This introduces an interface for the freelist, splits it into two concrete
implementations.
fixes etcd-io#773
Signed-off-by: Thomas Jungblut <tjungblu@redhat.com>
Benchmark with this change:
BenchmarkBucket_CreateBucketIfNotExists-10 123 9573035 ns/op 17930 B/op 37 allocs/op
Benchmark with old implementnation:
BenchmarkBucket_CreateBucketIfNotExists-10 121 10474415 ns/op 18147 B/op 46 allocs/op
Signed-off-by: Benjamin Wang <wachao@vmware.com>
This moves the error variables that had been moved to the
internal/common package during recent refactoring to a non-internal
errors package, once again allowing consumers to test for particular
error conditions.
To preserve API compatibility with bbolt v1.3, these error variables
are also redefined in the bbolt package, with deprecation notice to
migrate to bbolt/errors.
Signed-off-by: Josh Rickmar <jrick@zettaport.com>
So far the code was frequently traversing all the keys (ignoring flag whether key is a bucket)
and trying to open each of the keys as bucket (seeking the same entry from the scratch).
In this proposal, we iterate only through bucket keys.
Signed-off-by: Piotr Tabor <ptab@google.com>
The tests were failing when doing development on M1 mac. This was due to page-size difference (4KB assumed in tests) vs. the
OS defaults.
Signed-off-by: Piotr Tabor <ptab@google.com>
As of now the test fails and tries to delete the next key returned by
the cursor, which happens to be a nested bucket. That's why Delete is
fails.
Signed-off-by: Pavel Borzenkov <pavel.borzenkov@gmail.com>
Default/nil quick.Config uses 1000 rounds, causing TestBucker_Put_Single to
run for over 3 minutes in CI. The default count (via qconfig()) for boltdb
is 5, so use that.
This commit sets the capacity on slices returned from
`Bucket.Get()` to match the slice length. Previously
the capacity would be the size of the mmap max size.
This does not cause any backwards compatibility issues,
however, it does allow users to `append()` to the returned
slice since that will cause Go to realloc a new slice on the
heap.
Fixes#544
This commit refactors the test suite to make it cleaner and to use the
standard testing library better. The `assert()`, `equals()`, and `ok()`
functions have been removed and some test names have been changed for
clarity.
No functionality has been changed.
When removing all the keys from the bucket, use Cursor.Delete directly, which
avoids a seek through the bucket to find the element to delete. This knocks
1/5 of the time off my laptop when running this test while not reducing the
amount of overflow pages cleared out of the free list.
This commit ignores the TestBucket_Put_ValueTooLarge test on
Drone CI because the containers do not have enough memory. This
test will still be run by anyone else who runs the test suite so
that will still provide coverage.
This commit fixes a bug where only calling NextSequence() on a Bucket does not cause the Bucket to be
peristed. The simple fix is to simply materialize the root node so that the bucket is flushed out
during commit.
Thanks to Matthew Dawson (@MJDSys) for reporting.
https://github.com/boltdb/bolt/issues/296
This commit moves the DB.FillPercent field to Bucket.FillPercent. This
allows the fill percentage to be specified per-bucket, per-tx. This
value is not persisted and should be set whenever using it.
This commit fixes a bug where a root split on a very large insert would
cause an overflow of the root node. The problem was that the new root
was not split when it was created so a new root with more than 64K child
nodes would overflow the page.count (uint16).
This commit is a backwards compatible change that allows the freelist to overflow the
page.count (uint16). It works by checking if the overflow will occur and marking the
page.count as 0xFFFF and setting the actual count to the first element of the freelist.
This approach was used because it's backwards compatible and it doesn't make sense to
change the data type of all page counts when only the freelist's page can overflow.
Fixes#192.
This commit changes NextSequence() to return a uint64 instead of an int. This also
removes the ErrSequenceOverflow error condition since overflowing a uint64 is unlikely.
Fixes#39.
This commit changes Open() to provide an additional Options argument. The options
argument currently only has a Timeout which will cause the Open() to return
ErrTimeout if a file lock cannot be obtained in time.
Fixes#207.
This commit reverts merge-split and fixes the node.split() to do a multi-page split. This issue
caused problems with bulk loading because it would split into a small page and a very large page.
The very large page, in turn, would be an arbitrary size so when it was freed later it would be
difficult to reuse and would cause serious fragmentation issues.
This commit fixes a bug where deletions that caused merge-left rebalances were updating
the parent node which caused a node to "reappear" even after it had been deleted. This was
fixed in merge-right rebalances a while ago but merge-left is less frequent so it was
missed until now.
Many thanks to Jordan Sherer (@jsherer) for finding and reporting the bug.
This commit changes the node.split() functionality to check if the next node has
available space and, if so, it will merge the newly split keys into the next node.
Previously, keys could be continually put into the left side of a split causing that
first half to split off small right side nodes. This was especially problematic with
databases with a high fill percent.
This commit fixes the freelist so that it frees from the beginning of the data file
instead of the end. It also adds a fast path for pages which can be allocated from
the first free pages and it includes read transaction stats.
Thomas Jungblut
Benjamin Wang
Benjamin Wang
Ishan Tyagi
Marek Siarkowicz
Josh Rickmar
Kishen V
Piotr Tabor
wpedrak
Gyuho Lee
Gyu-Ho Lee
Anthony Romano
Pavel Borzenkov
Xiang
Ben Johnson
Matthew Dawson
Matt Layher
yosssi
Martin Kobetic