Prototyping applications with heroku and elasticsearch

Prototyping Apps with Elasticsearch and Heroku October 29th 2015

Protofy Martin and Mike: Prototyping with Heroku and Elasticsearch

Protofy builds Prototypes.Salad-Delivery-Service: From idea to first shipped salad -> 4days.

- Validation of the concept in a prototype - Live-Launch in March 2015- Continuous prototyping while quickly growing key KPIs- Seed-Financing in October, November

Protofy builds Prototypes.Voicefile transcoding and indexing for callcenters

client-a.com:- Validation of the concept in a prototype - Excessive usage of Elasticsearch as main database=> THIS is the FIRST project we will show deeper.

Protofy builds Prototypes.Automatic news aggregation by given list of keywords and synonyms.

client-b.com:- Validation of the concept in a prototype - Excessive usage of Elasticsearch to filter feed items and merge them=> THIS is the SECOND project we will show deeper.

And big infrastructure.Education Community Framework with Log-Everything strategy.

PokerStrategy.com:- 7 Mio members (2007-2013)- up to 1 Billion pageviews/year- sold in mid 2013After that 2 companies have been found: DECK36 and Feelgood. Both merged in early 2015 to PROTOFY.

Prototyping with Heroku.Focus on the app.

Heroku: Platform as a service.Prebuilt VMs for different programming languages- deployment via git- customizable with build-packs and add-ons- easily scalable- full logging of each part of the app and process- releases: Easy rollback on errors- heroku toolbelt to support local execution

Heroku: Prepare your app- Apps using other infrastructural services like MongoDB or Redis need to be aware of

environment variables

For example: Elasticsearch-service BONSAI provides: BONSAI_URL=https://user:pw@host.bonsai.io

- Use environment variables for everthing dependent.

- In general relay on the (Attention. Buzzword.) http://12factor.net/ methodology.

Buzzwording: 12 FactorsI. Codebase One codebase tracked in revision control, many deploysII. Dependencies Explicitly declare and isolate dependenciesIII. Config Store config in the environmentIV. Backing Services Treat backing services as attached resourcesV. Build, release, run Strictly separate build and run stagesVI. Processes Execute the app as one or more stateless processes

VII. Port binding Export services via port bindingVIII. Concurrency Scale out via the process modelIX. Disposability Maximize robustness with fast startup and graceful shutdownX. Dev/prod parity Keep development, staging, and production as similar as possibleXI. Logs Treat logs as event streamsXII. Admin processes Run admin/management tasks as one-off processes

Heroku: add-ons- Logentries

- NewRelic

- Bonsai-Elasticsearch

- MongoLabs

- Scheduler

- SSL

TIP: Care about backups! Even if they promise to do.

Heroku: Test before deployCONTINUOUS DEPLOYMENT using codehip.io (or others)

git

bitbucket

codeship test heroku

Terraforming.Infrastructure as code.

Heroku: Infrastructure as a service.- deployment via git

- a lot of add ons

- individual scaling of parts of the app

- process isolation

- full logging of each part of the app and process

- easy-to-use command line tools

- supports several languages (NodeJS, PHP, Rails, etc.

- releases: Easy rollback on errors.

- heroku toolbelt to support local execution like it would be on heroku with ForemanTERRAFORMBuild,'Combine,'and'Launch'Infrastructure

from automatic provisioning of servers ...

Configuration as Code

Infrastructure as Code… to automatic provisioning of services.

Why do we need that?As with Configuration Management:

-Replace “click-paths” with source code

-Reproducible Environment

-Versioning in SCM

-Specification and Documentation

What does it do?Configuration Language for Services

Actions:

-Plan

-Apply

-Refresh

-Destroy

What does it manage?Providers:

- Google Cloud- AWS- Azure- Heroku- DNSMadeEasy- …

Resources:

- aws_instance- aws_vpc- azure_instance- heroku_app- …

Provisioners:

- chef- file- exec

Example (part 1)###  AWS  Setupprovider  "aws"  {    access_key  =  "${var.aws_access_key}"    secret_key  =  "${var.aws_secret_key}"    region          =  "${var.aws_region}"}

#  Queue  between  importer  and  analyzerresource  "aws_sqs_queue"  "importqueue"  {    name  =  "${var.app_name}-­‐${var.app_env}-­‐import-­‐queue"}

resource  "aws_s3_bucket"  "importdisk"  {    bucket  =  "${var.app_name}-­‐${var.app_env}-­‐app-­‐importer"    acl        =  "private"}  

Example (part 2)###  Heroku  Setupprovider  "heroku"  {...}

#  App  EntityImporterresource  "heroku_app"  "importer"  {    name  =  "${var.app_name}-­‐${var.app_env}-­‐importer"    config_vars  {        SQS_REGION        =  "${var.aws_region}"        SQS_QUEUE_URL  =  "${aws_sqs_queue.importqueue.id}"        S3_BUCKET          =  "${aws_s3_bucket.importdisk.id}"        NODE_ENV            =  "${var.app_env}"    }}

resource  "heroku_addon"  "mongolab"  {    app    =  "${heroku_app.importer.name}"    plan  =  "mongolab:sandbox"}

Graph

Live-Demo

Launch application

terraform plan

terraform apply

terraform show

terraform destroy

Comparable Software– AWS CloudFormation

– HEAT, OpenStack orchestration

– boto, Python AWS library

– fog, Ruby cloud abstraction library

Problems– Version 0.6

– Still a few bugs

– Provider coverage

– Modules too simple

– Lacking syntactic sugar

Software as a service. Elasticsearch.

Elasticsearch ServiceLet other do the dirty work.

- Relatively complex setup with Shards and Replicas is maintained by specialists.

