Change-Id: I30909fa4e77bc5a41b45230f04ba5fe75b172dbf
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
This will need to be configured globally for all channels.
Change-Id: I773252f220373617f8d09d1f24243db8095cf8a4
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
This patch address the issue:
https://github.com/spdk/spdk/issues/151.
For cache_append_no_cache in cache_ut testcase,
there is resource contention for buffer among two
threads in the following two functions.
Thread 0: cache_free_buffers
Thread1: __file_flush_done
When the thread1 execuctes __file_flush_done,
it calls the call back: __sem_post defined in
following statement in spdk_file_sync
_file_sync(file, channel, __sem_post, &channel->sem);
Thus Thread 0 will execute next function
cache_buffers, and it frees the buffer.
Then Thread 1 continues executing the remaining statements
in __file_flush_done with the assert function, and touches
the space already freed.
So it will be safe to move ahead the next buffer check.
Change-Id: Ic007b3481f4e3a17d47eeca5c9c802001949a5ab
Signed-off-by: Ziye Yang <ziye.yang@intel.com>
The code will locks clear to put those definitions into tree.h header
Change-Id: Ib1a34f19d9849acd7ea979eb0a6e153b0e8e39de
Signed-off-by: Ziye Yang <ziye.yang@intel.com>
This avoids corner case where a buffer gets allocated on the 100th
try.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: If65053d539d458d9a53c8850bbb4cbe4ee84f604
This is the initial commit for "blobfs", a lightweight
filesystem built on top of the SPDK blobstore.
Also included in this patch:
1) a shim for using SPDK bdevs as the backing store for
SPDK blobstore/blobfs
2) documentation for using blobfs as the storage engine
with RocksDB
3) scripts for running a set of workloads and collecting
profiling data with RocksDB and blobfs
See doc/blobfs/getting_started.md included in this commit
for more details on blobfs, including some of the current
limitations.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I2a6d3d4b87236730051228ed62c0c04e04c42c73
