| Age | Commit message (Collapse) | Author | |
|---|---|---|---|
| 2014-03-28 | lint | Alex Gorrod | |
| Originally fixed as part of larger change: cc25faad1f24b2024b60ef242e60c77e51d38d93 | |||
| 2014-03-28 | Don't maintain the per-size skiplist unless it's an element on the live | Keith Bostic | |
| "avail" list. The problem is we're holding the live system's spinlock when updating the per-size skiplist, and that appears to be a bottleneck. Reference #924. | |||
| 2014-03-28 | Update sharec_cache configuration again. | Alex Gorrod | |
| Remove enabled option, set default value for name to empty, and require that a name is provided if any shared_cache configuration is provided. Also update upgrading documentation and tests. | |||
| 2014-03-28 | Update shared cache configuration options: | Alex Gorrod | |
| * Change enable to enabled to match log setting. * Return an error if there is a shared_cache configuration string but it's not enabled and not explicit cache_size configuration. | |||
| 2014-03-28 | Take a copy of the update transaction ID in reconciliation in case we race ↵ | Michael Cahill | |
| with a rollback. Go straight to aggressive eviction once the cache gets stuck. refs #894 | |||
| 2014-03-03 | Read threads should not bump the cache read generation: eviction is now ↵ | Michael Cahill | |
| responsible for that. Somehow this part of 30b1a822b26ff0531a0d31aa2895bc98b6f53d88 got lost. | |||
| 2014-03-03 | Use one check for whether to try forced eviction in a given session. Skip ↵ | Michael Cahill | |
| eviction of clean pages with modifications that are too recent, unless eviction gets aggressive. | |||
| 2014-02-28 | Add comments explaining the reconciliation transactional read logic, ↵ | Michael Cahill | |
| including when pages become clean. | |||
| 2014-02-28 | Rename disk_txn to reflect what we're actually using it for. | Michael Cahill | |
| 2014-02-28 | Fix another bug regarding when pages can be evicted: have reconcilition ↵ | Michael Cahill | |
| track the maximum transaction ID on the page (regardless of whether it is skipped). A page can only be evicted when it is clean and has no updates that are too new. refs #884 | |||
| 2014-02-27 | Merge pull request #847 from wiredtiger/checkpoint-directio | Alex Gorrod | |
| Do direct I/O reads on checkpoints to reduce the VM pressure of LSM workloads | |||
| 2014-02-27 | Update documentation for direct_io configuration. | Alex Gorrod | |
| 2014-02-27 | Don't automatically enable direct I/O on checkpoints. | Alex Gorrod | |
| Using direct I/O requires careful selection of page sizes, and doesn't always improve performance. It's not worth enabling it automatically. | |||
| 2014-02-26 | Merge branch 'develop' into checkpoint-directio | Michael Cahill | |
| 2014-02-26 | Merge branch 'develop' into lsm-fixes | Michael Cahill | |
| 2014-02-26 | Update documentation for wiredtiger_config_parser_open | Alex Gorrod | |
| 2014-02-26 | Typo fix that got lost in merge. | Alex Gorrod | |
| 2014-02-26 | Switch to incrementing the global read generation from the eviction server: ↵ | Michael Cahill | |
| this reduces collisions between updates from multiple threads and avoids updates entirely for in-memory workloads. | |||
| 2014-02-26 | Merge branch 'develop' into config-parse-api | Alex Gorrod | |
| Conflicts: src/include/wiredtiger_ext.h | |||
| 2014-02-26 | Add wiredtiger_config_parser_open and wiredtiger_version into extension API. | Alex Gorrod | |
| This is necessary so that statically linked applications can always find the symbols. | |||
| 2014-02-25 | Typo. | Susan LoVerso | |
| 2014-02-25 | Allow some internal pages to be evicted without eviction getting aggressive. | Michael Cahill | |
| Don't require eviction to visit pages twice before they can be evicted: in a large cache, it may take a very long time for eviction to move all the way through the cache, and this was keeping pages pinned for much longer than they needed to be. Remember if some files are not generating many eviction candidates, and visit them less often than files with many evictable pages. | |||
| 2014-02-25 | Configuration parser API, update documentation and examples. | Alex Gorrod | |
| 2014-02-25 | Update the configuration parsing API. Including: | Alex Gorrod | |
| * Don't copy the input string. * Remove reset and remove complex semantics. * Rename search back to get. | |||
| 2014-02-24 | Update the configuration parsing API. Not finalized, but a step closer. | Alex Gorrod | |
| 2014-02-24 | Merge branch 'develop' into lsm-fixes | Michael Cahill | |
