* blob: changed default shards from {3,3} to {1,3}
Turns out for very large repository around 100TB (5M blobs),
we end up creating max ~16M directories which is way too much
and slows down listing. Currently each leaf directory only has a handful
of files.
Simple sharding of {3} should work much better and will end up creating
directories with meaningful shard sizes - 12 K files per directory
should not be too slow and will reduce the overhead of listing by
4096 times.
The change is done in a backwards-compatible way and will respect
custom sharding (.shards) file written by previous 0.9 builds
as well as older repositories that don't have the .shards file (which
we assume to be {3,3}).
* fixed compat tests
The dual time measurement is described in
https://go.googlesource.com/proposal/+/master/design/12914-monotonic.md
The fix is to discard hidden monotonic time component of time.Time
by converting to unix time and back.
Reviewed usage of clock.Now() and replaced with timetrack.StartTimer()
when measuring time.
The problem in #1402 was that passage of time was measured using
the monotonic time and not wall clock time. When the computer goes
to sleep, monotonic time is still monotonic while wall clock time makes
a leap when the computer wakes up. This is the behavior that
epoch manager (and most other compontents in Kopia) rely upon.
Fixes#1402
Co-authored-by: Julio Lopez <julio+gh@kasten.io>
* refactor: move from io/ioutil to io and os package
The io/ioutil package has been deprecated as of Go 1.16, see
https://golang.org/doc/go1.16#ioutil. This commit replaces the existing
io/ioutil functions with their new definitions in io and os packages.
Signed-off-by: Eng Zer Jun <engzerjun@gmail.com>
* chore: remove //nolint:gosec for os.ReadFile
At the time of this commit, the G304 rule of gosec does not include the
`os.ReadFile` function. We remove `//nolint:gosec` temporarily until
https://github.com/securego/gosec/pull/706 is merged.
Signed-off-by: Eng Zer Jun <engzerjun@gmail.com>
* cli: added 'repository validate-provider' which runs a set of tests against blob storage provider to validate it
This implements a provider tests which exercises subtle behaviors which are not always correctly implemented by providers claiming compatibility with S3, for example.
The test checks:
- not found behavior
- prefix scans
- timestamps
- write atomicity
* retry: improved error message on failure
* rclone: fixed stats reporting and awaiting for completion
* webdav: prevent panic when attempting to mkdir with empty name
* testing: run providervalidation.ValidateProvider as part of regular provider tests
* cli: print a recommendation to validate provider after repository creation
* testing: removed testutil.Retry because all providers now have internal retries
* testing: simplified and unified cleanup for all cloud providers using shared buckets
* added framework for unit testing against remote real rclone remotes,
added google drive backend
* added parallelism to blobtesting which revealed some races during
PutBlob with WebDAV.
* content: fixed time-based auto-flush behavior to behave like Flush()
Previously it would sometimes be possible for a content whose write
started before time-based flush to finish writing afterwards (and it
would be included in the new index).
Refactored the code so that time-based flush happens before WriteContent
write and behaves exactly the same was as real Flush() so all writes
started before it will be awaited during the flush.
Also previous regression test was incorrect since it was mocking the
wrong blob method.
* content: refactored index blob manager crypto to separate file
This will be reused for encrypting session info.
* content: added support for session markers
Session marker (`s` blob) is written BEFORE the first data blob
(`p` or `q`) that belongs to new index segment (`n` is written).
Session marker is removed AFTER the index blob (`n`) has been written.
All pack and index blobs belonging to a session will have the session
ID as its suffix, so that if a reader can see `s<sessionID>` blob, they
will ignore any `p` and `q` blobs with the same suffix.
* maintenance: ignore blobs belonging to active sessions when running blob garbage collection
* cli: added 'sessions list' for listing active sessions
* content: added retrying writing previously failed blobs before writing new one
* linter: upgraded to 1.33, disabled some linters
* lint: fixed 'errorlint' errors
This ensures that all error comparisons use errors.Is() or errors.As().
We will be wrapping more errors going forward so it's important that
error checks are not strict everywhere.
Verified that there are no exceptions for errorlint linter which
guarantees that.
* lint: fixed or suppressed wrapcheck errors
* lint: nolintlint and misc cleanups
Co-authored-by: Julio López <julio+gh@kasten.io>
Globally replaced all use of time with internal 'clock' package
which provides indirection to time.Now()
Added support for faking clock in Kopia via KOPIA_FAKE_CLOCK_ENDPOINT
logfile: squelch annoying log message
testenv: added faketimeserver which serves time over HTTP
testing: added endurance test which tests kopia over long time scale
This creates kopia repository and simulates usage of Kopia over multiple
months (using accelerated fake time) to trigger effects that are only
visible after long time passage (maintenance, compactions, expirations).
The test is not used part of any test suite yet but will run in
post-submit mode only, preferably 24/7.
testing: refactored internal/clock to only support injection when
'testing' build tag is present
* blob: added DisplayName() method to blob.Storage
* cli: added 'kopia repo sync-to <provider>' which replicates BLOBs
Usage demo: https://asciinema.org/a/352299Fixes#509
* implemented suggestion by Ciantic to fail sync if the destination repository is not compatible with the source
* cli: added 'kopia repo sync --must-exist'
This ensures that target repository is not empty, otherwise syncing to
an accidentally unmounted filesystem directory might copy everything
again.
* content: added support for cache of own writes
Thi keeps track of which blobs (n and m) have been written by the
local repository client, so that even if the storage listing
is eventually consistent (as in S3), we get somewhat sane behavior.
