| b4a5a0d | brharrington | 05 June 2017, 22:18:13 UTC | fix possible deadlock during startup (#613) The healthcheck api was accessing the service manager from a provider when the routes were fetched. There doesn't appear to be any reason for this and it was removed already in 1.6 branch as part of #490. This was usually harmless, but on some systems threads would get scheduled in such a way that all of the actor threads were blocked with traces like: ``` "atlas-akka.actor.default-dispatcher-9" #18 prio=5 os_prio=0 tid=0x00007fb914010000 nid=0x6410 waiting on condition [0x00007fb9235f7000] java.lang.Thread.State: WAITING (parking) at sun.misc.Unsafe.park(Native Method) - parking to wait for <0x00000000838ae498> (a java.util.concurrent.locks.ReentrantLock$NonfairSync) at java.util.concurrent.locks.LockSupport.park(LockSupport.java:175) at java.util.concurrent.locks.AbstractQueuedSynchronizer.parkAndCheckInterrupt(AbstractQueuedSynchronizer.java:836) at java.util.concurrent.locks.AbstractQueuedSynchronizer.acquireQueued(AbstractQueuedSynchronizer.java:870) at java.util.concurrent.locks.AbstractQueuedSynchronizer.acquire(AbstractQueuedSynchronizer.java:1199) at java.util.concurrent.locks.ReentrantLock$NonfairSync.lock(ReentrantLock.java:209) at java.util.concurrent.locks.ReentrantLock.lock(ReentrantLock.java:285) at com.google.inject.internal.CycleDetectingLock$CycleDetectingLockFactory$ReentrantCycleDetectingLock.lockOrDetectPotentialLocksCycle(CycleDetectingLock.java:164) at com.google.inject.internal.SingletonScope$1.get(SingletonScope.java:185) at com.google.inject.internal.InternalFactoryToProviderAdapter.get(InternalFactoryToProviderAdapter.java:41) at com.google.inject.internal.InjectorImpl$2$1.call(InjectorImpl.java:1019) at com.google.inject.internal.InjectorImpl.callInContext(InjectorImpl.java:1085) at com.google.inject.internal.InjectorImpl$2.get(InjectorImpl.java:1015) at com.netflix.atlas.akka.HealthcheckApi.routes(HealthcheckApi.scala:36) at com.netflix.atlas.akka.RequestHandlerActor.receive(RequestHandlerActor.scala:41) at akka.actor.ActorCell.newActor(ActorCell.scala:568) at akka.actor.ActorCell.create(ActorCell.scala:588) at akka.actor.ActorCell.invokeAll$1(ActorCell.scala:461) at akka.actor.ActorCell.systemInvoke(ActorCell.scala:483) at akka.dispatch.Mailbox.processAllSystemMessages(Mailbox.scala:282) at akka.dispatch.Mailbox.run(Mailbox.scala:223) at akka.dispatch.Mailbox.exec(Mailbox.scala:234) at scala.concurrent.forkjoin.ForkJoinTask.doExec(ForkJoinTask.java:260) at scala.concurrent.forkjoin.ForkJoinPool$WorkQueue.runTask(ForkJoinPool.java:1339) at scala.concurrent.forkjoin.ForkJoinPool.runWorker(ForkJoinPool.java:1979) at scala.concurrent.forkjoin.ForkJoinWorkerThread.run(ForkJoinWorkerThread.java:107) ``` The user would see this as the application never fully starting and not being accessible. This should fix issue #612. | 05 June 2017, 22:18:13 UTC |
| 7f660b3 | brharrington | 03 May 2017, 14:41:35 UTC | backport bug fixes to 1.5.x (#570) | 03 May 2017, 14:41:35 UTC |
| 01becfe | brharrington | 15 March 2017, 23:25:38 UTC | fix bounds exception for Integer.MIN_VALUE hash (#515) If an object hashes to `Integer.MIN_VALUE`, then the absolute value will be `Integer.MIN_VALUE` and a negative values was used for looking up data in the underlying array. This change will map those to be `0` instead. | 03 May 2017, 14:31:44 UTC |
| 0c78d5f | brharrington | 25 March 2017, 20:16:39 UTC | improve performance of SmallHashMap.{+,-} (#526) Changes the small hash map implementation of `+` and `-` to retain the type. This is primarily to avoid using a less memory efficient map type if the map is modified somewhere along the way. A side benefit is better performance and reduced allocations. **Before** ``` Benchmark Mode Cnt Score Error Units addPair thrpt 10 781344.330 ± 25173.790 ops/s removePair thrpt 10 738796.252 ± 44849.997 ops/s ``` **After** ``` Benchmark Mode Cnt Score Error Units addPair thrpt 10 1637470.772 ± 29948.315 ops/s removePair thrpt 10 2520939.662 ± 32320.759 ops/s ``` | 03 May 2017, 14:29:11 UTC |
