Refactor Sniffer and make it testable #29638
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Introduced `Scheduler` abstraction to make `Sniffer` testable so that tasks scheduling is isolated and easily mockable. Also added a bunch of TODOs mainly around new tests that should be added soon-ish.
javanna
added
>enhancement
:Clients/Java Low Level REST Client
|
Pinging @elastic/es-core-infra |
javanna
commented
Apr 20, 2018
| * a new round is scheduled with delay 0, a following one is scheduled with sniffAfterFailure delay, after | ||
| * which the ordinary sniffing rounds get scheduled with sniffInterval delay | ||
| */ | ||
| public void testOnFailureSingleRound() throws Exception { |
There was a problem hiding this comment.
onFailure behaviour is really hard to test, too moving parts, so I went for testing different aspects of it with the following three test methods. Suggestions are welcome on whether we think coverage is good enough. Certainly better than before as it was 0% up until now :(
javanna
commented
Apr 20, 2018
| verify(executor, times(2)).shutdown(); | ||
| verify(executor, times(2)).awaitTermination(1000, TimeUnit.MILLISECONDS); | ||
| verifyNoMoreInteractions(executor); | ||
| } |
There was a problem hiding this comment.
I wonder whether it makes sense to add a test that uses the sniffer with on failure on against a proper http server. but that would add timing aspects which are not desirable. My concern is that we never test the sniffer in real-life. Maybe it would be nice to turn sniffing on when executing yaml tests?
There was a problem hiding this comment.
I think we can manage that in a followup.
nik9000
reviewed
Apr 20, 2018
| Future<?> nextFuture = task.schedule(delay); | ||
| Future<?> previousFuture = nextTask.getAndSet(nextFuture); | ||
| if (mustCancelNextRound) { | ||
| previousFuture.cancel(false); |
There was a problem hiding this comment.
When I was playing with cancellation as part of reindex I found that canceling a Runnable was sort of "best effort". If you make a test that calls sniffOnFailure a bunch of time really fast together I'll bet you get multiple rounds of sniffing in parallel.
There was a problem hiding this comment.
I guess it depends on whether the runnable has already started or not? That is what I've seen, but hard to test in real-life though from a unit test...
| logger.error("error while scheduling next sniffer task", e); | ||
| } | ||
| } | ||
| Future<?> schedule(long delay) { |
There was a problem hiding this comment.
Maybe drop this method and call scheduler.schedule(theTask, delay) all the places?
| } | ||
| final void sniff() throws IOException { | ||
| List<HttpHost> sniffedHosts = hostsSniffer.sniffHosts(); | ||
| logger.debug("sniffed hosts: " + sniffedHosts); |
There was a problem hiding this comment.
Probably best to wrap this if (logger.isDebugEnabled()) so we don't build the string if we don't need it.
| sniffer.sniff(); | ||
| if (hostsSniffer.failures.get() > failures) { | ||
| failures++; | ||
| fail("should have failed"); |
There was a problem hiding this comment.
Do you want to include something in this message so it is easier to debug?
| executor.shutdown(); | ||
| assertTrue(executor.awaitTermination(1000, TimeUnit.MILLISECONDS)); | ||
| } | ||
| int totalRuns = hostsSniffer.runs.get(); |
nik9000
approved these changes
Apr 20, 2018
There was a problem hiding this comment.
I do think it'd be nice to test the sniffer against a real server at some point too. Though I was using it in my work on the host selector so maybe that is good?
I think it'd be nice to, probably in a followup, write a test that triggers that cancellation many times to see if the cancellation sticks. I have a suspicion we'll need something stronger than this but this is better than we had!
|
@nik9000 I was not 100% satisfied with this so I got back to it and made some improvements. Your suspicions on the future cancellation were right, I did some digging around it and we cannot rely on it alone. With the latest commit I added more accurate tests and improved the implementation to behave the way we need it to. One change I made is that we skip onFailure rounds if at the time of the failure there's already another sniffing round in progress. Also, concurrent onFailure rounds don't end up cancelling each other like before, but rather go through or not depending on whether there's another ongoing round or not. Let me know what you think, sorry it took me forever to finish this up, it was quite tricky to get to the bottom of it. |
|
retest this please |
|
retest this please |
nik9000
reviewed
May 29, 2018
| import java.util.Map; | ||
| import org.apache.http.nio.entity.NStringEntity; |
There was a problem hiding this comment.