Note that this is still assumming read-after-create semantics, which
S3 also guarantees, otherwise it's very hard to do anything useful.
* compaction: support for compaction logs
Instead of compaction immediately deleting source index blobs, we now
write log entries (with `m` prefix) which are merged on reads
and applied only if the blob list includes all inputs and outputs, in
which case the inputs are discarded since they are known to have been
superseded by the outputs.
This addresses eventual consistency issues in stores such as S3,
which don't guarantee list-after-put or list-after-delete. With such
stores the repository is ultimately eventually consistent and there's
not much that can be done about it, unless we use second strongly
consistent storage (such as GCS) for the index only.
* content: updated list cache to cache both `n` and `m`
* repo: fixed cache clear on windows
Clearing cache requires closing repository first, as Windows is holding
the files locked.
This requires ability to close the repository twice.
* content: refactored index blob management into indexBlobManager
* testing: fixed blobtesting.Map storage to allow overwrites
* blob: added debug output String() to blob.Metadata
* testing: added indexBlobManager stress test
This works by using N parallel "actors", each repeatedly performing
operations on indexBlobManagers all sharing single eventually consistent
storage.
Each actor runs in a loop and randomly selects between:
- *reading* all contents in indexes and verifying that it includes
all contents written by the actor so far and that contents are
correctly marked as deleted
- *creating* new contents
- *deleting* one of previously-created contents (by the same actor)
- *compacting* all index files into one
The test runs on accelerated time (every read of time moves it by 0.1
seconds) and simulates several hours of running.
In case of a failure, the log should provide enough debugging
information to trace the exact sequence of events leading up to the
failure - each log line is prefixed with actorID and all storage
access is logged.
* makefile: increase test timeout
* content: fixed index blob manager race
The race is where if we delete compaction log too early, it may lead to
previously deleted contents becoming temporarily live again to an
outside observer.
Added test case that reproduces the issue, verified that it fails
without the fix and passed with one.
* testing: improvements to TestIndexBlobManagerStress test
- better logging to be able to trace the root cause in case of a failure
- prevented concurrent compaction which is unsafe:
The sequence:
1. A creates contentA1 in INDEX-1
2. B creates contentB1 in INDEX-2
3. A deletes contentA1 in INDEX-3
4. B does compaction, but is not seeing INDEX-3 (due to EC or simply
because B started read before #3 completed), so it writes
INDEX-4==merge(INDEX-1,INDEX-2)
* INDEX-4 has contentA1 as active
5. A does compaction but it's not seeing INDEX-4 yet (due to EC
or because read started before #4), so it drops contentA1, writes
INDEX-5=merge(INDEX-1,INDEX-2,INDEX-3)
* INDEX-5 does not have contentA1
7. C sees INDEX-5 and INDEX-5 and merge(INDEX-4,INDEX-5)
contains contentA1 which is wrong, because A has been deleted
(and there's no record of it anywhere in the system)
* content: when building pack index ensure index bytes are different each time by adding 32 random bytes
This is done by introducing N unsynchronized caches, which simulate
what frontend of a cloud storage system might do, that causes eventual
consistency behavior.
, where blob.Storage.PutBlob gets a list of slices and writes them sequentially
* performance: added gather.Bytes and gather.WriteBuffer
They are similar to bytes.Buffer but instead of managing a single
byte slice, they maintain a list of slices that and when they run out of
space they allocate new fixed-size slice from a free list.
This helps keep memory allocations completely under control regardless
of the size of data written.
* switch from byte slices and bytes.Buffer to gather.Bytes.
This is mostly mechanical, the only cases where it's not involve blob
storage providers, where we leverage the fact that we don't need to
ever concatenate the slices into one and instead we can do gather
writes.
* PR feedback
This is mostly mechanical and changes how loggers are instantiated.
Logger is now associated with a context, passed around all methods,
(most methods had ctx, but had to add it in a few missing places).
By default Kopia does not produce any logs, but it can be overridden,
either locally for a nested context, by calling
ctx = logging.WithLogger(ctx, newLoggerFunc)
To override logs globally, call logging.SetDefaultLogger(newLoggerFunc)
This refactoring allowed removing dependency from Kopia repo
and go-logging library (the CLI still uses it, though).
It is now also possible to have all test methods emit logs using
t.Logf() so that they show up in failure reports, which should make
debugging of test failures suck less.
fixed some issues that prevented go test from passing on Windows:
- webdav client used \ instead of /
- need retries around mmap.Open()
- paths are prefixed with C:\ on windows
- time.Now() does not always move forward on Windows
Previously, it was possible for Flush() to miss in-flight writes,
but only when using repository manually since Uploader guarantees
there are no in-flight writes when it completes.
With this change Flush() will guarantee that any pending writes
completed before Flush() has started are guaranteed to be committed
to the repository before Flush() returns.
This was actually a regression introduced in #105.
Added regression test to prevent it from reoccurring.
Also introduced strongly typed content.ID and manifest.ID (instead of string)
This aligns identifiers across all layers of repository:
blob.ID
content.ID
object.ID
manifest.ID
This updates the terminology everywhere - blocks become blobs and
`storage.Storage` becomes `blob.Storage`.
Also introduced blob.ID which is a specialized string type, that's
different from CABS block ID.
Also renamed CLI subcommands from `kopia storage` to `kopia blob`.
While at it introduced `block.ErrBlockNotFound` and
`object.ErrObjectNotFound` that do not leak from lower layers.
