Sparse files contain large regions containing only zero bytes. Checking
that a blob only contains zeros is possible with over 100GB/s for modern
x86 CPUs. Calculating sha256 hashes is only possible with 500MB/s (or
2GB/s using hardware acceleration). Thus we can speed up the hash
calculation for all zero blobs (which always have length
chunker.MinSize) by checking for zero bytes and then using the
precomputed hash.
The all zeros check is only performed for blobs with the minimal chunk
size, and thus should add no overhead most of the time. For chunks which
are not all zero but have the minimal chunks size, the overhead will be
below 2% based on the above performance numbers.
This allows reading sparse sections of files as fast as the kernel can
return data to us. On my system using BTRFS this resulted in about
4GB/s.
Sending data through a channel at very high frequency is extremely
inefficient. Thus use simple callbacks instead of channels.
> name old time/op new time/op delta
> MasterIndexEach-16 6.68s ±24% 0.96s ± 2% -85.64% (p=0.008 n=5+5)
Ignored packs were reported as an empty pack by EachByPack. The most
immediate effect of this is that the progress bar for rebuilding the
index reports processing more packs than actually exist.
Previously the buffer was grown incrementally inside `repo.LoadUnpacked`.
But we can do better as we already know how large the index will be.
Allocate a bit more memory to increase the chance that the buffer can be
reused in the future.
For large pack sizes we might be only interested in the first and last
blob of a pack file. Thus stream a pack file in multiple parts if the
gaps between requested blobs grow too large.
Use runtime.GOMAXPROCS(0) as worker count for CPU-bound tasks,
repo.Connections() for IO-bound task and a combination if a task can be
both. Streaming packs is treated as IO-bound as adding more worker
cannot provide a speedup.
Typical IO-bound tasks are download / uploading / deleting files.
Decoding / Encoding / Verifying are usually CPU-bound. Several tasks are
a combination of both, e.g. for combined download and decode functions.
In the latter case add both limits together. As the backends have their
own concurrency limits restic still won't download more than
repo.Connections() files in parallel, but the additional workers can
decode already downloaded data in parallel.
Use only a single not completed pack file to keep the number of open and
active pack files low. The main change here is to defer hashing the pack
file to the upload step. This prevents the pack assembly step to become
a bottleneck as the only task is now to write data to the temporary pack
file.
The tests are cleaned up to no longer reimplement packer manager
functions.
Previously, SaveAndEncrypt would assemble blobs into packs and either
return immediately if the pack is not yet full or upload the pack file
otherwise. The upload will block the current goroutine until it
finishes.
Now, the upload is done using separate goroutines. This requires changes
to the error handling. As uploads are no longer tied to a SaveAndEncrypt
call, failed uploads are signaled using an errgroup.
To count the uploaded amount of data, the pack header overhead is no
longer returned by `packer.Finalize` but rather by
`packer.HeaderOverhead`. This helper method is necessary to continue
returning the pack header overhead directly to the responsible call to
`repository.SaveBlob`. Without the method this would not be possible,
as packs are finalized asynchronously.
As MergeFinalIndex and index uploads can occur concurrently, it is
necessary for MergeFinalIndex to check whether the IDs for an index were
already set before merging it. Otherwise, we'd loose the ID of an index
which is set _after_ uploading it.
github.com/pkg/errors is no longer getting updates, because Go 1.13
went with the more flexible errors.{As,Is} function. Use those instead:
errors from pkg/errors already support the Unwrap interface used by 1.13
error handling. Also:
* check for io.EOF with a straight ==. That value should not be wrapped,
and the chunker (whose error is checked in the cases changed) does not
wrap it.
* Give custom Error methods pointer receivers, so there's no ambiguity
when type-switching since the value type will no longer implement error.
* Make restic.ErrAlreadyLocked private, and rename it to
alreadyLockedError to match the stdlib convention that error type
names end in Error.
* Same with rest.ErrIsNotExist => rest.notExistError.
* Make s3.Backend.IsAccessDenied a private function.
