summaryrefslogtreecommitdiff
Commit message (Collapse)AuthorAgeFilesLines
* Cut release 1.6.01.6.0Alex Gorrod2013-05-166-12/+48
|
* Merge pull request #542 from wiredtiger/name-hashMichael Cahill2013-05-156-5/+13
|\ | | | | Add name hash to avoid excessive strcmp on common path.
| * Use the name hash to avoid string comparisons in the session list of data ↵Michael Cahill2013-05-164-4/+7
| | | | | | | | handles as well as the shared connection list.
| * Add name hash to avoid excessive strcmp on common path.Susan LoVerso2013-05-152-1/+6
| |
* | Fix a broken link in README.Michael Cahill2013-05-162-4/+4
| | | | | | | | closes #541
* | Use a Berkeley DB home directory relative to the WiredTiger home.Keith Bostic2013-05-151-5/+17
| |
* | Restructure LSM and KVS data-sources; the top-level object is a "file:"Keith Bostic2013-05-153-10/+23
| | | | | | | | | | or a "table:", KVS devices are configured on create with "source:XXX", LSM is configured on create with "type:lsm".
* | Be more specific about LSM's requirement for non-zero-length data items,Keith Bostic2013-05-151-1/+2
|/ | | | the generic 'cursor not set' error message is misleading.
* Merge pull request #539 from wiredtiger/evict-clean-bugAlex Gorrod2013-05-148-57/+128
|\ | | | | Fix a bug where clean pages could be evicted when they contained updates required for a running transaction
| * Add a comment about tracking the largest transaction ID during reconciliation.Michael Cahill2013-05-151-0/+7
| |
| * typoMichael Cahill2013-05-151-10/+10
| |
| * Fix ordering of application eviction so a session's read-committed ↵Michael Cahill2013-05-152-25/+58
| | | | | | | | transaction can't block eviction (for forced eviction of big pages). Fix a race calculating the oldest snapshot ID required for eviction.
| * Include the current transaction when calculating the oldest snapshot ID.Michael Cahill2013-05-151-3/+3
| |
| * Fix a bug where clean pages could be evicted when they contained updates ↵Michael Cahill2013-05-157-37/+68
| | | | | | | | required for a running transaction.
* | Acquire the file-list spin lock when checking the WT_FH.refcnt duringKeith Bostic2013-05-141-19/+34
| | | | | | | | | | | | | | close, there's a race across the reference count test and decrement. Check for the name again before we link a duplicate file descriptor on the list of open files, it doesn't cost us anything to avoid the race.
* | fix a commentKeith Bostic2013-05-141-2/+2
|/
* Merge pull request #534 from wiredtiger/file-extendKeith Bostic2013-05-1419-40/+134
|\ | | | | file extension
| * Change file_extend syntax from file_extend=(type=[data,log],size=XXX)Keith Bostic2013-05-149-47/+39
| | | | | | | | | | | | to file_extend=(data=XXX,log=XXX). Add period testing of file_extend to test/format.
| * Merge branch 'develop' into file-extendKeith Bostic2013-05-146-27/+43
| |\ | |/ |/|
* | minor rephrasing so two patterns matchKeith Bostic2013-05-141-5/+4
| |
* | Only track whether eviction is stuck if the cache is full.Michael Cahill2013-05-141-3/+8
| |
* | Merge pull request #537 from wiredtiger/eviction-isolationAlex Gorrod2013-05-134-18/+30
|\ \ | | | | | | Fix an off-by-one error in the transaction used for eviction.
| * | Snapshots (including eviction) *should* include the last committed transaction.Michael Cahill2013-05-143-14/+15
| | |
| * | Fix an off-by-one error in the transaction used for eviction. We track the ↵Michael Cahill2013-05-142-4/+15
| | | | | | | | | | | | oldest ID that can *not* be written.
