Testing and Releasing Elasticsearch and the Elastic Stack

A presentation at Agile & Automation Days in October 2019 in Gdańsk, Poland by Alexander Reelsen

Testing & Releasing Elasticsearch & the Elastic Stack Alexander Reelsen @spinscale alex@elastic.co

Today’s goal Transfer Elasticsearch testing & automation knowledge into your own project

Elasticsearch in 10 seconds Search Engine (FTS, Analytics, Geo), real-time Distributed, scalable, highly available, resilient Interface: HTTP & JSON Heart of the Elastic Stack

Agenda Unit Testing Integration Testing Package Testing BWC Testing Documentation Testing CI: OS/JVM Testing Releasing

Random methods public void testRandomStringLength() { String randomString = randomAlphaOfLength(10); assertThat(randomString.length(), is(10)); boolean randomBoolean = randomBoolean(); assertThat(randomBoolean, is(anyOf(is(true), is(false)))); final Integer randomInt = randomFrom(1, 22, 11, 41, 58); assertThat(randomInt, is(allOf(greaterThanOrEqualTo(1), lessThanOrEqualTo(58)))); final int randomIntBetween = randomIntBetween(1, 10); assertThat(randomIntBetween, is(allOf(greaterThanOrEqualTo(1), lessThanOrEqualTo(10)))); final long nonNegativeLong = randomNonNegativeLong(); assertThat(nonNegativeLong, is(greaterThanOrEqualTo(0L))); final ZoneId zoneId = randomZone(); assertThat(ZoneId.getAvailableZoneIds(), hasItem(zoneId.toString())); final Locale locale = randomLocale(random()); assertThat(Arrays.asList(Locale.getAvailableLocales()), hasItem(locale)); }

Random methods public void testRandomStringLength() { String randomString = randomAlphaOfLength(10); assertThat(randomString.length(), is(10)); }

Random methods public void testRandomStringLength() { String randomString = randomAlphaOfLength(10); assertThat(randomString.length(), is(10)); } public void testRandomStringLength() { String randomString = randomAlphaOfLength(10); assertThat(randomString.length(), is(Integer.MAX_VALUE)); }

Randomization Example private int add (int a, int b) { assert a > 0; assert b > 0; return a + b; }

Randomization Example private int add (int a, int b) { assert a > 0; assert b > 0; return a + b; } public void testAdd() { assertThat(add(1, 1), is(2)); assertThat(add(2, 2), is(4)); }

Randomization Example private int add (int a, int b) { assert a > 0; assert b > 0; return a + b; } public void testAdd() { final int a = randomIntBetween(0, Integer.MAX_VALUE); final int b = randomIntBetween(0, Integer.MAX_VALUE); assertThat(add(a, b), allOf(greaterThanOrEqualTo(a), greaterThanOrEqualTo(b))); }

Randomization Example private int add (int a, int b) { assert a > 0; assert b > 0; return a + b; } @Repeat(iterations = 1000) public void testAdd() { final int a = randomIntBetween(0, Integer.MAX_VALUE); final int b = randomIntBetween(0, Integer.MAX_VALUE); assertThat(add(a, b), allOf(greaterThanOrEqualTo(a), greaterThanOrEqualTo(b))); }

Randomization Example private int add (int a, int b) { assert a > 0; assert b > 0; return a + b; } @Seed(“EC29A319F9128F9A”) public void testAdd() { final int a = randomIntBetween(0, Integer.MAX_VALUE); final int b = randomIntBetween(0, Integer.MAX_VALUE); assertThat(add(a, b), allOf(greaterThanOrEqualTo(a), greaterThanOrEqualTo(b))); }

ESTestCase goodies Thread leak detection Change log level for single test/package Checks for excessive sysout/err writes Test method order is randomized Test class order is randomized Deprecation logging checks No inflight search contexts

HTTP Integration tests Run tests against Elasticsearch HTTP interface Tests are written in YAML API definition is written in JSON ESClientYamlSuiteTestCase to extend All official clients are using the same spec to guarantee feature parity

API definition { } “get”:{ “documentation”:{ “url”:”https://www.elastic.co/guide/en/elasticsearch/reference/master/docs-get.html”, “description”:”Returns a document.” }, “stability”:”stable”, “url”:{ “paths”:[ { “path”:”/{index}/_doc/{id}”, “methods”:[ “GET” ], “parts”:{ “id”:{ “type”:”string”, “description”:”The document ID” }, “index”:{ “type”:”string”, “description”:”The name of the index” } } } ] }, “params”:{ “preference”:{ “type”:”string”, “description”:”Specify the node or shard the operation should be performed on (default: random)” } … } }

Test definition —”Basic”: - do: index: index: test_1 id: 中⽂文 body: { “foo”: “Hello: 中⽂文” } - do: get: index: test_1 id: 中⽂文 - match: { _index: - match: { _id: - match: { _source: test_1 } 中⽂文 } { foo: “Hello: 中⽂文” } }