When given a buf that is big enough for a compressed blob but not its
decompressed contents, the copy at the end of LoadBlob would skip the
last part of the contents.
Fixes #3783.
fd05037e1a changed the allocation batch
size from 256 to 128 under the assumption that an indexEntry is 60 bytes
on amd64, but it's 64: structs are padded out to a multiple of 8 for
alignment reasons. That means we'd waste no space in malloc even without
the batch allocation, at least on 64-bit machines. While that strategy
cuts the overallocation down dramatically for many small indexes, it also
seems to slow allocation down (Go 1.18, Linux, amd64, -benchtime=2s):
name old time/op new time/op delta
DecodeIndex-8 4.67s ± 5% 4.60s ± 1% ~ (p=0.953 n=10+5)
DecodeIndexParallel-8 4.67s ± 3% 4.60s ± 1% ~ (p=0.953 n=10+5)
IndexHasUnknown-8 37.8ns ± 8% 36.5ns ±14% ~ (p=0.841 n=5+5)
IndexHasKnown-8 38.5ns ±12% 37.7ns ±10% ~ (p=0.968 n=5+5)
IndexAlloc-8 615ms ±18% 607ms ± 1% ~ (p=1.000 n=10+5)
IndexAllocParallel-8 245ms ±11% 285ms ± 6% +16.40% (p=0.001 n=10+5)
MasterIndexAlloc-8 286ms ± 9% 275ms ± 2% ~ (p=1.000 n=10+5)
LoadIndex/v1-8 27.0ms ± 4% 26.8ms ± 1% ~ (p=0.690 n=5+5)
LoadIndex/v2-8 22.4ms ± 1% 22.8ms ± 2% +1.48% (p=0.016 n=5+5)
name old alloc/op new alloc/op delta
IndexAlloc-8 446MB ± 0% 446MB ± 0% -0.00% (p=0.000 n=8+4)
IndexAllocParallel-8 446MB ± 0% 446MB ± 0% -0.00% (p=0.008 n=8+5)
MasterIndexAlloc-8 213MB ± 0% 159MB ± 0% -25.47% (p=0.000 n=10+5)
name old allocs/op new allocs/op delta
IndexAlloc-8 913k ± 0% 2632k ± 0% +188.19% (p=0.008 n=5+5)
IndexAllocParallel-8 913k ± 0% 2632k ± 0% +188.21% (p=0.008 n=5+5)
MasterIndexAlloc-8 318k ± 0% 1172k ± 0% +267.86% (p=0.008 n=5+5)
Instead, this patch sets a batch size of 4, which means no space is
wasted by malloc on 64-bit and very little on 32-bit. It still gets very
close to the savings from not allocating in batches, without requiring
special code for bits.UintSize==64. Benchmark results, again for
Linux/amd64:
name old time/op new time/op delta
DecodeIndex-8 4.67s ± 5% 4.83s ± 9% ~ (p=0.315 n=10+10)
DecodeIndexParallel-8 4.67s ± 3% 4.68s ± 4% ~ (p=0.315 n=10+10)
IndexHasUnknown-8 37.8ns ± 8% 44.5ns ±19% ~ (p=0.095 n=5+5)
IndexHasKnown-8 38.5ns ±12% 36.9ns ± 8% ~ (p=0.690 n=5+5)
IndexAlloc-8 615ms ±18% 628ms ±18% ~ (p=0.218 n=10+10)
IndexAllocParallel-8 245ms ±11% 262ms ± 9% +7.02% (p=0.043 n=10+10)
MasterIndexAlloc-8 286ms ± 9% 287ms ±13% ~ (p=1.000 n=10+10)
LoadIndex/v1-8 27.0ms ± 4% 26.8ms ± 0% ~ (p=1.000 n=5+5)
LoadIndex/v2-8 22.4ms ± 1% 22.5ms ± 0% ~ (p=0.056 n=5+5)
name old alloc/op new alloc/op delta
IndexAlloc-8 446MB ± 0% 446MB ± 0% ~ (p=1.000 n=8+10)
IndexAllocParallel-8 446MB ± 0% 446MB ± 0% -0.00% (p=0.000 n=8+8)
MasterIndexAlloc-8 213MB ± 0% 160MB ± 0% -25.02% (p=0.000 n=10+9)
name old allocs/op new allocs/op delta
IndexAlloc-8 913k ± 0% 1333k ± 0% +45.94% (p=0.000 n=8+10)
IndexAllocParallel-8 913k ± 0% 1333k ± 0% +45.94% (p=0.000 n=8+8)
MasterIndexAlloc-8 318k ± 0% 525k ± 0% +64.99% (p=0.000 n=10+10)
The allocation method indexmap.newEntry has also been rewritten in a
form that is a few instructions shorter.