| ae410c5 | brharrington | 26 March 2017, 03:36:22 UTC | add additional build metadata to jar (#527) Add the build date and number to the jar manifest to aid in debugging. | 03 May 2017, 14:28:18 UTC |
| c5cff6f | brharrington | 27 March 2017, 15:51:39 UTC | add `atlas.rollup` to allowed keys (#528) Updates the default allowed keys list for `atlas.` prefix to include `atlas.rollup`. | 03 May 2017, 14:27:06 UTC |
| 7f011d6 | brharrington | 05 April 2017, 18:00:41 UTC | fix prefix for cloudwatch sqs metrics (#532) It was incorrectly specified as `s3`. | 03 May 2017, 14:22:22 UTC |
| bb8bfd0 | brharrington | 11 February 2017, 17:37:52 UTC | 1.5.x backport (#502) Backport select changes from 1.6 to 1.5. | 11 February 2017, 17:37:52 UTC |
| ec146ca | Brian Harrington | 11 February 2017, 03:54:17 UTC | fix #498, always honor order (#500) The order was getting ignored if there wasn't an explicit sort setting. | 11 February 2017, 04:05:40 UTC |
| 87a0168 | brharrington | 28 January 2017, 05:55:08 UTC | fix possible IAE if NaN values are in sorted set (#487) In some cases the sort would fail with: ``` IllegalArgumentException: Comparison method violates its general contract! ``` This was caused by having a mix of NaNs in the set of values being sorted. This change switches to use `Double.compare` which takes NaN values into account. | 11 February 2017, 03:41:05 UTC |
| 0ae65b1 | Brian Harrington | 11 February 2017, 03:40:35 UTC | update license headers for 2017 | 11 February 2017, 03:40:35 UTC |
| f455edb | brharrington | 26 January 2017, 04:46:44 UTC | fix bug with equality check on SmallHashMap (#482) For maps with the same length and hashCodes the equality check would fall back to `dataEquals`. That method had a bug where if different keys were in the same position of the underlying data array, then the values would not get checked and the entry would get skipped. This could cause some maps to incorrectly get treated as being equal even though they are not. | 11 February 2017, 03:30:35 UTC |
| 37771dd | brharrington | 08 November 2016, 15:09:40 UTC | backport select changes from 1.6 to 1.5.x branch (#454) | 08 November 2016, 15:09:40 UTC |
| e093545 | brharrington | 08 November 2016, 14:16:44 UTC | some patterns incorrectly interpretted as indexOf (#453) For simple patterns the `StringMatcher` will convert them to basic string operations that perform better than using regular expressions. If a simple pattern had an explicit end anchor, then it was incorrectly choosing the `IndexOf` matcher rather than falling back to using `Regex`. This change fixes the `IndexOf` pattern and also adds an `Equals` matcher to cover the simple case using `String.equals`. | 08 November 2016, 14:21:10 UTC |
| ec1d41a | brharrington | 08 November 2016, 13:52:27 UTC | fix #451: port setting in sample memory.conf (#452) Port setting in sample memory.conf was not updated to reflect changes introduced in #396. | 08 November 2016, 14:20:51 UTC |
| ed0f6cf | Brian Harrington | 03 November 2016, 14:46:40 UTC | update to spray 1.3.4 This version fixes a security vulerability for directory traversals on Windows: https://groups.google.com/forum/#!topic/spray-user/4V3F_oLm_TA We aren't using that method directly and do not run on Windows, but that might not be true downstream. | 08 November 2016, 14:20:40 UTC |
| 1fea402 | brharrington | 03 November 2016, 15:10:21 UTC | docs: getting started version to 1.5.0 (#449) | 08 November 2016, 14:20:25 UTC |
| 69aa97f | brharrington | 29 October 2016, 14:14:06 UTC | update dependencies (#447) | 29 October 2016, 14:14:06 UTC |
| 99ce6cd | Brian Harrington | 29 October 2016, 14:04:55 UTC | scalatest 3.0.0 | 29 October 2016, 14:04:55 UTC |
| d46ba5a | Brian Harrington | 29 October 2016, 13:55:40 UTC | equalsverifier 2.1.6 | 29 October 2016, 13:55:40 UTC |
| a8021e7 | Brian Harrington | 29 October 2016, 13:54:06 UTC | iep 0.4.9 | 29 October 2016, 13:54:06 UTC |
| 756b68a | Brian Harrington | 29 October 2016, 13:50:56 UTC | sbt-jmh 0.2.16 | 29 October 2016, 13:50:56 UTC |
| 7c20b73 | Brian Harrington | 29 October 2016, 13:44:17 UTC | aws-java-sdk 1.11.49 | 29 October 2016, 13:45:08 UTC |
| daed62a | Brian Harrington | 29 October 2016, 13:42:52 UTC | jol 0.6 | 29 October 2016, 13:42:52 UTC |
| 5bb4e6d | Brian Harrington | 29 October 2016, 13:41:59 UTC | log4j 2.7 | 29 October 2016, 13:41:59 UTC |
| abd41f1 | Brian Harrington | 29 October 2016, 13:34:05 UTC | akka 2.4.12 | 29 October 2016, 13:34:05 UTC |
| d784f60 | Brian Harrington | 29 October 2016, 13:25:12 UTC | spectator 0.42.0 | 29 October 2016, 13:25:12 UTC |
| e01a836 | Brian Harrington | 29 October 2016, 13:21:29 UTC | jackson 2.8.3 | 29 October 2016, 13:21:29 UTC |
| faee3cb | brharrington | 26 October 2016, 20:31:08 UTC | improve generic type detection (#446) Uses the type information from the bean description when available so we have more context as to the requested types for generic parameters. The `generics 2` test case would fail before this change because it would decode to a `Map` instead of `Inner`. | 26 October 2016, 20:31:08 UTC |
| 6c8aade | brharrington | 24 October 2016, 22:18:44 UTC | do not call readValue if current token is null (#445) If the current token for the parser is `VALUE_NULL`, then use the default rather than call readValue. The readValue call can throw which will cause it to incorrectly fail to parse the object. Exceptions like: ``` com.fasterxml.jackson.databind.JsonMappingException: Can not deserialize instance of scala.collection.immutable.List out of VALUE_NULL token ``` | 24 October 2016, 22:18:44 UTC |
| 9747d94 | brharrington | 10 October 2016, 20:13:16 UTC | fix update of invalid datapoints metric (#444) The counter was not getting incremented. | 10 October 2016, 20:13:16 UTC |
| 27036e4 | brharrington | 01 October 2016, 21:48:40 UTC | fill with zero for periods of inactivity (#442) CloudWatch graphs will frequently show gaps if there is low activity and if the reporting frequency is much larger than the step size used with Atlas. With this change we'll use an explicit zero value. | 01 October 2016, 21:48:40 UTC |
| 5a657ee | brharrington | 01 October 2016, 21:10:19 UTC | indicate filtering behavior in graph legend (#441) Filters can be confusing when users get an empty graph because nothing matched the filter criteria. The legend should be updated to indicate how filtering modified the result set. | 01 October 2016, 21:10:19 UTC |
| d587ad1 | Brian Harrington | 01 October 2016, 20:58:21 UTC | add message to legend to summarize filters For the simple case where there is a single group by expression that is filtered, a message is now added to the legend indicating how many matches there were to the filter. These can quickly get complicated when used with other features. For example, sorting can mix and match lines across multiple expressions. Also binary math operations that combine the results of multiple filter expressions or multi-level group by with filtered input. For now this is just an experiment for the common simple case to see how it impacts usability when dealing with filter expressions that remove some of the lines. | 01 October 2016, 20:58:21 UTC |
| 71ef776 | Brian Harrington | 01 October 2016, 19:52:07 UTC | remove old diff files if images are now the same Since we no longer generate the diff image if the test and golden files are the same, the old diff images will not get overridden and show up as a diff even if the problem has since been fixed. Now the old diff image will get deleted. | 01 October 2016, 19:52:07 UTC |
| 43da9ec | Brian Harrington | 01 October 2016, 19:48:20 UTC | add test case for MessageDef rendering | 01 October 2016, 19:48:20 UTC |
| f799e76 | Brian Harrington | 01 October 2016, 17:35:49 UTC | use error image for new tests with no golden file Updates the report to make it easier to see images for new tests without having to bless first. | 01 October 2016, 17:35:49 UTC |
| 7213191 | Brian Harrington | 01 October 2016, 17:15:04 UTC | include all DataDef types in legend Before the legend was only including LineDefs. Now spans are also include and a new MessageDef type is added to allow legend entries that are not tied to visual data. For example, information about filtering the results of a group by where all lines have been removed. | 01 October 2016, 17:15:04 UTC |
| 0dead52 | Brian Harrington | 01 October 2016, 17:04:01 UTC | restrict image report to diffs by default Updates the GraphAssertions to only show the entries where there are differences by default. The previous behavior is still available by calling `generateReport` with the `diffsOnly` param set to `false`. Also changes the code to only write the diff image to disk if there is an actual difference. | 01 October 2016, 17:04:01 UTC |
| f75fe35 | brharrington | 30 September 2016, 20:19:22 UTC | fix minor issues with case class deserializer (#440) This change fixes two issues with the case class deserializer added in #439: 1. We now check the initial token type to verify it is an object. 2. Avoids creating the object before parsing the fields. That was done before to avoid the allocating the argument array for the constructor. However, it prevents users from being able to have additional validation as part of the constructor logic. Now the object will be created after the fields have all been processed and placed in the arguments array. | 30 September 2016, 20:19:22 UTC |
| 294dc6b | brharrington | 28 September 2016, 22:54:05 UTC | add custom case class deserializer (#439) Adds a custom case class deserializer to the json helper that supports default values used in the case class constructor. This is the more intuitive behavior most users expect and avoids the need for wrapping in Option or having custom deserializers for a given type to fill in the defaults. | 28 September 2016, 22:54:05 UTC |
| 77b8eef | brharrington | 25 September 2016, 02:06:03 UTC | dependency updates (#437) | 25 September 2016, 02:06:03 UTC |
| ea92e4d | brharrington | 25 September 2016, 02:04:42 UTC | fix #436, QueryIndex misses some entries (#438) Once there was a match at a given level it would check the sub-tree, but never check for matches among siblings. Adds the test case from the issue and a second that demonstrates the issue for the `matches` function as well. The original test case would pass for `matches` because both sub-trees in the index would match. | 25 September 2016, 02:04:42 UTC |
| 30a5f86 | Brian Harrington | 25 September 2016, 01:17:56 UTC | config 1.3.1 | 25 September 2016, 01:17:56 UTC |
| 0541fcf | Brian Harrington | 25 September 2016, 01:16:05 UTC | caffeine 2.3.3 | 25 September 2016, 01:16:05 UTC |
| 31f41b2 | Brian Harrington | 25 September 2016, 01:15:32 UTC | aws-java-sdk 1.11.37 | 25 September 2016, 01:15:32 UTC |
| 8f08e5b | Brian Harrington | 25 September 2016, 01:11:17 UTC | iep 0.4.8 | 25 September 2016, 01:11:17 UTC |
| 136ec1c | brharrington | 07 September 2016, 12:00:39 UTC | jackson 2.8.2 (#434) This version fixes a bug with `Option[Trait]` getting encoded as an empty object. | 07 September 2016, 12:00:39 UTC |
| 2cac104 | brharrington | 05 September 2016, 22:28:30 UTC | less repetitive expression normalization (#433) Update the `/api/v1/expr/normalize` endpoint to be less repetitive. In particular, the macros like `:avg` and `:dist-stddev` are no longer expanded. Common queries will also get extracted. Analyzing around 60k expressions from logs: | Result | Before | After | |--------------|----------|---------| | longer | 28,682 | 8,968 | | shorter | 30,678 | 46,170 | | same length | 2,630 | 6,852 | | 05 September 2016, 22:28:30 UTC |
| 922aeb8 | Brian Harrington | 05 September 2016, 21:26:18 UTC | update normalize to be less repetitive This mainly consists of two main changes: 1. Not all macros are expanded. Some such as, `:avg` and `:dist-stddev`, will be preserved so that the user sees the simple operation rather than the full math to compute those operations. 2. Common-queries will now be extracted and added as a final `:list,(,{query},:cq,),:each` expression if it reduces the overall length of the expression. If present, then that will always be the last element in the list of normalized expressions. This change also updates the DES helper macros to avoid expansion. | 05 September 2016, 22:12:21 UTC |
