This helps prevent starvation in the case where a thread
is started but remains idle, while other threads consume
all of the spdk_bdev_io buffers in the global pool.
This starvation issue is fairly theoretical at this point,
but future patches will be adding the ability for callers
to be notified when an spdk_bdev_io becomes available if
the pool is exhausted. We will add tests to stress
pool exhaustion at which point this patch will become
much more important.
While here, increase the minimum bdev_io_pool_size to
account for the mgmt_ch getting destroyed and then
immediately created again on the master core. In this
case there is a window where both channels exist at
once - the one being destroyed won't free its cached
spdk_bdev_ios until the deferred spdk_put_io_channel
event executes.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I3a2fc80bc2bfd78b098bcbfce456d7a433cd64e9
Reviewed-on: https://review.gerrithub.io/415039
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>