[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Virt Storage/Block meeting minutes - 01-Jul-2020
From: |
John Ferlan |
Subject: |
Virt Storage/Block meeting minutes - 01-Jul-2020 |
Date: |
Wed, 22 Jul 2020 08:49:31 -0400 |
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101 Thunderbird/68.9.0 |
<forgot to send these after the last meeting>
Meeting date/time: Jul 01, 13:00 UTC
Previous meetings: http://etherpad.corp.redhat.com/PlatformVirtStorageMeetings
Prior meeting:
http://etherpad.corp.redhat.com/PlatformVirtStorageMeetings-2020-Jun-10
Join the meeting:
Bluejeans: https://redhat.bluejeans.com/1471327918/
(phone numbers are available in the website above)
*4 to mute/unmute
IRC: #block-call
Attendees: jferlan, philmd, pkrempa, mreitz, stefanha, eblake
Updates and announcements:
* FYI - QEMU 5.1 Upstream
* softfreeze 2020-07-07
* hardfreeze 2020-07-14
* release 2020-08-11 or 2020-08-18
* FYI - RHEL schedules:
* 8.x-AV - https://pp.engineering.redhat.com/pp/product/rhel_av/schedule
*
http://batcave.lab.eng.brq.redhat.com/reports/full-rhel8/short_report.html
* 8.2.1 - Changes require exception+/blocker+
* GA: Tue 2020-07-28
* 8.3.0 - Will get rebase of qemu-5.1 and libvirt-6.7
* Feature Freeze: Tue 2020-09-01
* Bug Freeze: Mon 2020-09-14
* Final Freeze: Tue 2020-10-06
* 8.3.0
* Backport from RHEL AV 8.2.1:
* https://bugzilla.redhat.com/show_bug.cgi?id=1844296
* Unless High/Urgent, avoid clone from RHEL AV 8.3.0+
* 7.9 - changes require blocker+
* GA: Tue 2020-08-11
* All future changes occur via zstream processing
* FYI - Layered Products
* RHV 4.4 will use on RHEL-AV-8.2.x.
* GA Planned: 2020-08-04
* OSP
* OSP 15 and 16.0 are using RHEL-AV-8.1.1
* OSP 16.1 compose (as of 11-Mar-2020) is using RHEL-AV-8.2
* GA Planned: June 2020
* CNV will use RHEL-AV-8.2.x and stay up to date with RHEL-AV in future
versions
* 2.3 GA: 2020-05-04
* 2.4 GA Planned: 2020-06-24
* FYI - General QEMU/libvirt Release Rebase schedule:
https://docs.google.com/spreadsheets/d/1izPBg1YSeEHDcuy8gtfcbVgHdx_xkRYubHnbe6BkV8c
----------------------------------------------------------------------------------------------------------------------------
Projects:
* Incremental Backup
* 8.2.1: https://projects.engineering.redhat.com/browse/RHELPLAN-39267
* Completed:
* bz1779904 - RFE: ability to estimate bitmap space utilization for qcow2
* bz1779893 - RFE: Copy bitmaps with qemu-img convert
* bz1778593 - Qemu coredump when backup to a existing small size image
* bz1804593 - RFE: Allow incremental backups with externally-created
snapshots
* Peter generated kbase/docs on usage
* Moved to 8.3.0
* bz1780409 - RFE: Add block-dirty-bitmap-generate block job to
facilitate bitmap chain recovery
* Hope to get this completed before qemu-5.1 soft freeze
* plan for qemu is to get x-block-dirty-bitmap-populate into 5.1 with
x- prefix, to allow us the ability to change interface as we experiment with
it, since libvirt is not depending on it yet
* Questions/Discussion over current implementation
*
* RHV progress [nsoffer, eshenitz]
* not present
* 8.3.0: https://projects.engineering.redhat.com/browse/RHELPLAN-38611
* libvirt: [pkrempa]
* Need bz for libvirt work to utilize qemu block-dirty-bitmap-generate
bz1780409
*
* bz1829829 - backups after disk hotplug
* refactored some internals which will make fixing this simpler
* Patches are done but need to be tested prior to sending
* bz1799011 - bitmaps + live migration
*
* bz1799010 - bitmaps + virDomainBlockPull
* [jferlan] Previous meeting minutes indicate this is not needed by
RHV, so should we close the bz and remove from Epic then?
* patches are done but need to be tested prior to sending
* bz1799015 - Enable Incremental Backup by default
* Update depends on?
* all of the above :)
* qemu [eblake, kwolf, mreitz]
* bz1780409 - RFE: Add block-dirty-bitmap-generate block job to
facilitate bitmap chain recovery (eblake)
* Hope to get this completed before qemu-5.1 soft freeze
* v3 sent upstream, review had some command suggestions, and need to
investigate those options more
* Considering using the x- prefix before softfreeze to allow for
changes
* bz1790492 - bitmaps + live migration (mreitz)
* First non-RFC posted:
https://lists.nongnu.org/archive/html/qemu-block/2020-06/msg01669.html
* goal to get this in before 5.1 softfreeze
* bz1801625 - block-stream job without dirtying bitmaps (kwolf)
* [jferlan] Previous meeting minutes indicate this may not be
necessary, so should we close the bz and remove from Epic then?
