| a2142f3a | 28-Dec-2017 |
GangCao <gang.cao@intel.com> |
bdev/qos: add the QoS rate limiting support on bdev
This patch is to add the basic support of QoS on bdev.
Including two major functionalities:
1. The QoS rate limiting algorithm:
a. New IO will
bdev/qos: add the QoS rate limiting support on bdev
This patch is to add the basic support of QoS on bdev.
Including two major functionalities:
1. The QoS rate limiting algorithm:
a. New IO will be always queued first also under
the no memory condition
b. Start the QoS IO operation based on the limit
c. A poller started in each millisecond to reset
the rate limit and send new IOs down
d. The rate limit is based on the millisecond and
converted from user configurable IOsPerSecond
2. The Master Thread management:
a. Add a per bdev channel_count
b. Whenever QoS is enabled on bdev, if QoS bdev
channel is not created, create the QoS bdev
channel and assign the QoS thread
c. When new IOs coming from different channels
(threads), pass the IOs to the QoS bdev channel
through the thread event
d. When the IOs are completed from the QoS bdev
channel, pass the IOs back to its orignal
channel(thread)
e. Destroy the QoS bdev channel when it is the
last bdev channel for this bdev. Defer the
destruction if current thread is not QoS thread
Change-Id: Ie4444551d7c3c7de52f6513c9db926628796adb4
Signed-off-by: GangCao <gang.cao@intel.com>
Reviewed-on: https://review.gerrithub.io/393136
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Daniel Verkamp <daniel.verkamp@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
show more ...
|
| 583a24a4 | 14-Jan-2018 |
Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com> |
bdev: share nomem_io data between bdevs built on the same device
When there are two bdevs built on the same io_device,
it is possible that one bdev entirely saturates
underlying queue, not letting t
bdev: share nomem_io data between bdevs built on the same device
When there are two bdevs built on the same io_device,
it is possible that one bdev entirely saturates
underlying queue, not letting the second bdev issue
a single I/O. The second bdev will silently fail any
subsequent I/O and append it to the nomem_io list.
However, since we resend I/O only from I/O completion
callback and there's no outstanding I/O for that bdev
(io_outstanding==0), the I/O will never be resent.
It'll be stuck in nomem_io forever.
This patch makes nomem_io list to be shared between
bdevs built on the same device. It is now possible
that I/O completion callback from one bdev will retry
sending I/O from other bdev.
The shared bdev data is based on thread-local
bdev_mgmt_channel, so doesn't need any external
synchronization.
Change-Id: Ia5ac3a1627ce3de4087e43907c329aa7d07ed7c7
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
Reviewed-on: https://review.gerrithub.io/394658
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ziye Yang <optimistyzy@gmail.com>
show more ...
|
| 94bc8cfd | 15-Sep-2017 |
Jim Harris <james.r.harris@intel.com> |
bdev: add ENOMEM handling
At very high queue depths, bdev modules may not have enough
internal resources to track all of the incoming I/O. For example,
we allocate a finite number of nvme_request o
bdev: add ENOMEM handling
At very high queue depths, bdev modules may not have enough
internal resources to track all of the incoming I/O. For example,
we allocate a finite number of nvme_request objects per allocated
queue pair. Currently if these resources are exhausted, the
bdev module will return failure (with no indication why) which
gets propagated all the way back to the application.
So instead, add SPDK_BDEV_IO_STATUS_NOMEM to allow bdev modules
to indicate this type of failure. Also add handling for this
status type in the generic bdev layer, involving queuing these
I/O for later retry after other I/O on the failing channel have
completed.
This does place an expectation on the bdev module that these
internal resources are allocated per io_channel. Otherwise we
cannot guarantee forward progress solely on reception of
completions. For example, without this guarantee, a bdev
module could theoretically return ENOMEM even if there were
no I/O oustanding for that io_channel. nvme, aio, rbd,
virtio and null drivers comply with this expectation already.
malloc only complies though when not using copy offload.
This patch will fix malloc w/ copy engine to at least
return ENOMEM when no copy descriptors are available. If the
condition above occurs, I/O waiting for resources will get
failed as part of a subsequent reset which matches the
behavior it has today.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Iea7cd51a611af8abe882794d0b2361fdbb74e84e
Reviewed-on: https://review.gerrithub.io/378853
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Daniel Verkamp <daniel.verkamp@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
show more ...
|
| 1f935c7a | 12-Sep-2017 |
Jim Harris <james.r.harris@intel.com> |
bdev: properly handle aborted resets
Previously, we naively assumed that a completed
reset was the reset in progress, and would
unilaterally set reset_in_progres to false.
So change reset_in_progre
bdev: properly handle aborted resets
Previously, we naively assumed that a completed
reset was the reset in progress, and would
unilaterally set reset_in_progres to false.
So change reset_in_progress to a bdev_io pointer
instead. If this is not NULL, a reset is not in
progress. Then when a reset completes, we only
set the reset_in_progress pointer to NULL if we
are completing the reset that is in progress.
We also were not aborting queued resets when
destroying a channel so that is fixed here too.
The added unit test covers both fixes above - it will
submit two resets on a different channels, then destroy
the second channel. This will abort the second reset
and check that the bdev still sees the first reset as in
progress.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I61df677cfa272c589ca03cb81753f71b0807a182
Reviewed-on: https://review.gerrithub.io/378199
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
Reviewed-by: Daniel Verkamp <daniel.verkamp@intel.com>
show more ...
|
| ab29d2ce | 12-Sep-2017 |
Jim Harris <james.r.harris@intel.com> |
bdev: improve handling of channel deletion with queued resets
Upper layers are not supposed to put an I/O channel if there
are still I/O outstanding. This should apply to resets as well.
To better
bdev: improve handling of channel deletion with queued resets
Upper layers are not supposed to put an I/O channel if there
are still I/O outstanding. This should apply to resets as well.
To better detect this case, do not remove the reset from
the channel's queued_reset list until it is ready to be
submitted to the bdev module. This ensures:
1) We can detect if a channel is put with a reset outstanding.
2) We do not access freed memory, when the channel is destroyed
before the reset message can submit the reset I/O.
3) Abort the queued reset if a channel is destroyed.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I0c03eee8b3642155c19c2996e25955baac22d406
Reviewed-on: https://review.gerrithub.io/378198
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
Reviewed-by: Daniel Verkamp <daniel.verkamp@intel.com>
show more ...
|