* | | Don't preload the file system cache when compression is enabled.Alex Gorrod2013-05-141-1/+1
|/ / | | | | | | | | Since we don't have easy access to the file offset for particular page when applications are using compression.
| * Merge branch 'develop' into file-extendKeith Bostic2013-05-131-1/+1
| |\ | |/ |/|
* | typoKeith Bostic2013-05-131-1/+1
| |
| * Merge branch 'develop' into file-extendKeith Bostic2013-05-132-1/+3
| |\ | |/ |/|
* | Mention dd explicitly, it has an option to configure O_DIRECT access.Keith Bostic2013-05-132-1/+3
| |
| * Merge branch 'develop' into file-extendKeith Bostic2013-05-131-5/+14
| |\ | |/ |/|
* | imported patch madvise-alignmentMichael Cahill2013-05-131-5/+14
| |
| * When unloading the live checkpoint, discard any extended blocks so filesKeith Bostic2013-05-131-0/+5
| | | | | | | | | | don't stay longer than they need to be. This means we truncate file on both "open" and "close", but that doesn't seem too awful.
| * Merge branch 'develop' into file-extendKeith Bostic2013-05-133-22/+24
| |\ | |/ |/|
* | whitespaceMichael Cahill2013-05-131-1/+0
| |
* | Merge pull request #536 from wiredtiger/lru-queue-clearagorrod2013-05-122-22/+14
|\ \ | | | | | | Clear pages from the LRU queue as soon as they are locked.
| * | Clear pages from the LRU queue as soon as they are locked. Threads getting ↵Michael Cahill2013-05-132-22/+14
|/ / | | | | | | pages to evict look inside the WT_PAGE structure before switching the WT_REF state, so it isn't safe to evict a page that is queued.
* | The underlying file format just changed, note it in the upgrading document.Keith Bostic2013-05-111-0/+11
| |
| * Support for a new configuration option, file_extend, that extends the fileKeith Bostic2013-05-1116-43/+140
|/ | | | in chunks instead of doing it a block at a time.
* Rename the WT_FH.file_size field to be WT_FH.size.Keith Bostic2013-05-1110-32/+33
|
* Shorten long lines to 80 column max.Keith Bostic2013-05-111-7/+7
|
* Add more warnings about direct I/O, put them all in one place.Keith Bostic2013-05-113-16/+28
|
* Increase the Linux default buffer alignment to 4KB, there are known casesKeith Bostic2013-05-114-8/+19
| | | | | | | where I/O to/from buffers with lesser alignment causes reads to go through the buffer cache, leading to multiple versions of a block in the system. Add asserts to the read/write calls to ensure we detect mis-aligned buffers.
* Files for which we only hvae fake checkpoints might be verified, the blockKeith Bostic2013-05-101-25/+19
| | | | manager has to check.
* Merge pull request #533 from wiredtiger/desc-allocsizeKeith Bostic2013-05-1011-78/+91
|\ | | | | Use the allocation size for the file "desc" block.
| * __wt_block_ckpt_init() shouldn't set the checkpoint's file size to aKeith Bostic2013-05-107-42/+39
| | | | | | | | | | | | | | | | | | | | | | newly created file size, that should be set by the caller. (There's only a couple places that aren't cracking an existing checkpoint block and would theoretically need to set it, and actually none of them need to set it, it's really only useful when it's taken out of an existing, written, checkpoint that we're subsequently reading.) Review everywhere "sector" is used, and change it to something less specific.
| * Use the allocation size for the file "desc" block. This is important for ↵Michael Cahill2013-05-0911-64/+80
|/ | | | direct I/O, so that reads and writes are aligned as expected.
* Merge pull request #530 from wiredtiger/terminate-methodMichael Cahill2013-05-0813-706/+1025
|\ | | | | Terminate method
| * Merge branch 'develop' into terminate-methodMichael Cahill2013-05-0911-82/+78
| |\ | |/ |/| | | | | Conflicts: src/conn/conn_api.c
* | Merge pull request #532 from wiredtiger/error-leakMichael Cahill2013-05-083-61/+58
|\ \ | | | | | | Call destroy functions on error. (#531)
| * | All cleanup happens in the top-level open call.Michael Cahill2013-05-091-8/+2
| | | | | | | | | | | | refs #531
View Lorry Controller Status