Commit message (Collapse) | Author | Age | Files | Lines | |
---|---|---|---|---|---|
* | Remove "vds.filestor.*" metrics from consumer fleetcontroller ↵ | Geir Storli | 2018-03-20 | 2 | -22/+4 |
| | | | | (clustercontroller) that are not used. | ||||
* | Merge pull request #5378 from vespa-engine/balder/no-more-thread-priorities | Tor Brede Vekterli | 2018-03-20 | 3 | -73/+17 |
|\ | | | | | Balder/no more thread priorities | ||||
| * | No need for an array when you just need a count | Henning Baldersheim | 2018-03-19 | 3 | -73/+17 |
| | | |||||
* | | Merge pull request #5377 from vespa-engine/bratseth/reduce-old-junit-usage | Jon Bratseth | 2018-03-19 | 6 | -12/+10 |
|\ \ | |/ |/| | Nonfunctional changes only | ||||
| * | Nonfunctional changes only | Jon Bratseth | 2018-03-19 | 6 | -12/+10 |
| | | |||||
* | | Add coverage metrics to default vespa metric set | Henning Baldersheim | 2018-03-16 | 1 | -0/+4 |
| | | |||||
* | | Prepare for removing some unused or noop methods from filedistribution API | Harald Musum | 2018-03-15 | 3 | -12/+20 |
|/ | |||||
* | Merge pull request #5285 from vespa-engine/arnej/add-gzip-accesslogs | Jon Bratseth | 2018-03-15 | 2 | -3/+18 |
|\ | | | | | hack in gzip of access logs | ||||
| * | ensure gzip of access logs for hosted vespa only | Arne Juul | 2018-03-12 | 2 | -3/+18 |
| | | |||||
* | | Prefix macro names by modelName to avoid name conflicts | Jon Bratseth | 2018-03-13 | 1 | -10/+10 |
| | | |||||
* | | Merge pull request #5309 from vespa-engine/bratseth/disambiguate-constants | Jon Bratseth | 2018-03-13 | 3 | -45/+49 |
|\ \ | | | | | | | Prefix constants by model name | ||||
| * | | Prefix constants by model name | Jon Bratseth | 2018-03-13 | 3 | -45/+49 |
| | | | | | | | | | | | | | | | | | | | | | | | | | | | Large constants are cross rank profiles. This avoids name conflicts when multiple models are used. It is not strictly necessary because the user can always disambiguate when chosing names, but there is a scenario where conflicts are plausible and leaving this to users is inconvenient: Multiple versions of the "same" model are tested in different rank profiles. | ||||
* | | | Tensor fields are out of beta | Jon Bratseth | 2018-03-12 | 1 | -8/+0 |
|/ / | |||||
* / | Use tensorflow 1.6 | Henning Baldersheim | 2018-03-12 | 1 | -2/+2 |
|/ | |||||
* | filedistributorservice does not exist anymore | Harald Musum | 2018-03-10 | 1 | -16/+12 |
| | |||||
* | Merge pull request #5267 from vespa-engine/lesters/tensorflow-broadcasting | Jon Bratseth | 2018-03-09 | 1 | -4/+4 |
|\ | | | | | Make TensorFlow import joins compatible with broadcasting | ||||
| * | Make TensorFlow import joins compatible with broadcasting | Lester Solbakken | 2018-03-08 | 1 | -4/+4 |
| | | |||||
* | | Merge pull request #5264 from vespa-engine/gjoranv/secret-store | gjoranv | 2018-03-09 | 7 | -10/+94 |
|\ \ | | | | | | | Gjoranv/secret store | ||||
| * | | Allow setting up a secret store from services.xml. | gjoranv | 2018-03-08 | 6 | -5/+88 |
| | | | | | | | | | | | | - The actual implementation must be added by a model amender. | ||||
| * | | Extract method to keep addClusterContent clean and "short". | gjoranv | 2018-03-08 | 1 | -2/+4 |
| | | | |||||
| * | | Add myself as author. | gjoranv | 2018-03-08 | 1 | -0/+1 |
| | | | |||||
| * | | Improve comment. | gjoranv | 2018-03-08 | 1 | -1/+1 |
| | | | |||||
| * | | Remove invalid TODO. DomQrserverClusterBuilder n'existe plus. | gjoranv | 2018-03-08 | 1 | -2/+0 |
| | | | |||||
* | | | Merge pull request #5265 from ↵ | Geir Storli | 2018-03-08 | 1 | -0/+8 |
|\ \ \ | | | | | | | | | | | | | | | | | vespa-engine/toregge/expose-threading-service-metrics Expose threading service metrics. | ||||
| * | | | Expose threading service metrics. | Tor Egge | 2018-03-08 | 1 | -0/+8 |
| |/ / | |||||
* | | | Merge pull request #5266 from ↵ | Jon Bratseth | 2018-03-08 | 1 | -2/+1 |
|\ \ \ | |_|/ |/| | | | | | | | | vespa-engine/bratseth/convert-tf-tensor-on-command-line Bratseth/convert tf tensor on command line | ||||
