Commit message (Collapse) | Author | Age | Files | Lines | ||
---|---|---|---|---|---|---|
... | ||||||
* | Merge pull request #2478 from ↵ | Tor Brede Vekterli | 2017-05-29 | 7 | -151/+203 | |
|\ | | | | | | | | | yahoo/vekterli/remove-special-handling-of-trusted-replicas-for-merge-source-only-labeling-rebased-fixed Remove special handling of trusted replicas for merge source only labeling (rebased/fixed) | |||||
| * | Remove leftover debugging statement. | Tor Brede Vekterli | 2017-05-16 | 1 | -1/+0 | |
| | | ||||||
| * | Simplify source-only selection logic for replicas. | Tor Brede Vekterli | 2017-05-16 | 3 | -96/+78 | |
| | | | | | | | | | | | | | | | | | | | | | | | | | | Don't take trusted flag into account as it's not to be trusted (the irony is not lost on anyone) when deciding what nodes to mark as source-only during a merge. Now we always mark non-ideal replicas that exceed the redundancy level as source-only to ensure they are removed when the merge is complete. This prevents former ideal replicas from not being marked source-only simply because they happen to be in sync with the now-ideal replicas. Still exist some edges where a source-only replica isn't removed, such as when it's marked active at the time of the merge's completion. | |||||
| * | Ensure merge node limiter handles trusted source-only replicas | Tor Brede Vekterli | 2017-05-16 | 4 | -55/+126 | |
| | | | | | | | | | | | | Could previously risk having a resulting node set be comprised solely of source-only replicas, which luckily gets rejected by the merge throttling component. | |||||
* | | Transfer ownership. | Haakon Dybdahl | 2017-05-23 | 1 | -1/+1 | |
|/ | ||||||
* | Revert "Simplify source-only selection logic for replicas." | Harald Musum | 2017-05-13 | 3 | -78/+96 | |
| | ||||||
* | Simplify source-only selection logic for replicas. | Tor Brede Vekterli | 2017-05-11 | 3 | -96/+78 | |
| | | | | | | | | | | | | | Don't take trusted flag into account as it's not to be trusted (the irony is not lost on anyone) when deciding what nodes to mark as source-only during a merge. Now we always mark non-ideal replicas that exceed the redundancy level as source-only to ensure they are removed when the merge is complete. This prevents former ideal replicas from not being marked source-only simply because they happen to be in sync with the now-ideal replicas. Still exist some edges where a source-only replica isn't removed, such as when it's marked active at the time of the merge's completion. | |||||
* | Deinline as compiler makes the best choices. | Henning Baldersheim | 2017-05-09 | 13 | -38/+54 | |
| | ||||||
* | Initialize all native members. | Henning Baldersheim | 2017-05-06 | 1 | -3/+3 | |
| | ||||||
* | Move code from .hpp to cpp file to avoid unused anonymous function ↵ | Henning Baldersheim | 2017-05-05 | 4 | -32/+31 | |
| | | | | complaints from some compilations units that do not use it. | |||||
* | Merge pull request #2391 from ↵ | Henning Baldersheim | 2017-05-04 | 4 | -0/+8 | |
|\ | | | | | | | | | yahoo/geirst/make-searchcore-compile-without-optimizations Make searchcore compile without optimizations (-Og). | |||||
| * | Make searchcore compile without optimizations (-Og). | Geir Storli | 2017-05-04 | 4 | -0/+8 | |
| | | ||||||
* | | Must give path for configs to build out of source and to build in CLion. | Arnstein Ressem | 2017-05-04 | 2 | -2/+2 | |
|/ | ||||||
* | Add missing file | Henning Baldersheim | 2017-05-02 | 1 | -0/+14 | |
| | ||||||
* | Fix warnings hidden earlier due to including application headers as system ↵ | Henning Baldersheim | 2017-05-02 | 89 | -1049/+644 | |
| | | | | includes | |||||
* | Add operation sequencing for put, remove and update operations (#2252) | Tor Brede Vekterli | 2017-04-25 | 19 | -47/+543 | |
| | | | | | | | | | | | Keeps track of GIDs of pending operations and bounces operations that arrive to these GIDs before the original operation has completed. RemoveLocation operations are currently not handled due to these covering an a priori unknown set of GIDs, but sequencer can be extended to support these as well with some extra work. Enabled by default, but may be disabled via config. | |||||
* | optimize includes. | Henning Baldersheim | 2017-04-25 | 12 | -46/+22 | |
| | ||||||
* | Add missing override. | Henning Baldersheim | 2017-04-25 | 5 | -60/+46 | |
| | ||||||
* | - Optimize includes. | Henning Baldersheim | 2017-04-25 | 98 | -936/+612 | |
| | | | | | - Move htmltable code to implementation file and add add override. - No virtual on override. | |||||
* | Add missing files. | Henning Baldersheim | 2017-04-24 | 2 | -0/+109 | |