| c34cd38 | Brian Harrington | 05 September 2016, 21:16:10 UTC | retain rewrite information for aggr macros For macros that are meant to appear to the user like additional aggregate functions it is preferrable if we keep track of the rewrite rather than fully expand it when presenting the expression to the user. This change adds a NamedRewrite operation that keeps track of the original expression for display as well as the rewritten expression for evaluation. | 05 September 2016, 21:16:10 UTC |
| d1e9318 | Brian Harrington | 05 September 2016, 21:10:56 UTC | do not emit trailing comma when settings is empty | 05 September 2016, 21:10:56 UTC |
| c891340 | Brian Harrington | 05 September 2016, 21:00:31 UTC | :cq should elide :true for final query When using `:cq` it should elide `:true` if present rather than generating an expression ANDd with `:true`. For example: ``` :true,:sum,name,foo,:eq,:cq ``` Before this change would expand to: ``` :true,name,foo,:eq,:and:sum ``` Now expands to: ``` name,foo,:eq,:sum ``` | 05 September 2016, 21:00:31 UTC |
| 9fa557e | Brian Harrington | 04 September 2016, 01:22:54 UTC | add helper methods to normalize a query Adds helper functions to normalize a query to CNF or DNF. | 04 September 2016, 01:22:54 UTC |
| 4dff5ec | brharrington | 01 September 2016, 21:25:03 UTC | update wiki examples to use 1.4.7 (#432) | 01 September 2016, 21:25:03 UTC |
| ccb01d7 | brharrington | 26 August 2016, 03:57:53 UTC | fix NPE if keyString is null for subset of values (#429) If a subset of the time series passed to `GroupBy.eval` are missing keys, then a NullPointerException will result when sorting the groups. This change filters out the entries with missing keys and adds some additional test cases for group by. | 26 August 2016, 03:57:53 UTC |
| b5014bc | brharrington | 25 August 2016, 22:49:34 UTC | add vpc and subnet to default allowed-keys (#428) Updates the allowed `nf.` keys to include vpc and subnet. | 25 August 2016, 22:49:34 UTC |
| 5fea792 | brharrington | 25 August 2016, 03:58:39 UTC | add test case for ActorService (#427) Verifies that the actor is created on the expected path. See #417 for more details. | 25 August 2016, 03:58:39 UTC |
| 5f1b5a3 | brharrington | 25 August 2016, 03:42:58 UTC | fix scaladoc warnings (#426) | 25 August 2016, 03:42:58 UTC |
| 081e539 | brharrington | 25 August 2016, 03:42:10 UTC | fix warning in test case (#425) Fixes a compiler warning in ClientActorSuite: ``` Warning:(173, 15) non-variable type argument spray.http.HttpResponse in type pattern scala.util.Try[spray.http.HttpResponse] is unchecked since it is eliminated by erasure case t: Try[HttpResponse] => response = t.asInstanceOf[Try[HttpResponse]] ``` | 25 August 2016, 03:42:10 UTC |
| d299de0 | brharrington | 25 August 2016, 03:41:57 UTC | add lambda to default list of categories (#424) The conf file is already there, but it got left out of the main list. | 25 August 2016, 03:41:57 UTC |
| 5a5ff0d | brharrington | 24 August 2016, 04:34:29 UTC | fix batch that is written back (#423) Use the immutable copy rather than the mutable list for the batch that has already been cleared. | 24 August 2016, 04:34:29 UTC |
| d90a0fa | brharrington | 24 August 2016, 01:25:07 UTC | update modules to dedup (#422) Changes the guice modules to be final and implement equals and hashCode so that guice will dedup if it is installed multiple times. | 24 August 2016, 01:25:07 UTC |
| 304bc26 | brharrington | 23 August 2016, 22:33:32 UTC | add poller to collect CloudWatch metrics data (#421) Poller for collecting CloudWatch data and publishing it into Atlas. This is a restructured from the legacy db implementation used internally to wrap CloudWatch. Overall the polling approach is preferred because: * It makes CloudWatch data follow the same path as all other data. This means it is available for teeing to stream processing systems, long term storage in hive, etc. * Querying CloudWatch on-demand does not scale well and it fails frequently due to throttles. In this case there is a consistent load all the time, but we can tune that for the needed set and avoid the really high bursts needed for on-demand queries. * If we are lucky Amazon will support a way to get a stream of all CloudWatch metric data and we can switch to consuming that in the future. In the mean time we can make it look like a stream for most of our use-cases. | 23 August 2016, 22:33:32 UTC |