This removes RunWorkers, which had become mere overhead by successive
refactors. It also ensures that each former user of that function
returns any context error that occurs, so failure to complete an
operation is always reported as an error.
Apparently it can take a moment between closing a tempfile marked as
DELETE_ON_CLOSE and it actually being deleted. During that time the file
is inaccessible. Thus just skip deleting the temp file on windows.
A compressed index is only about one third the size of an uncompressed
one. Thus increase the number of entries in an index to avoid cluttering
the repository with small indexes.
The config file is not compressed as it should remain readable by older
restic versions such that these can return a proper error.
As the old format for unpacked data does not include a version header,
make use of a trick: The old data is always encoded as JSON. Thus it can
only start with '{' or '['. For any other value the first byte indicates
a versioned format. The version is set to 2 for now. Then the zstd
compressed data follows.
As repack streams packs these occupy one backend connection. Uploading a
new pack also requires a backend connection. To prevent a deadlock
during repack when reaching the backend connections limit, simply limit
the repackWorker count to always leave one connection for uploading.
The repack operation copies all selected blobs from a set of pack files
into new pack files. For prune the source and destination repositories
are identical. To implement copy, just use a different source and
destination repository.
The function supports efficiently loading a specified list of blobs from
a single pack in a streaming fashion. That is there's no need for
temporary files independent of the pack size.
This enables the backends to request the calculation of a
backend-specific hash. For the currently supported backends this will
always be MD5. The hash calculation happens as early as possible, for
pack files this is during assembly of the pack file. That way the hash
would even capture corruptions of the temporary pack file on disk.
This can be used to check how large a backup is or validate exclusions.
It does not actually write any data to the underlying backend. This is
implemented as a simple overlay backend that accepts writes without
forwarding them, passes through reads, and generally does the minimal
necessary to pretend that progress is actually happening.
Fixes #1542
Example usage:
$ restic -vv --dry-run . | grep add
new /changelog/unreleased/issue-1542, saved in 0.000s (350 B added)
modified /cmd/restic/cmd_backup.go, saved in 0.000s (16.543 KiB added)
modified /cmd/restic/global.go, saved in 0.000s (0 B added)
new /internal/backend/dry/dry_backend_test.go, saved in 0.000s (3.866 KiB added)
new /internal/backend/dry/dry_backend.go, saved in 0.000s (3.744 KiB added)
modified /internal/backend/test/tests.go, saved in 0.000s (0 B added)
modified /internal/repository/repository.go, saved in 0.000s (20.707 KiB added)
modified /internal/ui/backup.go, saved in 0.000s (9.110 KiB added)
modified /internal/ui/jsonstatus/status.go, saved in 0.001s (11.055 KiB added)
modified /restic, saved in 0.131s (25.542 MiB added)
Would add to the repo: 25.892 MiB
The io.Reader interface does not support contexts, such that it is
necessary to embed the context into the backendReaderAt struct. This has
the problem that a reader might suddenly stop working when it's
contained context is canceled. However, this is now problem here as the
reader instances never escape the calling function.
This is no change in behavior as a canceled context did later on cause
the config file creation to fail. Therefore this change just lets the
repository initialization fail a bit earlier.
This allows creating multiple repositories with identical chunker
parameters which is required for working deduplication when copying
snapshots between different repositories.
