This series implements an active and synchronous mirroring mode.
Currently, the mirror block job is passive an asynchronous: Depending on
your start conditions, some part of the source disk starts as "dirty".
Then, the block job will (as a background operation) continuously copy
dirty parts to the target disk until all of the source disk is clean.
In the meantime, any write to the source disk dirties the affected area.
One effect of this operational mode is that the job may never converge:
If the writes to the source happen faster than the block job copies data
to the target, the job can never finish.
When the active mode implemented in this series is enabled, every write
request to the source will automatically trigger a synchronous write to
the target right afterwards. Therefore, the source can never get dirty
faster than data is copied to the target. Most importantly, once source
and target are in sync (BLOCK_JOB_READY is emitted), they will not
diverge (unless e.g. an I/O error occurs).
Active mirroring also improves on a second issue of the passive mode: We
do not have to read data from the source in order to write it to the
target. When new data is written to the source in active mode, it is
automatically mirrored to the target, which saves us the superfluous
read from the source.
Things to do on top of this series:
- Allow switching between active and passive mode at runtime: Mainly
hinges on the question of how to expose it to the user (ideally
through a generic block-job-set-option command)
- Implement an asynchronous active mode (launch both write operations to
the source and the target at the same time, and do not wait for the
target operation to finish)
- Integrate the mirror BDS more tightly into the BDS graph: Both source
and target should be BdrvChildren (and the source should not be the
"backing" child). I'm working on this in a follow-up.
- Improve the mirror job coroutine use: Currently more of a hack, a
follow-up will make this nicer.
- Add read-write-blocking mode: This series adds the write-blocking
mode, where every write blocks until the data has been mirrored to the
target. read-write-blocking would also mirror data on reads from the
source, which saves some performance (because that data does not have
to be read twice) at the cost of latency on mirroring read operations.
(Will be in the same follow-up.)
v2:
- Dropped some work, moving it to a follow-up series (making source the
file child, adding the read-write-blocking mode)
- Patches 1 and 2: Dropped BdrvDeletedStatus, instead replacing it by
bdrv_ref()/bdrv_unref() pairs (with some precautions so we don't get
into an infinite recursion)
- And added a test for this, patch 3
- Patch 4: Just rebase conflicts
- Patch 5: Reworked the use of bytes_copied/now "bytes_handled", fixed
the mirror_co_read() comment -- everything will be made nicer
in a follow-up, I promise!
- Patch 7: mirror_wait_for_free_in_flight_slot() may not wait put its
own coroutine into a pseudo operation's waiting queue, or we may
starve. (I noticed this when running the iotest added by this
series under qcow v1, so the test is there!)
- Patch 8: Drop the MirrorBlockJob.source field
- Patch 10: Fixed the function description, and there was a rebase
conflict (due to an additional user of hbitmap_iter_next()).
- Patch 11: Block bitmaps use bytes instead of sectors now, so this
patch gets a bit easier
- Patch 12: Rebase conflict because of the change to patch 7
- Patch 14:
- Dropped read-write-block mode (or rather, put it off to a follow-up)
- Renamed "passive" -> "background" and
"active-write" -> "write-blocking"
- Some rebase conflicts
- Reset dirty bitmap before the active write, re-set it on failure
(instead of only clearing it on success); important to prevent
double-writes (from background and active operations)
- Use a bounce buffer for active operations
- Patch 15:
- Renamed "passive" -> "background" and
"active-write" -> "write-blocking"
- Aimed for 2.12 now...
- Patch 16: Use BlockBackends and aio_write instead of an own write -B
git-backport-diff to v1:
Key:
[] : patches are identical
[] : number of functional differences between upstream/downstream patch
[down] : patch is downstream-only
The flags [FC] indicate (F)unctional and (C)ontextual differences, respectively
001/16:[0054] [FC] 'block: BDS deletion during bdrv_drain_recurse'
002/16:[down] 'block: BDS deletion in bdrv_do_drained_begin()'
003/16:[down] 'tests: Add bdrv-drain test for node deletion'
004/16:[0006] [FC] 'block/mirror: Pull out mirror_perform()'
005/16:[0050] [FC] 'block/mirror: Convert to coroutines'
006/16:[] [-C] 'block/mirror: Use CoQueue to wait on in-flight ops'
007/16:[0021] [FC] 'block/mirror: Wait for in-flight op conflicts'
008/16:[0016] [FC] 'block/mirror: Use source as a BdrvChild'
009/16:[] [--] 'block: Generalize
