summaryrefslogtreecommitdiff
Commit message (Collapse)AuthorAgeFilesLines
* Enhance error handling in LevelDB API cursor implementation.Alex Gorrod2014-06-131-33/+92
|
* In LevelDB API remove asserts from functions that return values.Alex Gorrod2014-06-131-34/+78
|
* Don't search for LevelDB headers now they are in our tree.Michael Cahill2014-06-131-5/+0
|
* Standardize whitespace in LevelDB source code.Alex Gorrod2014-06-132-382/+382
| | | | Use two spaces for tabs to be consistent with LevelDB.
* Add in the a copy of the LevelDB header files, and use them in our LevelDB API.Alex Gorrod2014-06-1315-11/+1796
| | | | | | | This avoids us needing to maintain compatability with the numerous different versions of LevelDB in the wild. Header files will be installed to $include/wiredtiger/leveldb
* Merge branch 'develop' into leveldb-apiMichael Cahill2014-06-126-8/+98
|\
| * Merge pull request #1057 from wiredtiger/lsm-snapshot-update-check-onlyAlex Gorrod2014-06-124-4/+81
| |\ | | | | | | Support snapshots in LSM via update checks rather than actual updates.
| | * Merge pull request #1060 from wiredtiger/lsm-snapshot-update-check-only-2Michael Cahill2014-06-125-24/+78
| | |\ | | | | | | | | Lsm snapshot update check only, review changes.
| | | * Wrap update_check in a public cursor call, so that session->dhandle is set ↵Michael Cahill2014-06-124-5/+29
| | | | | | | | | | | | | | | | | | | | | | | | correctly. refs #1060
| | | * Don't use a cursor flag for update checks: override the insert method ↵Michael Cahill2014-06-124-30/+62
| | | | | | | | | | | | | | | | | | | | | | | | instead. This gets conflict checking out of the common fast path for updates, and avoids a new flag in the public API that we don't want to document. refs #1047, #1057, #1060
| | | * Trivial style changes: join a pair of lines, re-order a test so it looksKeith Bostic2014-06-111-4/+3
| | | | | | | | | | | | | | | | more like the test that precedes it.
| | | * Move the WT_CBT_CONFLICT_CHECK flag from the btree-specific cursorKeith Bostic2014-06-114-16/+15
| | |/ | | | | | | | | | | | | | | | | | | | | | structure into the WT_CURSOR structure as WT_CURSTD_CONFLICT_CHK. It's possible we could return from the row-search function with a match not found in the insert list, check for a NULL insert before indirecting through it.
| | * Enhance comment.Alex Gorrod2014-06-111-0/+4
| | |
| | * Support snapshots in LSM via update checks, rather than actual updates.Alex Gorrod2014-06-113-0/+19
| | | | | | | | | | | | | | | | | | | | | The current code inserts updates into old chunks, in order to check for conflicts with snapshot transactions. It's enough to do the conflict check without actually updating, and doing a check saves some mess with tracking when it's OK to checkpoint non-primary chunks.
| * | Add wtperf option to execute lots of checkpoints during populate phase.Alex Gorrod2014-06-122-4/+17
| | | | | | | | | | | | Useful for catching races in LSM switch.
* | | Merge branch 'develop' into leveldb-apiMichael Cahill2014-06-1263-355/+546
|\ \ \ | |/ /
| * | Fix a race in the populate phase of wtperf with multiple threads.Alex Gorrod2014-06-121-2/+2
| | |
| * | Allow merges to move past most recent maximum-sized chunk. #1053Susan LoVerso2014-06-111-1/+2
| | |
| * | If the string isn't configured at all, output an empty string.Keith Bostic2014-06-111-1/+2
| | |
| * | Reinstate a check after deepening a tree that the child pages are internal ↵Michael Cahill2014-06-111-0/+6
| | | | | | | | | | | | | | | | | | pages before descending into them. refs #1048, #1056.
| * | Merge pull request #1056 from wiredtiger/stall-eviction-1048Michael Cahill2014-06-114-18/+52
| |\ \ | | |/ | |/| Don't free created internal pages from deepen-split until it's safe.
| | * Check on transaction release if we can release any segments from theKeith Bostic2014-06-101-0/+7
| | | | | | | | | | | | | | | session's free-on-transaction generation list, we were accumulating them until connection close, which wasn't correct.
| | * When deepening the tree, don't free the newly created child pages until afterKeith Bostic2014-06-103-4/+42
| | | | | | | | | | | | | | | | | | | | | | | | | | | it's known that all threads have exited the original page's page index array. This fixes two problems: first, threads in the original page index array can see freed WT_REF structures if the newly created child pages are evicted too soon, and at the end of the function that deepens the tree (after making the deepen-the-tree split "real", we read through the newly created child pages without holding hazard references. Reference #1048.
| | * The parent page type must be an internal page, don't bother testing forKeith Bostic2014-06-101-6/+2
| | | | | | | | | | | | leaf pages, it's confusing.