The slicing operator `slice[low:high]` default to 0 for the lower bound and
len(slice) for the upper bound when either or both are not specified.
Fix the code to use `cap(slice)` to check for the slice capacity.
If a blob in a pack file can be decrypted successfully but contains data
that results in a different hash than stated in the header pack, then
abort repacking. As both the pack header and the blob are
cryptographically verified this either means than a malicious entity
tampered with the backup or indicates hardware problems on the client.
prune should fail with an error in both cases.
The test now uses the fact that the sort is stable. It's not guaranteed
to be, but the test is cleaner and more exhaustive. sortCachedPacksFirst
no longer needs a return value.
The benchmark was actually testing the speed of index lookups.
name old time/op new time/op delta
SaveAndEncrypt-8 101ns ± 2% 31505824ns ± 1% +31311591.31% (p=0.000 n=10+10)
name old speed new speed delta
SaveAndEncrypt-8 41.7TB/s ± 2% 0.0TB/s ± 1% -100.00% (p=0.000 n=10+10)
name old alloc/op new alloc/op delta
SaveAndEncrypt-8 1.00B ± 0% 20989508.40B ± 0% +2098950740.00% (p=0.000 n=10+10)
name old allocs/op new allocs/op delta
SaveAndEncrypt-8 0.00 123.00 ± 0% +Inf% (p=0.000 n=10+9)
(The actual speed is ca. 131MiB/s.)
A side remark to the definition of Index.blob:
Another possibility would have been to use:
blob map[restic.BlobHandle]*indexEntry
This would have led to the following sizes:
key: 32 + 1 = 33 bytes
value: 8 bytes
indexEntry: 8 + 4 + 4 = 16 bytes
each packID: 32 bytes
To save N index entries, we would therefore have needed:
N * OF * (33 + 8) bytes + N * 16 + N * 32 bytes / BP = N * 82 bytes
More precicely, using a pointer instead of a direct entry is the better memory choice if:
OF * 8 bytes + entrysize < OF * entrysize <=> entrysize > 8 bytes * OF/(OF-1)
Under the assumption of OF=1.5, this means using pointers would have been the better choice
if sizeof(indexEntry) > 24 bytes.
- The SaveBlob method now checks for duplicates.
- Moves handling of pending blobs to MasterIndex.
-> also cleans up pending index entries when they are saved in the index
-> when using SaveBlob no need to care about index any longer
- Always check for full index and save it when storing packs.
-> removes the need of an index uploader
-> also removes the verbose "uploaded intermediate index" messages
- The Flush method now also saves the index
- Fix race condition when checking and saving full/non-finalized indexes
The username and hostname for new keys can be specified with the new
--user and --host flags, respectively. The flags are used only by the
`key add` command and are otherwise ignored.
This allows adding keys with for a desired user and host without having
to run restic as that particular user on that particular host, making
automated key management easier.
Co-authored-by: James TD Smith <ahktenzero@mohorovi.cc>
When loading a blob, restic first looks up pack files containing the
blob. To avoid unnecessary work an already cached pack file is preferred.
However, if there is only a single pack file to choose from (which is
the normal case) sorting the one-element list won't change anything.
Therefore avoid the unnecessary cache check in that case.
The previous benchmark spent much of its time allocating RNGs and
generating too many random numbers. It now spends 90% of its time
hashing and half of the rest writing to files.
name old time/op new time/op delta
PackerManager-8 319ms ± 1% 247ms ± 1% -22.48% (p=0.000 n=20+18)
name old speed new speed delta
PackerManager-8 143MB/s ± 1% 213MB/s ± 1% +48.63% (p=0.000 n=10+18)
name old alloc/op new alloc/op delta
PackerManager-8 635kB ± 0% 92kB ± 0% -85.48% (p=0.000 n=10+19)
name old allocs/op new allocs/op delta
PackerManager-8 1.64k ± 0% 1.43k ± 0% -12.76% (p=0.000 n=10+20)