| * | | Log OrderedTensorType of imported TensorFlow variables | Jon Bratseth | 2018-03-08 | 1 | -2/+1 |
| |/ | |||||
* / | Add batch dimension reduction in generated macros | Lester Solbakken | 2018-03-08 | 2 | -64/+113 |
|/ | | | | | | Check for inputs/placeholders containing batch dimensions inside generated macros, and insert reduce operations. Expand those dimensions back at the top level expression. | ||||
* | Merge pull request #5241 from vespa-engine/bratseth/tf-constants-in-parent-doc | Lester Solbakken | 2018-03-07 | 2 | -21/+129 |
|\ | | | | | Bratseth/tf constants in parent doc | ||||
| * | Allow macros to replace TenorFlow variables | Jon Bratseth | 2018-03-07 | 2 | -21/+129 |
| | | | | | | | | | | | | Also, remove quoting of constant arguments generated in TensorFlow as that is unnecessary now and is interpreted as a string constant argument to a macro. | ||||
* | | Fix unit test after correct dimension renaming | Lester Solbakken | 2018-03-07 | 1 | -2/+2 |
|/ | |||||
* | Put generated TensorFlow macros in replicated dir | Lester Solbakken | 2018-03-06 | 1 | -1/+1 |
| | |||||
* | Put TensorFlow expression back to replicated dir | Lester Solbakken | 2018-03-06 | 1 | -1/+1 |
| | |||||
* | Add testing of macro generation from stored model | Lester Solbakken | 2018-03-05 | 1 | -14/+21 |
| | |||||
* | Generate macros for TensorFlow nodes with multiple outputs | Lester Solbakken | 2018-03-05 | 2 | -2/+73 |
| | |||||
* | No wildcard static imports | Jon Bratseth | 2018-03-05 | 4 | -8/+17 |
| | |||||
* | Remove junit.fdramework usage | Jon Bratseth | 2018-03-02 | 21 | -55/+97 |
| | |||||
* | Add batch dimension reduction for TensorFlow import | Lester Solbakken | 2018-03-01 | 2 | -4/+98 |
| | |||||
* | Merge pull request #5169 from ↵ | Håkon Hallingstad | 2018-02-28 | 3 | -32/+122 |
|\ | | | | | | | | | vespa-engine/hakonhall/avoid-slobrok-in-node-admin-cluster Avoid Slobrok in node-admin cluster | ||||
| * | Add test | Håkon Hallingstad | 2018-02-28 | 2 | -22/+91 |
| | | |||||
| * | Avoid Slobrok in node-admin cluster | Håkon Hallingstad | 2018-02-28 | 1 | -10/+31 |
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | When this rolls out, two Slobroks on the node-admin will be removed, and 1 additional Slobrok will be put on one of the proxy nodes if the routing cluster has 3 or more nodes. The Orchestrator sees the latest model, i.e. expect 3 Slobroks on the proxy nodes. Since the 1 additional Slobrok service is down at the start of the rollout, only that host will be allowed to suspend among the 3. But if that 1 additional Slobrok service comes up quickly, then any of the 3 can suspend. I think this will just work - no deadlocks or similar. I couldn't find any test that would covers this code change, but will write tests if someone can help me with this, if deemed necessary. | ||||
* | | Add switch to enable multiple bucket spaces in cluster controller. | Geir Storli | 2018-02-27 | 2 | -5/+22 |
|/ | |||||
* | Don't validate when reloading models | Jon Bratseth | 2018-02-26 | 103 | -486/+585 |
| | |||||
* | Disable tests | Jon Bratseth | 2018-02-26 | 1 | -0/+8 |
| | |||||
* | Disable type validation for now | Jon Bratseth | 2018-02-26 | 1 | -1/+1 |
| | |||||
* | Merge pull request #5141 from vespa-engine/bratseth/simplify-tensor-transformer | Jon Bratseth | 2018-02-26 | 21 | -274/+105 |
|\ | | | | | Bratseth/simplify tensor transformer | ||||
| * | Remove println | Jon Bratseth | 2018-02-26 | 1 | -1/+0 |
| | | |||||
| * | Refactor | Jon Bratseth | 2018-02-24 | 3 | -42/+13 |
| | | |||||
| * | Remove dead code | Jon Bratseth | 2018-02-24 | 1 | -163/+15 |
| | | |||||
| * | Cleanup and more type inference | Jon Bratseth | 2018-02-24 | 20 | -85/+94 |
| | | | | | | | | | | | | | | - Use a type context in TensorTransformer - Correct Join.outputType - Understand tensorFromLabels and tensorFromWeightedSet - Don't wrap simple tensor arguments in constant() |