diff options
| author | Etienne Dechamps <[email protected]> | 2012-07-31 10:45:37 +0200 |
|---|---|---|
| committer | Brian Behlendorf <[email protected]> | 2012-07-31 09:46:09 -0700 |
| commit | 7c0e570888532b3cecf66459e55688df18946be0 (patch) | |
| tree | 2f78e22bc0a86c02e9d07f36781b74e699deaf1a | |
| parent | 330d06f90d143b41b276796526a66a1c1fff046d (diff) | |
Limit the number of blocks to discard at once.
The number of blocks that can be discarded in one BLKDISCARD ioctl on a
zvol is currently unlimited. Some applications, such as mkfs, discard
the whole volume at once and they use the maximum possible discard size
to do that. As a result, several gigabytes discard requests are not
uncommon.
Unfortunately, if a large amount of data is allocated in the zvol, ZFS
can be quite slow to process discard requests. This is especially true
if the volblocksize is low (e.g. the 8K default). As a result, very
large discard requests can take a very long time (seconds to minutes
under heavy load) to complete. This can cause a number of problems, most
notably if the zvol is accessed remotely (e.g. via iSCSI), in which case
the client has a high probability of timing out on the request.
This patch solves the issue by adding a new tunable module parameter:
zvol_max_discard_blocks. This indicates the maximum possible range, in
zvol blocks, of one discard operation. It is set by default to 16384
blocks, which appears to be a good tradeoff. Using the default
volblocksize of 8K this is equivalent to 128 MB. When using the maximum
volblocksize of 128K this is equivalent to 2 GB.
Signed-off-by: Brian Behlendorf <[email protected]>
Closes #858
| -rw-r--r-- | module/zfs/zvol.c | 7 |
1 files changed, 6 insertions, 1 deletions
diff --git a/module/zfs/zvol.c b/module/zfs/zvol.c index 125d58de9..32e4f3c1e 100644 --- a/module/zfs/zvol.c +++ b/module/zfs/zvol.c @@ -49,6 +49,7 @@ unsigned int zvol_inhibit_dev = 0; unsigned int zvol_major = ZVOL_MAJOR; unsigned int zvol_threads = 32; +unsigned long zvol_max_discard_blocks = 16384; static taskq_t *zvol_taskq; static kmutex_t zvol_state_lock; @@ -1242,7 +1243,8 @@ __zvol_create_minor(const char *name) blk_queue_physical_block_size(zv->zv_queue, zv->zv_volblocksize); blk_queue_io_opt(zv->zv_queue, zv->zv_volblocksize); #ifdef HAVE_BLK_QUEUE_DISCARD - blk_queue_max_discard_sectors(zv->zv_queue, UINT_MAX); + blk_queue_max_discard_sectors(zv->zv_queue, + (zvol_max_discard_blocks * zv->zv_volblocksize) >> 9); queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zv->zv_queue); #endif #ifdef HAVE_BLK_QUEUE_NONROT @@ -1446,3 +1448,6 @@ MODULE_PARM_DESC(zvol_major, "Major number for zvol device"); module_param(zvol_threads, uint, 0444); MODULE_PARM_DESC(zvol_threads, "Number of threads for zvol device"); + +module_param(zvol_max_discard_blocks, ulong, 0444); +MODULE_PARM_DESC(zvol_max_discard_blocks, "Max number of blocks to discard at once"); |