I think this change is worse? Maybe revert it?
| this.sniffAfterFailureDelayMillis = sniffAfterFailureDelay; | ||
| this.scheduler = scheduler; | ||
| /* | ||
| The first sniffing round is async, so this constructor returns before nextScheduledTask is assigned to a task. |
There was a problem hiding this comment.
Could you add the traditional * before each line of the comment?
| } | ||
|
|
||
| /** | ||
| * Triggers a new sniffing round and explicitly takes out the failed host provided as argument | ||
| * Triggers a new immediate sniffing round, which will schedule a new round in sniffAfterFailureDelayMillis |
There was a problem hiding this comment.
Maybe "reschedule sniffing to run as soon as possible if it isn't already running".
| * already cancelled or already completed). | ||
| */ | ||
| boolean cancelOrSKip() { | ||
| return future.cancel(false) && task.skip(); |
There was a problem hiding this comment.
I'd do
task.cancel(false);
return task.skip();
I don't think it is worth relying on cancel returning true in the case when we want to cancel. Maybe I'm overly paranoid but I don't trust it.
There was a problem hiding this comment.
And I might rename this method to skip.
There was a problem hiding this comment.
I agree! I have tests that verify this and I ran them many times but I am playing with fire here. After all, if cancel returns false whenever a task has already run, its corresponding state change will also fail. With this change all tests are still green so I think it's a good one to stay on the safe side.
| verify(executor, times(2)).shutdown(); | ||
| verify(executor, times(2)).awaitTermination(1000, TimeUnit.MILLISECONDS); | ||
| verifyNoMoreInteractions(executor); | ||
| } |
There was a problem hiding this comment.
I think we can manage that in a followup.
|
retest this please |
|
retest this please |
|
retest this please |
bizybot
added a commit
that referenced
this pull request
Jun 1, 2018
* Remove AllocatedPersistentTask.getState() (#30858) This commit removes the method AllocatedPersistentTask.getState() that exposes the internal state of an AllocatedPersistentTask and replaces it with a new isCompleted() method. Related to #29608. * Improve allocation-disabling instructions (#30248) Clarify the “one minute” in the instructions to disable the shard allocation when doing maintenance to say that it is configurable. * Replace several try-finally statements (#30880) This change replaces some existing try-finally statements that close resources in their finally block with the slightly shorter and safer try-with-resources pattern. * Move list tasks under Tasks namespace (#30906) Our API spec define the tasks API as e.g. tasks.list, meaning that they belong to their own namespace. This commit moves them from the cluster namespace to their own namespace. Relates to #29546 * Deprecate accepting malformed requests in stored script API (#28939) The stored scripts API today accepts malformed requests instead of throwing an exception. This PR deprecates accepting malformed put stored script requests (requests not using the official script format). Relates to #27612 * Remove log traces in AzureStorageServiceImpl and fix test (#30924) This commit removes some log traces in AzureStorageServiceImpl and also fixes the AzureStorageServiceTests so that is uses the real implementation to create Azure clients. * Fix IndexTemplateMetaData parsing from xContent (#30917) We failed to register "aliases" and "version" into the list of keywords in the IndexTemplateMetaData; then fail to parse the following index template. ``` { "aliases": {"log": {}}, "index_patterns": ["pattern-1"] } ``` This commit registers that missing keywords. * [DOCS] Reset edit links (#30909) * Limit the scope of BouncyCastle dependency (#30358) Limits the scope of the runtime dependency on BouncyCastle so that it can be eventually removed. * Splits functionality related to reading and generating certificates and keys in two utility classes so that reading certificates and keys doesn't require BouncyCastle. * Implements a class for parsing PEM Encoded key material (which also adds support for reading PKCS8 encoded encrypted private keys). * Removes BouncyCastle dependency for all of our test suites(except for the tests that explicitly test certificate generation) by using pre-generated keys/certificates/keystores. * Upgrade to Lucene-7.4-snapshot-1cbadda4d3 (#30928) This snapshot includes LUCENE-8328 which is needed to stabilize CCR builds. * Moved keyword tokenizer to analysis-common module (#30642) Relates to #23658 * [test] packaging test logging for suse distros * Fix location of AbstractHttpServerTransport (#30888) Currently AbstractHttpServerTransport is in a netty4 module. This is the incorrect location. This commit moves it out of netty4 module. Additionally, it moves unit tests that test AbstractHttpServerTransport logic to server. * [test] packaging: use shell when running commands (#30852) When subprocesses are started with ProcessBuilder, they're