| | * The parent page of a split must be an internal page by definition, don'tKeith Bostic2014-06-101-5/+1
| | | | | | | | | | | | bother testing for leaf pages, it's confusing.
| | * If we're in the split-deepen code, there's no need to check if we're lookingKeith Bostic2014-06-101-3/+0
| | | | | | | | | | | | at tree internal pages, we are, by definition.
| * | We haven't been consistently checking for configuration options thatKeith Bostic2014-06-102-64/+52
| |/ | | | | | | | | | | aren't applicable to the underlying storage types of the run (for example, bloom filter settings for non-LSM runs). Just remove those CONFIG file messages, it's simpler, more consistent and less work.
| * Alex notes that the current code will continue running if the pluggableKeith Bostic2014-06-091-4/+4
| | | | | | | | | | | | | | decompression fails, but if the returned decompressed length isn't what we expect, we'll panic. For now, change the code to always panic -- it would be simple enough to continue in this case, should we want to. Reference #1055.
| * Add explicit os_cache_dirty_max settings for LSM wtperf tests. #1051Susan LoVerso2014-06-0934-38/+43
| | | | | | | | Add code to ignore the setting if sync_file_range is not available.
| * Set the WT_REF.home; not currently used for anything, but a valid WT_REFKeith Bostic2014-06-091-1/+3
| | | | | | | | should never have a NULL home field.
| * Assert the parent's WT_REF.home is correct in all cases.Keith Bostic2014-06-091-1/+1
| |
| * Update a comment, sentence no longer needed.Keith Bostic2014-06-071-3/+0
| |
| * Make the "leak_memory" configuration explicit in the test/format config, off ↵Michael Cahill2014-06-064-2/+10
| | | | | | | | by default. Fix a leak with repeated setting of configuration values.
| * Fix an implicit sign conversion found by clang 3.5.Michael Cahill2014-06-061-3/+4
| |
| * Add custom key sorting function for times. #1052Susan LoVerso2014-06-051-1/+9
| |
| * Change test1-4 to max out with 2 billion items.Susan LoVerso2014-06-054-1/+1
| |
| * Merge pull request #1045 from wiredtiger/recover-missing-filessueloverso2014-06-051-13/+26
| |\ | | | | | | Fix a bug in recovery with missing files
| | * Fix a bug in recovery with missing files (e.g., after a hotbackup that raced ↵Michael Cahill2014-06-031-13/+26
| | | | | | | | | | | | | | | | | | with file creation). refs #1042
| * | Merge pull request #1051 from wiredtiger/lsm-latenciessueloverso2014-06-053-48/+51
| |\ \ | | | | | | | | Fix some LSM latency issues
| | * | Fix some LSM latency issues:Michael Cahill2014-06-053-48/+51
| | | | | | | | | | | | | | | | | | | | * have merge threads periodically check whether the tree needs to be switched in case the checkpoint thread is blocked; and * don't turn on os_cache_dirty_max by default for LSM.
| * | | I believe the clang complaint is because the skiphigh/skiplow values areKeith Bostic2014-06-051-34/+31
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | longer than the 0th key on the internal page, and when we try and skip over the initial bytes, we read past the end of the key. Skiphigh/skiplow are a lot more valuable in terms of performance than avoiding the branch instruction, give up: don't do the 0th key comparison inside the internal loop.
| * | | Instead of byte-copying the WT_IKEY structure and data, call the standardKeith Bostic2014-06-051-4/+5
| | | | | | | | | | | | | | | | | | | | | | | | | | | | underlying supporting function. Clang is complaining about the internal page binary search function going off the end of the allocated memory: I don't see a bug here, but there's no reason to create the WT_IKEY structure in a non-standard way, and maybe I'm just missing something.
| * | | Merge pull request #1050 from wiredtiger/row-store-internal-pageKeith Bostic2014-06-053-100/+101
| |\ \ \ | | |/ / | |/| | Row store internal page search optimizations.
| | * | Merge branch 'develop' into row-store-internal-pageKeith Bostic2014-06-051-2/+47
| | |\ \ | | |/ / | |/| |
| * | | Merge pull request #1049 from wiredtiger/json-docKeith Bostic2014-06-041-2/+47
| |\ \ \ | | | | | | | | | | Json doc
| | * | | Add the word "dump" to the section headings for clarity.Keith Bostic2014-06-041-2/+2
| | | | |
| | * | | Add more detail about our JSON output, along with a small sample. refs #740.Don Anderson2014-06-041-0/+45
| | | |/ | | |/|
| | | * Reverse change to use -1, 0 and 1 tests against collation returns,Keith Bostic2014-06-051-8/+8
| | | | | | | | | | | | | | | | | | | | debugging application collators that return a range of values is a pain.
| | | * Minor code shuffle to make the flow a little more obvious.Keith Bostic2014-06-041-2/+2
| | | |
| | | * Re-order the comparison cascading if/else clauses to put the leastKeith Bostic2014-06-041-19/+16
| | | | | | | | | | | | | | | | likely option (a match, where cmp == 0), last.
View Lorry Controller Status