| | ||||||
* | - Deinline large metrics constructors/destructors. | Henning Baldersheim | 2017-04-24 | 13 | -360/+165 | |
| | | | | | - Optimize includes - remove virtual when override. | |||||
* | Revert "Balder/enforce override 2" | Arne H Juul | 2017-04-24 | 101 | -666/+1094 | |
| | ||||||
* | Follow api change | Henning Baldersheim | 2017-04-23 | 1 | -4/+2 | |
| | ||||||
* | Some cleanup after major rebase. | Henning Baldersheim | 2017-04-23 | 3 | -29/+35 | |
| | ||||||
* | Including storage now builds fine with override enforcement. | Henning Baldersheim | 2017-04-23 | 98 | -1061/+629 | |
| | ||||||
* | Merge pull request #2212 from ↵ | Tor Brede Vekterli | 2017-04-20 | 6 | -13/+166 | |
|\ | | | | | | | | | yahoo/vekterli/do-not-erroneously-mark-replica-as-trusted-during-bucket-info-merge Do not erroneously mark replicas as trusted when merging bucket info | |||||
| * | Do not erroneously mark replicas as trusted when merging bucket info | Tor Brede Vekterli | 2017-04-20 | 6 | -13/+166 | |
| | | | | | | | | | | | | | | | | | | | | | | | | | | Addresses two long-standing distributor bugs: 1) Fix removal loop of bucket replica information removal for nodes whose information has been fetched (i.e. "outdated" nodes). Could previously prematurely terminate loop. 2) Defer update of trusted flag on replicas until _after_ all replicas have been removed for outdated nodes AND new replica information has been merged in. Fixes edge case where e.g. transiently removing 1 of 2 mutually out of sync replicas before merging its info back in again would mark the remaining replica as trusted due to being the only remaining replica in this transient period. | |||||
* | | remove FastOS_Socket::getHostName | Haavard | 2017-04-19 | 1 | -1/+2 | |
|/ | ||||||
* | add override in storage module | Arne H Juul | 2017-04-12 | 1 | -1/+2 | |
| | ||||||
* | add override in storage tests | Arne H Juul | 2017-04-11 | 56 | -136/+136 | |
| | ||||||
* | add override in storage/visiting module | Arne H Juul | 2017-04-11 | 9 | -46/+41 | |
| | ||||||
* | add override in storage/tools module | Arne H Juul | 2017-04-11 | 2 | -15/+15 | |
| | ||||||
* | add override in storage/storageserver module | Arne H Juul | 2017-04-11 | 10 | -75/+68 | |
| | ||||||
* | add override in storage/persistence module | Arne H Juul | 2017-04-11 | 8 | -67/+60 | |
| | ||||||
* | add override in storage/frameworkimpl module | Arne H Juul | 2017-04-11 | 7 | -15/+13 | |
| | ||||||
* | add override in storage/distributor module | Arne H Juul | 2017-04-11 | 17 | -103/+101 | |
| | ||||||
* | add override in storage/config module | Arne H Juul | 2017-04-11 | 1 | -1/+1 | |
| | ||||||
* | add override in storage/common module | Arne H Juul | 2017-04-11 | 6 | -29/+31 | |
| | ||||||
* | add override in storage/bucketmover module | Arne H Juul | 2017-04-11 | 1 | -2/+1 | |
| | ||||||
* | add override in storage/bucketdb module | Arne H Juul | 2017-04-11 | 6 | -28/+21 | |
| | ||||||
* | let EntryWrapper expose internal value with get function | Haavard | 2017-03-28 | 1 | -0/+4 | |
| | ||||||
* | use std::move instead of release/capture | Haavard | 2017-03-28 | 1 | -1/+1 | |
| | ||||||
* | remove most usage of LinkedPtr from vespa | Haavard | 2017-03-27 | 45 | -97/+88 | |
| | ||||||
* | Ignore document sizes in DeleteBucket sanity checking | Tor Brede Vekterli | 2017-03-20 | 2 | -18/+57 | |
| | | | | | | | Just checking checksum + number of docs should suffice for this. Prevents false positives when Proton does background moving of documents that previously did not have an actual serialized size computed. | |||||
* | No more postfix operator. | Henning Baldersheim | 2017-03-15 | 1 | -1/+1 | |
| | ||||||
* | No more postfix operator. | Henning Baldersheim | 2017-03-15 | 1 | -1/+1 | |
| | ||||||
* | Merge pull request #1993 from ↵ | Henning Baldersheim | 2017-03-13 | 1 | -1/+1 | |
|\ | | | | | | | | | yahoo/vekterli/dont-return-stringref-from-temporary Don't return stringref to temporary buffer | |||||
| * | Don't return stringref to temporary buffer | Tor Brede Vekterli | 2017-03-13 | 1 | -1/+1 | |
| | | | | | | | | | | | | This line was last changed in 2011 so, uh, it's been buggy for a while. Presumably this has been aggressively inlined, so going via the stringref hasn't actually happened during normal compilation. | |||||
* | | Create explicit, non implicitly inlined function definitions | Tor Brede Vekterli | 2017-03-13 | 3 | -0/+9 | |
|/ | ||||||
* | Deinline destructorsi and use -Winline gcc option. | Henning Baldersheim | 2017-03-09 | 5 | -129/+136 | |
| |