More integration testing Test against kerberos/samba/openldap/shibboleth-idp using docker and docker-compose S3 integration testing using minio (via dockercompose) Testing the docker image via docker-compose

Elasticsearch: A lot of flavours tar.gz, zip, deb, rpm apt / yum repositories docker images osx brew tap Windows MSI helm chart, ECK (K8s operator) puppet, ansible, chef

Testing packages Use vagrant to download/install distributions ubuntu-1604, ubuntu-1804, debian-8, debian-9, centos-6, centos-7, oel-6, oel-7, fedora-28, fedora-29, opensuse-42, sles-12, rhel-8, windows-2012r2, windows-2016

Testing packages bats is used for testing bats: bash automated testing package installation/start/stop/uninstall Long term plan: Moving away from bats to java tests https://github.com/elastic/elasticsearch/tree/master/qa/os

BWC Testing You get an upgrade, everyone gets an upgrade!

BWC Testing Upgrades must be supported and tested! Full cluster restart Rolling restarts between minor versions Rolling restarts from the latest old major version to the current major version, i.e. 6.8.4 to 7.4.1

Rolling upgrade tests upgrade 7.4.x 6.8.x 6.8.x

Rolling upgrade tests upgrade 7.4.x 7.4.x 6.8.x

Rolling upgrade tests upgrade 7.4.x 7.4.x 7.4.x

Documentation Testing Y’all doing this, right?

Documentation Testing All documentation is written in asciidoc Fire up a test cluster Extract snippets from documentation Run snippets against cluster Optionally: check response contents

Sample snippet [source,console] ————————————————————————-GET twitter/_doc/0 ————————————————————————-// TEST[setup:twitter]

Sample snippet [source,console-result] ————————————————————————-{ “_index” : “twitter”, “_id” : “0”, “_version” : 1, “_seq_no” : 10, “_primary_term” : 1, “found”: true, “_source” : { “user” : “kimchy”, “date” : “2009-11-15T14:12:12”, “likes”: 0, “message” : “trying out Elasticsearch” } } ————————————————————————-// TESTRESPONSE[s/”_seq_no” : \d+/”_seq_no” : $body._seq_no/ s/”_primary_term” : 1/”_primary_term” : $body._primary_term/]

build.gradle FTW One build system to rule them all

All you need is gradle Elasticsearch uses gradle for everything All tests can be triggered via gradle Try to reduce dependency external tools

Github PR testing Every commit to an open PR triggers a build Trigger test run via github Trigger merge and test run via github

CI stats All active branches are tested Every pull request commit gets tested several thousand builds per day

Links https://elasticsearch-ci.elastic.co https://kibana-ci.elastic.co https://logstash-ci.elastic.co https://beats-ci.elastic.co https://benchmarks.elastic.co/ https://github.com/elastic/rally

Embrace the difference Every product in the stack will test differently Java, JRuby, Ruby, Go, Javascript, TypeScript, Scala, .NET, perl, PHP, python, bash, Groovy, C++, Haskell, Clojure JSON, YAML, terraform, hcl

Everyone be ready! Releases are time based Every product’s branch should be ready Build of release candidate artifacts Testing of these artifacts Repeat until done across products: Fix all blockers Releasing of those artifacts

Release manager A person responsible for the release Triggers the build of all artifacts Performs follow up tasks for each team after a release

Trust the process! Staged artifacts are the final release artifacts, no rebuilds! Push artifacts to their final destinations Official links, OSS sonatype, rubygems, docker images, package repositories

Trust the process! +200 artifacts are promoted to release bucket Using Tekton Pipelines instead of single process started on release manager machine promoting artifacts: 4x faster 3rd party promotion: 6x faster > 40 pods are doing a single release

More Work begins after releasing website communication release blog posts ensuring everything has been published

Products have a lifecycle Stop supporting old products Be consistent with support guarantees Provide maintenance policy Distinguish between end of life & fixing bugs

Summary Automate early, automate often Automate your tests, all of them Manual testing will still find bugs Automate your releases Ensure everyone can do a release Non-technical aspects are much harder to automate

Thanks for listening! Questions? Alexander Reelsen @spinscale alex@elastic.co

Alexander Reelsen
@spinscale

1 / 85

Elasticsearch is well known piece of software. This talk explains the different levels of testing along with packaging and releasing as part of the Elastic Stack.

Testing a well known software like Elasticsearch is not too different to any other software. In this session we will peak into the different testing strategies for unit and integration tests including randomized testing, how we leverage gradle, how we do packaging tests, how we test the REST layer, what our CI infrastructure and tooling around that looks like and finally what happens in order to release Elasticsearch and other parts of the Elastic Stack.

Be prepared to see a fair share of source code and snippets.

Testing and Releasing Elasticsearch and the Elastic Stack

Link for this presentation:

HTML code for embedding:

Share on social media: