summaryrefslogtreecommitdiff
Commit message (Collapse)AuthorAgeFilesLines
* Cut release 1.1.4.1.1.4Michael Cahill2012-04-165-6/+36
|
* Move the test/format threads setting into the CONFIG file.Michael Cahill2012-04-165-17/+21
|
* All dump configuration should be valid for WT_SESSION->create.Michael Cahill2012-04-164-33/+40
| | | | closes #194.
* Split WT_PAGE->flags so that there is no possibility of racing:Michael Cahill2012-04-167-96/+109
| | | | | | | 1) Move WT_PAGE_REC_* flags into WT_PAGE_MODIFY; 2) Use the WT_REF_EVICT_WALK state for all pages in the LRU queue and get rid of the WT_PAGE_EVICT_LRU flag. The only remaining flag is WT_PAGE_BUILD_KEYS, so there is no possibility of a race.
* Clear all unused eviction queue entries while holding the lru_lock.Michael Cahill2012-04-162-5/+10
|
* Fix an unlikely bug where the EVICT_LRU flag was cleared when a page in the ↵Michael Cahill2012-04-161-3/+3
| | | | LRU queue was overwritten with itself during a walk. This led to an assertion failure when the page was later evicted.
* whitespaceMichael Cahill2012-04-161-1/+1
|
* Don't worry about session.create for the table URI, only the individual fileKeith Bostic2012-04-101-10/+12
| | | | information, ref #194.
* lint (unsigned int to int)Keith Bostic2012-04-101-1/+1
|
* The dump utility should only output configuration that can be passed toKeith Bostic2012-04-103-14/+53
| | | | WT_SESSION::create. Close #194.
* Ref #193, replacing 98be8332986a663b9b2a441cbc9da68dfa974d4b andKeith Bostic2012-04-102-23/+20
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | b504394466dfe7e1d4b458765f655e5c93282a6d. The last two commits for issue #193 broke the tree: the initial problem was that in some cases we failed to migrate a leaf page's dirty-ness up the tree. A leaf page would be reconciled and mark its parent, a split-merge page, dirty, but the dirty flag would climb no further up the tree. That's because the way a dirty flag moves up the tree is that reconciling a dirty page cleans the page and marks the parent dirty. Since split-merge pages are never reconciled, the dirty flag would never climb the tree. The fix I initially put in was to change split-merge page reconciliation to not simply return, instead, split-merge page reconciliation dirtied its parent page. That was OK, but then I simplified eviction to reconcile any page it found dirty rather than skipping split-merge pages. That broke the world: the problem is we never mark split-merge pages clean, so a completely clean tree that was being evicted could re-dirty a page because the split-merge page was never marked cleaned and so would be reconciled as part of eviction, marking its clean parent dirty. I can think of two fixes. Fix 1, change split-merge page reconciliation to have two effects: mark the split-merge page clean, and dirty its parent page. Fix 2, don't ever mark a split-merge page dirty during reconciliation, climb the tree until we find a non-split-merge page and mark it dirty instead. Fix 1 makes split-merge pages behave more like other pages in the tree, but means more work, we have to reconcile split-merge pages for no other reason than to make the dirty flag climb the tree, that is, our only purpose in reconciling split-merge pages is to mark them clean and their parent pages dirty. This commit implements fix 2, reconciliation doesn't mark split-merge pages dirty, it climbs the tree until it finds a real page and marks it dirty.
* Reconcile dirty split-merge pages as part of doing eviction.Keith Bostic2012-04-101-5/+6
| | | | | | | | | Reconciliation of split-merge pages does have an effect, it marks the split-merge's parent page dirty. I can't think of a code path where we'd be looking at a split-merge page during eviction, where the split-merge page's parent isn't already marked dirty, but eviction is not that common, split-merge pages are less common, and so the cost of consistency is cheap. Ref #193.
* Merge branch 'develop' of github.com:wiredtiger/wiredtiger into developKeith Bostic2012-04-1018-143/+223
|\
| * Change test of "wt list" now that there is always a schema file.Michael Cahill2012-04-101-1/+1
| | | | | | | | refs #191
| * Create the schema table as part of creating the environment so that ↵Michael Cahill2012-04-103-4/+19
| | | | | | | | | | | | | | | | application threads don't race trying to create it later. This is not a complete solution: threads may still race creating other files or schema objects, but this fixes the most common "thundering herd" case. refs #191
| * Relax the version requirements for autoconf and libtool: we are testing with ↵Michael Cahill2012-04-041-2/+2
| | | | | | | | 2.63 and 2.2.6, respectively.
| * Check the versions of autoconf, automake and libtool to avoid failures when ↵Michael Cahill2012-04-042-1/+7
| | | | | | | | trying to build from the github tree.
| * Merge branch 'master' into developMichael Cahill2012-04-041-0/+1
| |\
| | * Added tag 1.1.3 for changeset a792d468beddMichael Cahill2012-04-041-0/+1
| | |
| * | Merge branch 'master' into developMichael Cahill2012-04-040-0/+0
| |\ \ | | |/
| | * Merge branches 'develop' and 'master' of github.com:wiredtiger/wiredtiger1.1.3Michael Cahill2012-04-040-0/+0
| | |
| * | Cut release 1.1.3.Michael Cahill2012-04-046-7/+33
| | |
| * | Merge pull request #188 from wiredtiger/feature/eviction-fixesMichael Cahill2012-04-036-127/+152
| |\ \ | | |/ | |/| Merge eviction-fixes into develop.
| | * Merge branch 'develop' into feature/eviction-fixesMichael Cahill2012-04-033-5/+14
| | |\ | | |/ | |/|
| * | Don't report range errors for config values that aren't well-formed integers.Michael Cahill2012-04-031-4/+11
| | |
| | * During an eviction walk, pin pages up to the root. Use the EVICT_LRUMichael Cahill2012-04-033-98/+78
| | | | | | | | | | | | page flag to avoid putting a page on the LRU queue multiple times.
| | * Reduce the impact of clearing a page from the LRU queue by marking pages on ↵Michael Cahill2012-03-303-5/+17
| | | | | | | | | | | | the queue with a flag.
| | * When evicting a page with children, remove the children from the LRU ↵Michael Cahill2012-03-303-27/+60
| | | | | | | | | | | | eviction queue.
* | | split-merge pages have to be reconciled to mark their parents dirtyKeith Bostic2012-04-101-3/+6
|/ / | | | | | | close #193.
* | Remove the release tree after creating the tarball.Michael Cahill2012-03-301-0/+2
| |
* | s_types should check in more directories than just src.Michael Cahill2012-03-301-1/+1
|/
* When we removed test/config.i, it broke s_types -- simplify s_types,Keith Bostic2012-03-281-5/+1
| | | | just do a find on src/.
* We no longer use __wt_buf_swap(), remove it (it's trivial to replaceKeith Bostic2012-03-282-15/+0
| | | | if it comes back).
* move CURDUMP_PASS into the WiredTiger name space, don't complain thatKeith Bostic2012-03-282-8/+9
| | | | it isn't used.
* Remove test/config*, it's obsolete now.Michael Cahill2012-03-283-97/+0
|
* Layer dump cursors on top of any cursor type.Michael Cahill2012-03-1113-135/+232
| | | | | --HG-- extra : rebase_source : a0ddaa6ba4d7ed5fdb7ca8811f2147e3ede0bad3
* Push handling of "standard" cursor config flags into __wt_cursor_init.Michael Cahill2012-03-114-73/+38
| | | | | --HG-- extra : rebase_source : dfa07973f66199329fede015f43ac813315bd1f8
* Remove the CURSTD_FILE and CURSTD_TABLE flags: if we need to look inside aMichael Cahill2012-03-1113-74/+51
| | | | | | | cursor beyond the public fields, we're breaking layering. --HG-- extra : rebase_source : 06d12a9724f81743a140a9875f816e857993e405
* Fix the "exclusive" config for WT_SESSION::create.Michael Cahill2012-03-287-43/+65
| | | | | | | | | | 1. Make it work for files within a single session. 2. Make it work for files across sessions. 3. Make other data sources consistent with files. closes #181 --HG-- extra : rebase_source : 29b07b914508b8fcef1a4e4c212d32e5f21dd50e
* Typo in usage method for "wt write".Michael Cahill2012-03-281-1/+1
| | | | | --HG-- extra : rebase_source : 32fbccc4b5c7c38b685ee00183f2e6f04e31c72c
* Add a couple more copyrights.Keith Bostic2012-03-263-1/+50
|
* update a comment, no real change.Keith Bostic2012-03-251-4/+4
|
* Don't force the delete percentage to 0 (so salvage runs), if theKeith Bostic2012-03-241-3/+12
| | | | configuration says differently.
* Add a section on replacing the default system memory allocator to theKeith Bostic2012-03-213-3/+16
| | | | | | | tuning page. An example setting the cache to 10MB isn't useful, crank the example to 500MB.
* Merge branch 'master' of github.com:wiredtiger/wiredtigerMichael Cahill2012-03-201-0/+2
|\
| * Don't complain about buffers that weren't discarded yet, it breaks theKeith Bostic2012-03-201-0/+2
| | | | | | | | test suite.
* | Cut release 1.1.2.Michael Cahill2012-03-205-12/+57
| | | | | | | | | | --HG-- extra : rebase_source : 2a3ef9165ed9001ebb2b013577652d831140c370
* | ../../../../ext/compressors/bzip2_compress/bzip2_compress.c:161: warning: ↵Michael Cahill2012-03-201-4/+4
|/ | | | | | | implicit conversion shortens 64-bit value into a 32-bit value --HG-- extra : rebase_source : 67f5d6fe20895f7f2891df98c1062b84a2db85da
* Remove core files in the clean target.1.1.2Keith Bostic2012-03-201-1/+1
|
* Complain if a scratch buffer is allocated but never discarded.Keith Bostic2012-03-201-2/+6
|
View Lorry Controller Status