aboutsummaryrefslogtreecommitdiffstats
path: root/storage
Commit message (Collapse)AuthorAgeFilesLines
* use new input/output model in slimeHaavard2017-02-035-7/+7
|
* Clean up a very dangerous interface to prevent potential misuse.Henning Baldersheim2017-01-201-4/+4
|
* Use more descriptive accessorTor Brede Vekterli2017-01-121-3/+3
|
* Track time spent transitioning between cluster statesTor Brede Vekterli2017-01-125-12/+120
| | | | | Allows observing whether state transitions take an unexpected amount of time.
* copy out the stuff you access after erasing the node.Henning Baldersheim2017-01-111-1/+2
|
* Rewrite the getNextMessage loop so that is easier to understand and followHenning Baldersheim2017-01-112-42/+29
|
* refactor for readability.Henning Baldersheim2017-01-112-27/+32
|
* Do not broadcast after fetching a message.Henning Baldersheim2017-01-111-2/+1
|
* Use std::moveHenning Baldersheim2017-01-111-13/+13
|
* back to using broadcastHenning Baldersheim2017-01-111-1/+1
|
* Signal instead of broadcast.Henning Baldersheim2017-01-111-1/+1
|
* Use emplace back and std::move to reduce expensive copying of shared pointers.Henning Baldersheim2017-01-113-7/+15
|
* Show that we hold the lock.Henning Baldersheim2017-01-112-5/+9
|
* Unlock before return.Henning Baldersheim2017-01-111-0/+1
|
* Move the pair of shared_ptrs on return since there is multiple return points.Henning Baldersheim2017-01-111-1/+1
|
* Move the shared_ptrHenning Baldersheim2017-01-111-2/+2
|
* Delay taking a shared_ptr until it is really necessary.Henning Baldersheim2017-01-111-14/+9
|
* Unify includesHenning Baldersheim2017-01-111-3/+3
|
* mkdir -p instead of rm -rfHenning Baldersheim2017-01-041-4/+4
|
* Ensure that we use unique directories.Henning Baldersheim2017-01-0211-43/+62
|
* Include the correct files the correct place.Henning Baldersheim2017-01-021-6/+4
|
* Remove unused files.Henning Baldersheim2017-01-022-233/+0
|
* Use 8 threads for testing storage.Henning Baldersheim2017-01-021-1/+1
|
* Invalid path to include file in case of out of source builds. C++11 for loops.Henning Baldersheim2016-12-221-16/+8
|
* Use correct metric value increment methodTor Brede Vekterli2016-12-211-3/+3
| | | | | | | Turns out, inc() and addValue() have completely different semantics for how values are actually aggregated. inc() maintains a running value while addValue() treats each invocation as a separate sample (which is what we want).
* Merge pull request #1362 from ↵Henning Baldersheim2016-12-212-3/+4
|\ | | | | | | | | yahoo/vekterli/use-floating-point-latency-timer-for-get-operation Use floating point timer for Get operation latency metric
| * Use floating point timer for Get operation latency metricTor Brede Vekterli2016-12-202-3/+4
| |
* | Deinline some non-trivial methods and do not provide assert.h in string.Henning Baldersheim2016-12-213-1/+4
| |
* | Avoid pulling all of config everywhere.Henning Baldersheim2016-12-205-11/+8
| |
* | Further decouple some hpp files, config and the attributevector.Henning Baldersheim2016-12-2032-154/+139
|/
* Update tests with explicitt includes.Henning Baldersheim2016-12-196-3/+7
|
* GC unused code and only include the stuff needed.Henning Baldersheim2016-12-191-0/+55
|
* Do not carry exceptions in header filesHenning Baldersheim2016-12-1913-104/+59
|
* GC unused code.Henning Baldersheim2016-12-193-13/+15
|
* GC unused code.Henning Baldersheim2016-12-193-808/+0
|
* Do not carry exceptions in header filesHenning Baldersheim2016-12-192-2/+3
|
* decouple slime from config.Henning Baldersheim2016-12-192-6/+4
|
* Do not include fastos.h in header files.Henning Baldersheim2016-12-193-15/+28
|
* Do not include fastos.h in header files.Henning Baldersheim2016-12-1911-13/+3
|
* Merge pull request #1338 from ↵Tor Brede Vekterli2016-12-161-4/+4
|\ | | | | | | | | yahoo/vekterli/increase-default-distributor-buckets-checked-per-second Increase number of buckets/sec processed by distributors
| * Increase number of buckets/sec processed by distributorsTor Brede Vekterli2016-12-151-4/+4
| | | | | | | | | | | | | | | | | | | | This greatly reduces time to process large bucket databases, especially when the number of buckets go into the millions. With the previous defaults, this could take absolute ages and fail to get in sync within a reasonable time. If we do not observe any negative effects as part of this change, we'll likely boost the ticks_before_wait default even more.
* | Complete the split also for tests.Henning Baldersheim2016-12-152-9/+3
| |
* | Split in hash_xxx, array, lru, cache ++ in hpp files. To reduce clinon buildHenning Baldersheim2016-12-1531-243/+153
|/
* Add correct include.Henning Baldersheim2016-12-121-6/+1
|
* Avoid pulling in all of slobrok and config everywhere.Henning Baldersheim2016-12-1212-12/+18
|
* Split out so documentprotocol.h does not pull in everything.Henning Baldersheim2016-12-1213-50/+35
|
* Do not bring in nbostream everywhere.Henning Baldersheim2016-12-123-1/+3
|
* Avoid building all the document code in every file.Henning Baldersheim2016-12-127-22/+18
|
* Include asciistream in implementation only.Henning Baldersheim2016-12-1211-47/+41
|
* Include asciistream in implementation only.Henning Baldersheim2016-12-127-239/+13
|