package qcow-tool
Install
Dune Dependency
Authors
Maintainers
Sources
sha256=0810e97f44d0105f9e183e4b744d619eeb8d1d531df9ff6de13dc00bf4dc6ecd
md5=fce49e5bdf48ee79bc334e41e562d2d1
CHANGES.md.html
0.10.0 (2017-05-13)
fix a major performance problem with
compact
split into 2 packages: qcow and qcow-tool
add
qcow-tool dehydrate
andqcow-tool rehydrate
for extracting metadata for debug/supportadd prometheus metrics for I/O and GC operations
restore the
qcow-tool compact --progress
progress baradd
qcow-tool compact --progress-fd
for json-formatted progressbuild via jbuilder
0.9.5 (2017-03-12)
CLI:
check
andsha
will nolonger resize the file as a side-effect (#84)Allow the number of
cluster_bits
to be set increate
0.9.4 (2017-03-07)
Strictly enforce the cluster move state machine
Don't start moving new blocks while existing moves are in progress (fix bug where the same destination block could be reused)
Hold a lock to exclude
flush
while updating references to ensure reference updates hit the disk before the move is considered completeSimplify allocator by always adding blocks to the Roots set before returning. The caller must transfer them somewhere else.
Simplify the cluster moving API by combining
get_moves
withstart_moves
, so it's not possible to block and affect the moves which can legally be startedWhen detecting a duplicate reference or hitting an I/O error, log analysis of the internal state
Check for move cancellation before copying a block to avoid accidentally copying a block which is now outside the file
Avoid adding a cluster to the Junk set twice during a reference update
Add lots of assertions
0.9.3 (2017-03-02)
Hold a read lock on the L1 during read/write
Minimise locking while updating references
When moving an L2 cluster, update the cluster map
0.9.2 (2017-02-26)
Don't hold the global lock while updating references
Log an error if a client I/O takes more than 30s
Improve the performance of discard by writing each L2 cluster to disk only once
Track clusters which are being erased and copied into, to prevent the file being shrunk, orphaning them (which typically manifests as a later double-allocation)
0.9.1 (2017-02-25)
Add configuration
runtime_assert
to check GC invariants at runtimeUse tail-recursive calls in the block recycler (which deals with large block lists)
Wait for the compaction work list to stabilise before processing it (otherwise we move blocks which are then immediately discarded)
Track the difference between blocks on the end of the file being full of zeroes due to ftruncate versus being full of junk due to discard
On open, truncate the file to erase trailing junk
Don't try to use free space between header structures for user data since we assume all blocks after the start of free space are movable and header blocks aren't (in this implementation)
Make cluster locks recursive, hold relevant metadata read locks while reading or writing data clusters to ensure they aren't moved while we're using them.
Add a debug testing mode and use it in a test case to verify that compact mid-write works as expected.
0.9.0 (2017-02-21)
Add online coalescing mode and background cluster recycling thread
Rename internal modules and types
Ensure the interval tree remains balanced to improve performance
0.8.1 (2017-02-13)
fix error in META file
0.8.0 (2017-02-13)
update to Mirage 3 APIs
now requires OCaml 4.03+
ensure the interval tree is kept balanced
0.7.2 (2016-12-21)
if
discard
is not enabled, faildiscard
callsif
discard
is enabled, enable lazy-refcounts and zero refcount clusters to avoid breaking refcounts overdiscard
,compact
0.7.1 (2016-12-15)
speed up
check
andcompact
up to 50xqcow-tool compact
work around files which aren't a whole number of sectors
0.7.0 (2016-12-10)
now functorised over
TIME
allow background compact to be cancelled
cancel background compact to allow regular I/O to go through
don't trigger the background compact until 1s after the last
discard
on
connect
, sanity-check the image
0.6.0 (2016-12-04)
rename ocamlfind package from
qcow-format
toqcow
for uniformityadd support for runtime configuration arguments to
connect
andcreate
add support for
discard
(aka TRIM or UNMAP) and online compaction (through a stop-the-world GC)switch the build from
oasis
totopkg
(thanks to @jgimenez)
0.5.0 (2016-11-26)
resize
now takes a new size in bytes (rather than sectors) and uses a labelled argumentqcow-tool info
now takes a--filter <expression>
for exampleqcow-tool info ... --filter .size
to view the virtual size
0.4.2 (2016-09-21)
Don't break the build if
Block.connect
has optional arguments
0.4.1 (2016-08-17)
Remove one necessary source of
flush
callsCLI: add
mapped
command to list the mapped regions of a file
0.4 (2016-08-03)
For buffered block devices, call
flush
to guarantee metadata correctnessIn lazy_refcounts mode (the default), do not compute any refcounts
CLI: the
repair
command should recompute refcounts
0.3 (2016-05-12)
Depend on ppx, require OCaml 4.02+
0.2 (2016-01-15)
Use qcow version 3 by default, setting
lazy_refcount=on
Unit tests now verify that
qemu-img check
is happy and thatqemu-nbd
sees the same data we wrote
0.1 (2015-11-09)
initial
V1_LWT.BLOCK
supportcaches metadata for performance
CLI tool for manipulating images
supports the
seek_mapped
seek_unmapped
interface for iterating over sparse regions