forked by the java process directly rather than from a shell, which can be surprising for our use case here in the packaging tests which is similar to scripting. This commit changes the tests to run their subprocess commands in a shell, using the bash -c <script> syntax for commands on linux and using the powershell.exe -Command <script> syntax for commands on windows. This syntax on windows is essentially what the tests were already doing. * [DOCS] Adds missing TLS settings for auditing (#30822) * stable filemode for zip distributions (#30854) Applies default file and directory permissions to zip distributions similar to how they're set for the tar distributions. Previously zip distributions would retain permissions they had on the build host's working tree, which could vary depending on its umask For #30799 * Minor clean-up in InternalRange. (#30886) * Make sure all instance variables are final. * Make generateKey a private static method, instead of protected. * Rename formatter -> format for consistency. * Serialize bucket keys as strings as opposed to optional strings. * Pull the stream serialization logic for buckets into the Bucket class. * [DOCS] Remove reference to platinum Docker image (#30916) * Use dedicated ML APIs in tests (#30941) ML has dedicated APIs for datafeeds and jobs yet base test classes and some tests were relying on the cluster state for this state. This commit removes this usage in favor of using the dedicated endpoints. * Update the version checks around range bucket keys, now that the change was backported. * [DOCS] Fix watcher file location * Rename methods in PersistentTasksService (#30837) This commit renames methods in the PersistentTasksService, to make obvious that the methods send requests in order to change the state of persistent tasks. Relates to #29608. * Rename index_prefix to index_prefixes (#30932) This commit also adds index_prefixes tests to TextFieldMapperTests to ensure that cloning and wire-serialization work correctly * Add missing_bucket option in the composite agg (#29465) This change adds a new option to the composite aggregation named `missing_bucket`. This option can be set by source and dictates whether documents without a value for the source should be ignored. When set to true, documents without a value for a field emits an explicit `null` value which is then added in the composite bucket. The `missing` option that allows to set an explicit value (instead of `null`) is deprecated in this change and will be removed in a follow up (only in 7.x). This commit also changes how the big arrays are allocated, instead of reserving the provided `size` for all sources they are created with a small intial size and they grow depending on the number of buckets created by the aggregation: Closes #29380 * Fsync state file before exposing it (#30929) With multiple data paths, we write the state files for index metadata to all data paths. We only properly fsync on the first location, though. For other locations, we possibly expose the file before its contents is properly fsynced. This can lead to situations where, after a crash, and where the first data path is not available anymore, ES will see a partially-written state file, preventing the node to start up. * Fix AliasMetaData parsing (#30866) AliasMetaData should be parsed more leniently so that the high-level REST client can support forward compatibility on it. This commit addresses this issue that was found as part of #28799 and adds dedicated XContent tests as well. * Cross Cluster Search: do not use dedicated masters as gateways (#30926) When we are connecting to a remote cluster we should never select dedicated master nodes as gateway nodes, or we will end up loading them with requests that should rather go to other type of nodes e.g. data nodes or coord_only nodes. This commit adds the selection based on the node role, to the existing selection based on version and potential node attributes. Closes #30687 * Fix missing option serialization after backport Relates #29465 * REST high-level client: add synced flush API (2) (#30650) Adds the synced flush API to the high level REST client. Relates to #27205. * Fix synced flush docs They had some copy and paste errors that failed the docs build. * Change ScriptException status to 400 (bad request) (#30861) Currently failures to compile a script usually lead to a ScriptException, which inherits the 500 INTERNAL_SERVER_ERROR from ElasticsearchException if it does not contain another root cause. Instead, this should be a 400 Bad Request error. This PR changes this more generally for script compilation errors by changing ScriptException to return 400 (bad request) as status code. Closes #12315 * Fix composite agg serialization error Fix serialization after backport Relates #29465 * Revert accidentally pushed changes in NoriAnalysisTests * SQL: Remove log4j and joda from JDBC dependencies (#30938) More cleanup of JDBC driver project Relates to #29856 * [DOCS] Fixes kibana security