| c13e114 | brharrington | 23 August 2016, 03:14:58 UTC | fix content type for client actor (#420) Updates the content-type to be smile to match the payload. This was mistakenly set as the accept header earlier. | 23 August 2016, 03:14:58 UTC |
| 5d74f35 | brharrington | 18 August 2016, 13:26:34 UTC | manager for polling data (#418) Adds in the base manager that supervises a set of poller actors. This is used for collecting metrics data from various other services such as edda, healthcheck service, cloudwatch, etc. | 18 August 2016, 13:26:34 UTC |
| 51a898b | Michael Graff | 16 August 2016, 20:50:33 UTC | Merge pull request #417 from skandragon/fix-actor-path Set actor name rather than using a generated one. | 16 August 2016, 20:50:33 UTC |
| 351f98f | Michael Graff | 16 August 2016, 20:33:41 UTC | Set actor name rather than using a generated one. | 16 August 2016, 20:33:41 UTC |
| df4590e | brharrington | 13 August 2016, 14:27:30 UTC | make ending '/' optional in path pattern (#416) Updates the default path pattern to make the ending slash optional. If it is a simple actor path it is usually not present. Makes it match the setting we are typically using internally. | 13 August 2016, 14:27:30 UTC |
| 329c331 | brharrington | 13 August 2016, 14:17:09 UTC | look for routing config on loaded actors (#415) Looks for routing config on routed actors including the request-handler. | 13 August 2016, 14:17:09 UTC |
| 6e37d0a | brharrington | 13 August 2016, 14:16:15 UTC | use tag map for block store lookup (#414) The datapoint passed in computes the id on demand which can be slow. The tag map is immutable and has a cached hash code and is thus much cheaper for the initial lookup. | 13 August 2016, 14:16:15 UTC |
| 86b7be8 | brharrington | 10 August 2016, 22:55:12 UTC | more robust hash code function (#413) Changes the hash code function to be more similar to the MurmurHash3.unorderedHash. This is mostly paranoia that the simple additive method will be terrible for some data sets. There is an added test case that demonstrates a fairly bad case. | 10 August 2016, 22:55:12 UTC |
| e2eef31 | brharrington | 10 August 2016, 21:16:51 UTC | improve performance of SmallHashMap equality check (#412) Overrides equals for SmallHashMap to improve the performance. The two main changes are to use the cache hash code as a filter and simply compare the elements of the array. This does much better than the default inherited from GenMapLike that more or less does: ```scala this.forall { t => other.get(t._1) == Some(t._2) } ``` If the elements are not inserted in the same order, then they may be in a different order in the array due to collisions. However, others will be in the correct position and we can limit the lookups to ones where there is a mismatch. **Equal Maps** Checking where the maps being compared are equal: ``` Benchmark Mode Cnt Score Error Units currentEquals thrpt 10 47004215.059 ± 2291847.719 ops/s inheritedEquals thrpt 10 2134457.383 ± 111821.520 ops/s dataEquals thrpt 10 51326103.645 ± 758297.787 ops/s selfEquals thrpt 10 351279563.043 ± 18578115.816 ops/s ``` Note: * `currentEquals` - is SmallHashMap.equals, it does some sanity checks and then calls `dataEquals`. * `inheritedEquals` - is the GenMapLike.equals. * `dataEquals` - is the method used for comparing the array. * `selfEquals` - is comparing against the same instance of the map. Just sanity checks that the special case is fast. **Not Equal Maps** Checking maps that are almost equal except that one data value is different. Using the cached hash code as a pre filter is the biggest contributor: ``` currentEqualsNot thrpt 10 273893522.221 ± 7980051.600 ops/s inheritedEqualsNot thrpt 10 2341207.187 ± 206356.584 ops/s dataEqualsNot thrpt 10 32392263.165 ± 1289262.059 ops/s ``` **Equal with Different Order** Comparing equal maps, but constructed such that the order in the underlying array is different. This means we have to do more lookups rather than just traversing the array. For the examples though there are typically still quite a few entries that are in the same position for both maps. ``` currentEqualHashCodes thrpt 10 14601802.119 ± 360902.793 ops/s inheritedEqualHashCodes thrpt 10 483515.784 ± 10044.781 ops/s ``` | 10 August 2016, 21:16:51 UTC |