* Will require mreitz series "Deal with filters" to expose the corner
case
* bz1780416 - RFE: qemu-img check --repair should optionally remove any
corrupted bitmaps (virt-maint)
* [Need owner]
* QEMU Backlog - Need to determine need, owner, and possible release
* bz1814664 - backup in push mode will actually write ZEROes to target
image [eblake]
*
* bz1816692 - when scratch file has no space during pull-mode backup, the
backup job will still run, this will produce a inconsistent backup file
[virt-maint]
*
* bz1802401 - The actual size of backup image bigger than base image after
dd data file in guest [virt-maint]
*
* bz1784266 - Error prompt info when do incremental backup with an invalid
"bitmap-mode" [virt-maint]
*
----------------------------------------------------------------------------------------------------------------------------
* Project Cheetah: [stefanha]
* Cheetah - performance improvements for fast storage in KVM:
https://docs.engineering.redhat.com/display/RHELPLAN/Cheetah+-+performance+improvements+for+fast+storage+in+KVM
* bz1827722 - QEMU multi-queue virtio-blk/scsi by default and associated
scalability optimizations that eliminate interprocessor interrupts.
* Benchmarked guest driver NUMA-awareness but there was no measurable
improvement, leaving it for now
* Found that automatic irq affinity of virtqueue interrupts does not
honor NUMA, so completion interrupts may be handled on a CPU in another NUMA
node. This is a problem for devices with few interrupts. Not a problem for true
multiqueue devices because we want an irq on each CPU anyway.
* Userspace cannot override irq affinity when the guest driver uses
automatic irq affinity, e.g. virtio_blk.ko
* bz1827756 - virtio-blk io_uring passthrough to provide guests direct
access to host storage APIs [sgarzare]
* Stefano on PTO
* QEMU nvme [philmd]
* series posted: "cleanups required to use multiple queues"
* https://lists.gnu.org/archive/html/qemu-devel/2020-06/msg10136.html
* Preliminary step required for BZ#1827750
* Add support for sharing PCI devices between nvme:// block driver
instances"
* qemu-storage-daemon [kwolf]
* Updates on Coiby Xu's vhost-user-blk server patches
* Review is moving along, getting closer to merge
* Upstream: [PATCH v6 00/12] monitor: Optionally run handlers in
coroutines
*
https://lists.nongnu.org/archive/html/qemu-devel/2020-05/msg08018.html
* bz1827724 - qemu-storage-daemon SPDK-like polling mode I/O [stefanha]
* No change/update
* virtio_blk.ko completion polling [stefanha]
* Update on extend cpuidle-haltpoll to peek at virtqueues w/ mtosatti
* No change/update
* bz1519010 - QEMU multi-queue block layer support [bonzini]
* Eliminate CPU bottlenecks by spreading I/O across CPUs on the host
* No change/update
* bz1806887 - High IOPS storage performance optimizations [stefanha]
* Generic BZ tracking upstream changes targeting RHEL-AV-8.3.0 due to a
rebase
* No change/update
----------------------------------------------------------------------------------------------------------------------------
* virtio-fs [mreitz, slp]
* Rust virtiofsd implementation - packaging progress/concerns
* Update to packaging conundrum?
* Upstream cloud-hypervisor community is moving virtio-fs daemon to
separate git repo \o/
* Contact w/ Don Bayly (Intel Partner Manager)
* Upstream code to enable on s390x (Marc Hartmayer <mhartmay@linux.ibm.com>)
*
* kubevirt/CNV
* Initial PR from Vladik: https://github.com/kubevirt/kubevirt/pull/3493
* kvm.ko async pf for virtio-fs DAX [future work]
* Making progress but Vivek is concerned that there is no mechanism yet
that is suitable for raising faults to the guest
----------------------------------------------------------------------------------------------------------------------------
* LUKS encryption slot management [mlevitsk, mreitz]
* Issues with build/test on PULL, Max has reposted upstream with iotest
adjustments
* Going to try to make soft freeze
----------------------------------------------------------------------------------------------------------------------------
* Project Capybara [areis, jsnow, stefanha]
* Capybara weekly meeting details:
https://docs.google.com/document/d/12N_Ml0_ZFEKgS609sDu7EaaU9ePYBM1QXV52sXVQy8o
* OpenShift Virtualization Storage presentation (pre-meeting):
https://docs.google.com/presentation/d/1iRBx15NMtWqeLE5QTQAY2W77GDBOnd4jRwDXyzif-hs/edit
* Various discussion with extended details on capybara-list
* See Kubernetes CSI (https://github.com/kubernetes-csi)
qemu-storage-daemon plugin​
* If qemu-storage-daemon is a storage provider, can we provide backups,
NVMe userspace driver, storage migration, I/O throttling, etc? The answer to
many of these is "yes" although we need to fit it into the CSI API.
*
----------------------------------------------------------------------------------------------------------------------------
* virtio-blk vDPA [stefanha]
* Mellanox looking at VFIO, might not use vDPA
----------------------------------------------------------------------------------------------------------------------------
* virtio-blk 2.0 [stefanha] - from <virt-devel>
* Please see discussion on mailing list if you are interested in details
----------------------------------------------------------------------------------------------------------------------------
* NVMe emulation device maintenance problem
* Only "include/block/nvme.h" is important for Red Hat
* Keith - current maintainer - is not very active
* recent high activity from Samsung and WDC
* Kevin suggests Klaus from Samsung step in to help
*
-----------------------------------------------------------------------------------------------------------------------
Next meeting will be 22-Jul. The 08-Jul meeting was canceled
* <feel free to add more>
- Virt Storage/Block meeting minutes - 01-Jul-2020,
John Ferlan <=