file location * [CI] Mute SamlAuthenticatorTests testIncorrectSigningKeyIsRejected Tracked by #30970 * Add Verify Repository High Level REST API (#30934) This commit adds Verify Repository, the associated docs and tests for the high level REST API client. A few small changes to the Verify Repository Response went into the commit as well. Relates #27205 * Add “took” timing info to response for _msearch/template API (#30961) Add “took” timing info to response for _msearch/template API Closes #30957 * Mute FlushIT tests We have identified the source causing these tests failed. This commit mutes them again until we have a proper fix. Relates #29392 * [CI] Mute HttpSecretsIntegrationTests#testWebhookAction test Tracked by #30094 * [Test] Prefer ArrayList over Vector (#30965) Replaces some occurances of Vector class with ArrayList in tests of the rank-eval module. * Fix license on AcitveDirectorySIDUtil (#30972) This code is from an Apache 2.0 licensed codebase and when we imported it into our codebase it carried the Apache 2.0 license as well. However, during the migration of the X-Pack codebase from the internal private repository to the elastic/elasticsearch repository, the migration tool mistakently changed the license on this source file from the Apache 2.0 license to the Elastic license. This commit addresses this mistake by reapplying the Apache 2.0 license. * [CI] Mute Ml rolling upgrade tests Tracked by #30982 * Make AllocatedPersistentTask.isCompleted() protected (#30949) This commit changes the isCompleted() method to be protected so that classes that extends AllocatedPersistentTask can use it. Related to #30858 * [CI] Mute Ml rolling upgrade test for mixed cluster too It can fail in either the mixed cluster or the upgraded cluster, so it needs to be muted in both. Tracked by #30982 * [Docs] Fix typo in Min Aggregation reference (#30899) * Refactor Sniffer and make it testable (#29638) This commit reworks the Sniffer component to simplify it and make it possible to test it. In particular, it no longer takes out the host that failed when sniffing on failure, but rather relies on whatever the cluster returns. This is the result of some valid comments from #27985. Taking out one single host is too naive, hard to test and debug. A new Scheduler abstraction is introduced to abstract the tasks scheduling away and make it possible to plug in any test implementation and take out timing aspects when testing. Concurrency aspects have also been improved, synchronized methods are no longer required. At the same time, we were able to take #27697 and #25701 into account and fix them, especially now that we can more easily add tests. Last but not least, unit tests are added for the Sniffer component, long overdue. Closes #27697 Closes #25701 * Deprecates indexing and querying a context completion field without context (#30712) This change deprecates completion queries and documents without context that target a context enabled completion field. Querying without context degrades the search performance considerably (even when the number of indexed contexts is low). This commit targets master but the deprecation will take place in 6.x and the functionality will be removed in 7 in a follow up. Closes #29222 * Core: Remove RequestBuilder from Action (#30966) This commit removes the RequestBuilder generic type from Action. It was needed to be used by the newRequest method, which in turn was used by client.prepareExecute. Both of these methods are now removed, along with the existing users of prepareExecute constructing the appropriate builder directly. * Ensure intended key is selected in SamlAuthenticatorTests (#30993) * Ensure that a purposefully wrong key is used Uses a specific keypair for tests that require a purposefully wrong keypair instead of selecting one randomly from the same pull from which the correct one is selected. Entropy is low because of the small space and the same key can be randomly selected as both the correct one and the wrong one, causing the tests to fail. The purposefully wrong key is also used in testSigningKeyIsReloadedForEachRequest and needs to be cleaned up afterwards so the rest of the tests don't use that for signing. Resolves #30970 * [DOCS] Update readme for testing x-pack code snippets (#30696) * Remove version read/write logic in Verify Response (#30879) Since master will always communicate with a >=6.4 node, the logic for checking if the node is 6.4 and conditionally reading and writing based on that can be removed from master. This logic will stay in 6.x as it is the bridge to the cleaner response in master. This also unmutes the failing test due to this bwc change. Closes #30807 * HLRest: Allow caller to set per request options (#30490) This modifies the high level rest client to allow calling code to customize per request options for the bulk API. You do the actual customization by passing a `RequestOptions` object to the API call which is set on the `Request` that is generated by the high level