| b1d4184 | brharrington | 09 August 2016, 23:13:29 UTC | switch to MurmurHash3.arrayHash (#411) Switches the hashCode implementation for SmallHashMap to call `MurmurHash3.arrayHash` instead of relying on the implementation from the super class which calls `MurmurHash3.mapHash`. The array variant is much faster (3-4x) and greatly reduces allocations. | 09 August 2016, 23:13:29 UTC |
| 4b0ecca | brharrington | 09 August 2016, 20:26:31 UTC | turn off interning while building the index (#410) Adds an option to support turning off string interning while building the index. In many cases the strings will already have been interned while parsing the incoming payload and this additional step is redundant. | 09 August 2016, 20:26:31 UTC |
| 69d5b50 | brharrington | 09 August 2016, 18:23:58 UTC | reduce size of publish failure response (#409) Fixes a bug where the failure message was getting encoded as a json string inside the message of the json response. Also changes the behavior on failures to have the error count explicitly in the response and limit the number of specific failure messages to the first 5. This significantly reduces allocations and response sizes for payloads with many failures. | 09 August 2016, 18:23:58 UTC |
| 7d984bb | brharrington | 08 August 2016, 20:06:41 UTC | disable jackson buffer recycling (#408) The thread local buffer recycling in jackson seems to cause corruption in some cases. It is hard to reproduce in a controlled setting, but we see it pretty reliably on a big cluster. After disabling, the corruption is no longer being seen. First observed in jackson 2.7.x. Current tests do not show much performance impact for our use-cases. | 08 August 2016, 20:06:41 UTC |
| cfb302e | brharrington | 06 August 2016, 23:37:22 UTC | reduce allocations for computing ids (#407) | 06 August 2016, 23:37:22 UTC |
| e91049b | Brian Harrington | 06 August 2016, 23:12:23 UTC | add naive method to computeId benchmark | 06 August 2016, 23:12:23 UTC |
| 5811f66 | Brian Harrington | 06 August 2016, 22:49:51 UTC | reuse CharBuffer for encoding tag values Most of the allocations for computing ids were for creating the wrapped CharBuffer objects around the strings. This change allocates a fixed buffer and reuses it for each run. Performance is a little bit better: ``` Before: computeIdSmallTagMap thrpt 10 252703.940 ± 11424.615 ops/s After: computeIdSmallTagMap thrpt 10 262549.444 ± 5860.124 ops/s ``` Allocations are a lot better, for the benchmark it reduced by ~6.5GB. | 06 August 2016, 22:49:51 UTC |
| b9aab2c | Brian Harrington | 06 August 2016, 22:27:45 UTC | add test case for computing ids | 06 August 2016, 22:27:45 UTC |
| 152b0da | Brian Harrington | 06 August 2016, 21:44:01 UTC | add benchmark for computing ids | 06 August 2016, 21:44:01 UTC |
| c8ced70 | brharrington | 04 August 2016, 03:13:59 UTC | fix #405 possible contract error using desc order (#406) If the input set has many equal values it was possible to get an exception: ``` IllegalArgumentException: Comparison method violates its general contract! ``` This is because the descending order was simply inverting the ascending order comparator. With this change we sort in ascending order and simply reverse the list if it should be descending. | 04 August 2016, 03:13:59 UTC |
| d7eac10 | brharrington | 30 July 2016, 21:36:24 UTC | misc build cleanup (#404) | 30 July 2016, 21:36:24 UTC |
| cff596e | Brian Harrington | 30 July 2016, 21:26:18 UTC | fix unused import and parens warnings | 30 July 2016, 21:26:18 UTC |
| 136943f | Brian Harrington | 30 July 2016, 21:14:30 UTC | store creds before doing snapshot build Update snapshot make target to store the credentials first before doing the publish build. This is necessary because if the creds file does not exist when the build starts the bintray plugin will no longer be enabled. | 30 July 2016, 21:14:30 UTC |