- Backups and version upgrades are done by these specialists, too.

- But 1: If version upgrades are announced YOU have to take action.

- But 2: Backups SHOULD be done by the specialists. In some cases they cannot provide consistent backups and that can lead to data loss. => Care about them yourself.

- But 3: If you need plugins: in the non-dedicated plans you cannot install them.

Decide well if or if not to use a service or do it yourself.

The projects. Short overview.

client-a.comVoicefile transcoding and indexing for callcenters

- Make telephone calls searchable- AccessManagement per Callcenter and Customer- Fast responses and results- Mobile- Be able to white label

Callcenter

client-b.comAutomatic content aggregation based on editor’s given input.

- Have up to 250.000 news items/day related to a topic from blogs, twitter/facebook/instagram and other configurable sources.- Have automatic sorting and merging of similar items into stories.- Be nearly realtime- Make editing of main stories possible- Mobile first

Elasticsearch. Some magic for the app.

Elasticsearch: GeneralSearch server based on Lucene. Providing RESTful web interface for JSON documents.

- Near real-time search.

- Sophisticated mapping configuration options. => Where the magic comes from.

- Highly scaleable and available.

- Conflict management with optimistic version control to avoid dataloss during concurrent write operations.

- Supporting Plugins for different areas (Like Filters, Queries, Analyzers etc.)

Elasticsearch: client-a.comElasticsearch as main database

- Provide several states of a document based on the state of processing. Always findable and restricted by ACLs

How to reach that?

Elasticsearch: client-a.comRestrict access by ACLs for „normal“ search

1. Check if user is allowed to access groups trying to request documents for.

2. If yes: Build query with filter restricting results to customers and callcenters based on ACL.

Find documentsrelated to callcenter1and callcenter2

{ "query": { "filtered": { "query": { "query_string": { "default_operator": "AND", "minimum_should_match": "55%", "auto_generate_phrase_queries": true, "phrase_slop": 3, "fields": [ "tags^2", "transscript" ], "query": "*" } }, "filter": { "bool": { "must": [ { "range": { "lastUpdated": { "gte": "now-24h", "lte": "2015-10-25T17:34:24+00:00" } } }, { "range": { "lastUpdated": { "gte": "2015-08-30T21:04:08+00:00" } } }, { "bool": { "should": [ { "term": { "source.callcenter": "callcenter1" } }, { "term": { "source.callcenter": "callcenter2" } } ] } } ] } } } }}

{ "query": { "filtered": { "query": { "query_string": { "default_operator": "AND", "minimum_should_match": "55%", "auto_generate_phrase_queries": true, "phrase_slop": 3, "fields": [ "tags^2", "transscript.texts.contents" ], "query": "*" } }, "filter": { "bool": { "must": [ { "range": { "lastUpdated": { "gte": "now-24h", "lte": "2015-10-25T17:34:24+00:00" } } }, { "range": { "lastUpdated": { "gte": "2015-08-30T21:04:08+00:00" } }

}, { "bool": { "should": [ { "term": { "source.callcenter": "callcenter1" } }, { "term": { "source.callcenter": "callcenter2" } } ] } } ] } } } }}

Elasticsearch: client-a.comRestrict access for suggests

1. Completion suggests are special handling for really fast autocompletionhttps://www.elastic.co/guide/en/elasticsearch/reference/current/search-suggesters-completion.html

2. How to make suggestions context (ACL) aware?

{ "body": { "suggest": { "text": "Agent 007", "completion": { "field": "agent.suggest", "size": 20, "fuzzy": false, "context": { "customer": [ "customer1", "customer4" ], "callcenter": [ "callcenter1", "callcenter2" ] } } } }}

Find suggestions related to context{ "agent": { "type": "multi_field", "fields": { "agent": { "type": "string", "copy_to": "autocompletion" }, "autocompletion": { "type": "string", "index_analyzer": "edgeNGram_analyzer_suggest" }, "suggest": { "type": "completion", "index_analyzer": "nGram_analyzer_suggest2", "search_analyzer": "whitespace_analyzer", "max_input_length": 20, "context": { "customer": { "type": "category", "path": "source.customer_lowercase" }, "callcenter": { "type": "category", "path": "source.callcenter_lowercase" } } } }, "include_in_all": false } }

Query Mapping

Elasticsearch: client-b.comElasticsearch to find similar articles and match them to stories

- Index stories and automatically find entities within the articles text

- Match similar articles to at least one story (based on entities) and context

How to do that?

Elasticsearch: client-b.comEntity matching by list of keep words and aliases

1. Create a list of synonyms and keep words to be used in filters.https://www.elastic.co/guide/en/elasticsearch/reference/current/analysis-keep-words-tokenfilter.htmlhttps://www.elastic.co/guide/en/elasticsearch/reference/current/analysis-synonym-tokenfilter.html

2. Index document 1st time to find entities based on keep words and synonyms.

3. Take document enriched with entities to build a query from it to match against the set of documents to find similar articles.

4. Combine them to a story.

Setting for matching entities

"settings": { "analysis": { "filter": {}, "analyzer": {

"entity_analyzer": { "tokenizer": "whitespace", "filter": [ "german_stop", "shingle", "entity_synonym", "shingle", "entity_keepwords" ] } } }},

Live-Demo

Check how entities are matched in a text

1. ./load_entities_list

2. curl -XGET "localhost:9200/talk/_analyze?analyzer=entity_analyzer&pretty=true" -d "Text"

=> Document is indexed with found entities on indexing time. Analyzing process is like operating on a stream.

Martin Schütte and Mike Lohmann

ProtofyKaiser-Wilhelm-Straße 8520355 Hamburg

martin@protofy.commike@protofy.com