client. It also makes the `RequestOptions` a thing in the low level rest client. For now that just means you use it to customize the headers and the `httpAsyncResponseConsumerFactory` and we'll add node selectors and per request timeouts in a follow up. I only implemented this on the bulk API because it is the first one in the list alphabetically and I wanted to keep the change small enough to review. I'll convert the remaining APIs in a followup. * [DOCS] Clarify not all PKCS12 usable as truststores (#30750) Although elasticsearch-certutil generates PKCS#12 files which are usable as both keystore and truststore this is uncommon in practice. Settle these expectations for the users following our security guides. * Transport client: Don't validate node in handshake (#30737) This is related to #30141. Right now in the transport client we open a temporary node connection and take the node information. This node information is used to open a permanent connection that is used for the client. However, we continue to use the configured transport address. If the configured transport address is a load balancer, you might connect to a different node for the permanent connection. This causes the handshake validation to fail. This commit removes the handshake validation for the transport client when it simple node sample mode. * Remove unused query methods from MappedFieldType. (#30987) * Remove MappedFieldType#nullValueQuery, as it is now unused. * Remove MappedFieldType#queryStringTermQuery, as it is never overridden. * Reuse expiration date of trial licenses (#30950) * Retain the expiryDate for trial licenses While updating the license signature to the new license spec retain the trial license expiration date to that of the existing license. Resolves #30882 * Watcher: Give test a little more time Changes watcher's integration tests to wait 30 seconds when starting watcher rather than 10 seconds because this build failed when starting took 12 seconds: https://elasticsearch-ci.elastic.co/job/elastic+elasticsearch+6.3+periodic/222/console
javanna
added a commit
that referenced
this pull request
Jun 1, 2018
This commit reworks the Sniffer component to simplify it and make it possible to test it. In particular, it no longer takes out the host that failed when sniffing on failure, but rather relies on whatever the cluster returns. This is the result of some valid comments from #27985. Taking out one single host is too naive, hard to test and debug. A new Scheduler abstraction is introduced to abstract the tasks scheduling away and make it possible to plug in any test implementation and take out timing aspects when testing. Concurrency aspects have also been improved, synchronized methods are no longer required. At the same time, we were able to take #27697 and #25701 into account and fix them, especially now that we can more easily add tests. Last but not least, unit tests are added for the Sniffer component, long overdue. Closes #27697 Closes #25701
dnhatn
added a commit
that referenced
this pull request
Jun 2, 2018
* 6.x: Adjust BWC version on client features Introduce client feature tracking (#31020) [DOCS] Make geoshape docs less memory hungry (#31014) Fix handling of percent-encoded spaces in Windows batch files (#31034) [Docs] Fix a typo in Create Index naming limitation (#30891) REST high-level client: add delete ingest pipeline API (#30865) Ensure that index_prefixes settings cannot be changed (#30967) REST high-level client: add get ingest pipeline API (#30847) Cross Cluster Search: preserve remote status code (#30976) High-level client: list tasks failure to not lose nodeId (#31001) Refactor Sniffer and make it testable (#29638) [ML][TEST] Fix bucket count assertion in all tests in ModelPlotsIT (#31026) Add an option to split keyword field on whitespace at query time (#30691) Allow rollup job creation only if cluster is x-pack ready (#30963) Fix interoperability with < 6.3 transport clients (#30971) [Tests] Fix alias names in PutIndexTemplateRequestTests (#30960) [DOCS] Fixes links (#31011) Watcher: Give test a little more time
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR reworks the
Sniffercomponent to simplify it and make it possible to test it.In particular, it no longer takes out the host that failed when sniffing on failure, but rather relies on whatever the cluster returns. This is the result of some valid comments from #27985. Taking out one single host is too naive, hard to test and debug.
A new
Schedulerabstraction is introduced to abstract the tasks scheduling away and make it possible to plug in any test implementation and take out timing aspects when testing.Concurrency aspects have also been improved, synchronized methods are no longer required. At the same time, we were able to take #27697 and #25701 into account and fix them, especially now that we can more easily add tests.
Last but not least, good news is that we now have unit tests for the
Sniffercomponent, long overdue.Closes #27697
Closes #25701