| e2036dc | Brian Harrington | 30 July 2016, 21:11:55 UTC | fix storeBintrayCredentials task Fixes the storeBintrayCredentials to work when the bintray plugin is disabled. This allows us to generate the file needed by the bintray plugin using the environment variables available in the travis build. | 30 July 2016, 21:11:55 UTC |
| c44b865 | Brian Harrington | 30 July 2016, 20:53:04 UTC | reduce repetition across sub-projects | 30 July 2016, 20:53:04 UTC |
| b4a61dc | Brian Harrington | 30 July 2016, 20:40:04 UTC | disable bintray unless credentials are present Disables the bintray plugin unless the credentials file is present. This avoids many regular warnings such as: ``` Missing bintray credentials /Users/brharrington/.bintray/.credentials. Some bintray features depend on this. [warn] Credentials file /Users/brharrington/.bintray/.credentials does not exist ``` | 30 July 2016, 20:40:04 UTC |
| d1dd88e | Brian Harrington | 30 July 2016, 20:05:43 UTC | fix "wrong forward reference" warnings | 30 July 2016, 20:05:43 UTC |
| 9235297 | Brian Harrington | 30 July 2016, 17:28:24 UTC | add jsr305 dependency This gets rid of the warnings like: ``` Class javax.annotation.CheckForNull not found - continuing with a stub. ``` This dependency can be removed after we update to scala 2.12. For details see: https://issues.scala-lang.org/browse/SI-9673 | 30 July 2016, 17:28:24 UTC |
| b166659 | Brian Harrington | 30 July 2016, 17:04:48 UTC | switch from Build.scala to build.sbt Support for Build.scala is going away in future versions. See discussion for more details: https://groups.google.com/forum/#!searchin/sbt-dev/build.scala%7Csort:relevance/sbt-dev/S-pr2nZQ3KY/YhUtksKf5PoJ Migration was surprisingly painless: just rename, delete some of the scaffolding (imports and build object), and cleanup whitespace (`%s/^ //`). | 30 July 2016, 17:04:48 UTC |
| 117764b | Brian Harrington | 30 July 2016, 16:53:28 UTC | sbt-dependency-graph 0.8.2 | 30 July 2016, 16:53:28 UTC |
| 50ffd7b | Brian Harrington | 30 July 2016, 16:48:37 UTC | remove unnecessary whitespace for plugins | 30 July 2016, 16:48:37 UTC |
| 2894903 | brharrington | 30 July 2016, 02:33:02 UTC | make it easier to reuse access logging (#403) Splits out a helper class for doing the access logging based on the spray http classes. Makes it easier to reuse in other contexts, in particular, for use with clients. | 30 July 2016, 02:33:02 UTC |
| 0e5b51f | brharrington | 25 July 2016, 23:27:02 UTC | fix #401, add scala version to artifact id (#402) Makes it easier to automatically detect for tools that want to index scala libraries. | 25 July 2016, 23:27:02 UTC |
| 4c360d3 | brharrington | 21 July 2016, 16:27:13 UTC | move appender setup to log4j config (#400) For more details see: http://netflix.github.io/spectator/en/latest/ext/log4j2/ Which is a bit out of date, with the change below the ordering issues are no longer a problem when using the appender via the config: https://github.com/Netflix/spectator/commit/33825db0b96e3337f68b85e9ecbc78567071aecc#diff-9bc272482ec3aa21411ff595a1f4225f | 21 July 2016, 16:27:13 UTC |
| 4188e3b | brharrington | 21 July 2016, 16:14:25 UTC | remove diff helper, can be done with lib if needed (#399) | 21 July 2016, 16:14:25 UTC |
| 857888c | brharrington | 21 July 2016, 16:14:09 UTC | create module packages for akka/webapi (#398) Pulls the setup logic in standalone into modules that we can just include in other projects. We don't use the standalone jar internally it is just for simple testing. This change allows us to further reduce deviation from oss and internal. | 21 July 2016, 16:14:09 UTC |
| 59024bb | brharrington | 21 July 2016, 04:42:08 UTC | inject config and actor system (#396) Updates the akka helpers to be more DI friendly. In particular, it removes the dependency on `atlas-config` and injects the `Config` as well as the `ActorSystem`. A helper setting `atlas.akka.actors` is now available in the config to configure an actor to load into the system. | 21 July 2016, 04:42:08 UTC |