| 2014-02-21 | Fix a long-standing bug where a page could be marked clean (and evicted) ↵ | Michael Cahill | |
| when it contained uncommitted changes. refs #835, #818(?), #581 | |||
| 2014-02-20 | Only evict Bloom filter pages if eviction gets aggressive. | Michael Cahill | |
| 2014-02-20 | Use a count in LSM cursors to decide whether to throttle: the shared count ↵ | Michael Cahill | |
| can race and with huge records, there may not be enough records per chunk to get effective throttling. Also tweak LSM merges to minimize throttling: stick with low-level merges whenever the throttle value is non-zero. refs #874, #663 | |||
| 2014-02-17 | Increase the skew against evicting internal pages: some wtperf workloads ↵ | Michael Cahill | |
| cause excessive eviction. refs #829 (etc.) | |||
| 2014-02-13 | Add configuration parsing to the public API. | Alex Gorrod | |
| Don't remove it from the extension API for now - to match the value packing code. Updated inline documentation, but not the official documentation - since I suspect this will change before it's merged. Update wtperf usage of configuration parsing API. | |||
| 2014-02-07 | Merge branch 'develop' into checkpoint-directio | Michael Cahill | |
| 2014-02-06 | Add statistics to track the amount of time application threads spend ↵ | Michael Cahill | |
| throttled for LSM checkpoints or merges. | |||
| 2014-02-03 | Only retry compress_raw if it returns EAGAIN. Otherwise, if no rows were ↵ | Michael Cahill | |
| compressed, assume that compression failed and fall back to standard compression. | |||
| 2014-01-31 | Rewrite the retry/no-retry paragraphs to echo each other "if no rows | Keith Bostic | |
| could be compressed" isn't the same as "don't try again". | |||
| 2014-01-31 | We use 0/zero inconsistently in the documentation, fix one place so we | Keith Bostic | |
| don't in adjacent paragraphs, talking about the same variables. | |||
| 2014-01-31 | Fix several problems with raw compression: | Michael Cahill | |
| * when shuffling entries around, ranges can overlap (found by valgrind). Use memmove rather than memcpy. * use the allocation size to figure out which keys can't be split: with 512 byte leaf pages, the first offset was always near 1KB so compression almost always failed. * provide a way for raw compression to indicate that compression failed so that WiredTiger immediately falls back to compressing the entire page, rather than calling raw compression again with more records. | |||
| 2014-01-30 | Merge pull request #851 from wiredtiger/merge-efficiency | Alex Gorrod | |
| Improve in-memory merge efficiency when there is a lot of forced eviction | |||
| 2014-01-30 | Remove WT_MERGE_MAX_REFS -- it was always arbitrary, and now when we create ↵ | Michael Cahill | |
| an internal page with more than 100 children, it won't be marked split merge, so we won't merge it again. | |||
| 2014-01-30 | Note that LSM's `chunk_size` setting overrides `memory_page_max`. | Michael Cahill | |
| 2014-01-30 | Try an in-memory split before evicting big pages: that avoids the tree ↵ | Michael Cahill | |
| growing arbitrarily deep for append-only workloads. | |||
| 2014-01-24 | If an in-memory split fails, unlock the page. | Michael Cahill | |
| 2014-01-23 | Merge branch 'develop' into merge-efficiency | Alex Gorrod | |
| 2014-01-22 | Fix some GCC 4.8 "maybe-uninitialized" warnings. | Michael Cahill | |
| 2014-01-21 | Try an in-memory split before forced eviction, also make the "promotion" ↵ | Michael Cahill | |
| code for merges apply in more cases than just a single first/last child. | |||
| 2014-01-17 | Do direct I/O reads on checkpoints to reduce the VM pressure of LSM workloads. | Michael Cahill | |
| 2014-01-14 | whitespace | Michael Cahill | |
| 2014-01-14 | Add a new mode to eviction that only considers internal pages. | Alex Gorrod | |
| The mode is triggered when we are force evicting pages, and the cache is not yet full. If we don't merge internal pages while doing forced eviction we can end up with very deep chains of WT_REC_SPLIT pages. | |||
| 2014-01-13 | Switch LSM chunks back to ordinary flags, protected by the LSM tree lock. | Michael Cahill | |
| refs #809 | |||
| 2014-01-10 | Add WT_CLEAR_INLINE which uses structure assignment to (hopefully) prevent | Keith Bostic | |
| any function call or loop setup/tear-down when clearing structures, call it when unpacking cells and doing row-store searches. | |||
 |
index : mongo.git | |
| Unnamed repository; edit this file 'description' to name the repository. | Lucas Castro |
| summaryrefslogtreecommitdiff |