Learning Elasticsearch with PowerShell

Reframing Elasticsearch

Before I talk about any topic, I like to reframe it away from the marketing, lame "Hello World" examples, and your own personal echo chamber. So, I'm going to begin my talking about what Elasticsearch ("ES") is. I do not consider it to be a "search engine", so... pay attention.

I'm not big on marketing introductions. They are usually filled with non-technical pseudo-truths and gibberish worthy of the "As seen on TV" warning label. So, what is Elasticsearch? Marketing says it's a search system. People who have used it say it's a hardcore Aristotelian database that makes for a fine primary datastore as well as for a fine search engine.

One of the major differences with MongoDB is that Elastic is more explicit about its indexing. Every database does indexing and everything has a schema. Have a data-structure? You have an index. Have fields? At a minimum, you have an implicit schema. This is what makes an Aristotelian-system Aristotelian.

See my video on Platonic and Aristotelian Data Philosophies for more information on why "NoSQL" is a modern marketing fiction similar to "AJAX".

More particularly, Elasticsearch has a strong focus on the schema than MongoDB.

You might find people say that Elastic is schemaless. These people have neither read nor peeked at the docs. Elastic is very explicit about it's indexes. Sometimes you'll hear that it's schemaless because it uses Lucene (the engine deep deep deep down that does the searching), which is schemaless. That's stupid. Lucene uses straight bytes and Elastic adds the schema on top of it. Your file system uses bytes and SQL Server adds a schema on top of it. Just because your file system uses bytes, not a schema, this doesn't mean that SQL Server "doesn't have a schema" because it uses MDF files on a a file system using bytes. SQL Server has a schema. Elastic has a schema. It might be implicit, but it has a schema. Even if you never create one, there is a schema. Elastic is explicit about having a schema.

Beyond the Aristotelian nature though, like MongoDB, ES is an object database (or "document database" as the marketing guys call it, but, unless we wildly redefine what "document" means, I've never stored a document in MongoDB / ES!) You work with objects anyway, why in the world are you translating them to and from relational patterns? Square peg -> round role. ES and MongoDB are perfect for systems that rely heavily on objects. JSON in / JSON out. No translation required. This one feature here is why many ditch SQL Server for ES or MongoDB. Translation for the sake of translation is insane.

Yet another difference is in terms of access: because MongoDB uses TCP and ES uses HTTP, for all practical purposes, MongoDB requires a library, but an ES library is redundant. When something is as dynamic as ES, using strongly-typed objects in .NET makes .NET fodder for ridicule. .NET developers with an old school mindset (read: inability to paradigm shift) in particular have an unhealthy attachment to make everything they touch into some strongly-typed abstraction layer (though not .NET, the blasphemy known as Mongoose comes to mind.) This is also everything right about the dynamic keyword in C#. It's good to see C# finally get close to the modern world.

The stereotype of a .NET developer (for non-.NET developers) is this: the dude is given a perfectly good HTTP endpoint, he then proceeds to express his misguided cleverness by destroying the open nature of the endpoint by wrapping it in yet another API you have to learn. You can simply look at Nuget to see the massive number of pointless abstraction layers everyone feels the need to dump onto the world. To make matters worst, when you want to make a simple call, you're forced to defeat the entire point of the clean RESTful API by using the pointless abstraction layer. Fail. Abstraction layers are fun to write to learn an API, but... dude... keep them to yourself. Go simple. Go raw. Nobody wants abstraction layer complexity analogous to WebForms; there's a reason Web API is so popular now. Don't ruin it. This lesson is for learning, not to create yet another pointless abstraction layer. Live and love dynamic programming. Welcome to the modern world.

This is most likely why REST is little more than a colloquialism. People who attempt to use hypermedia (a requirement for REST) either find it pointlessly complicated or impossible to wrap. REST is dead until something like the HAL spec gains widespread acceptance; for now, we use proper term "REST" to mean any HTTP verb-driven, resource-based API. ES is not REST, it's "REST" only in this latter sense. We usually refer to this, ironically, as "RESTful".

Game Plan

This leads to the point of this entire document: you will learn to use ES from first-principles; you will be able to write queries as wrapped or unwrapped as you want.

My game plan here will seem absolutely backward:

  • Abstract some lower-level ES functionality with PowerShell using search as an example.
  • Discuss ES theory, index setup, and data inserting.

Given that setup is a one-time thing and day-to-day stuff is... uh... daily, the day-to-day stuff comes first. The first-things-first fallacy can die.

I chose PowerShell for this because it's almost guaranteed to be something you've not seen before-- and it offends .NET developers, iOS developers, and Python developers equally. In practice, I use raw curl to manage my ES clusters and Python for mass importing (.NET has far too much bloat for one-off tools!)

As I demonstrate using ES from PowerShell, I will give commentary on what I'm doing with ES. You should be able to learn both ES and some practical, advanced PowerShell. If you don't care about PowerShell... even better! One reason I chose PowerShell was to make sure you focus on the underlying concepts.

There's a lot of really horrible PowerShell out there. I'm not part of the VB-style / Cmdlet / horribly-tedious-and-tiring-long-command-name PowerShell weirdness. If you insist on writing out Invoke-WebRequest instead of simply using wget, but use int and long instead of Int32 and Int64, then you have a ridiculous inconsistency to work out. Also, you're part of the reason PowerShell isn't more widespread.. You are making it difficult on everyone . In the following code, we're going to use PowerShell that won't make you hate PowerShell; it will be pleasant and the commands will ends up rolling off your fingers (something that won't happen with the horrible longhand command names). Linux users will feel at home with this syntax. VB-users will be wildly offended. EXCELLENT!

Flaw Workaround

Before we do anything, we have to talk about one of the greatest epic fails in recent software history...

While the Elasticsearch architecture is truly epic, it has a known design flaw (the absolute worst form of a bug): it allows a POST body in a GET request. This makes development painful:

  • Fiddler throws a huge red box at you.
  • wget in PowerShell gives you an error.
  • Postman in Chrome doesn't even try.
  • HttpClient in .NET throws System.Net.ProtocolViolationException saying "Cannot send a content-body with this verb-type."

.NET is right. Elasticsearch is wrong. Breaking the rules for the sake of what you personally feel makes you a vigilante. Forcing a square peg into a round hole just for the sake of making sure "gets" are done with GET makes you an extremist. Bodies belong in POST and PUT, not GET.

It's a pretty stupid problem having given how clever their overall architecture is. There's a an idiom for this in English: Homer Nodded.

To get around this flaw, instead of actually allowing us to search with POST (like normal people would), we are forced to use a hack: source query string parameter.

PowerShell Setup

When following along, you'll want to use the PowerShell ISE. Just type ISE in PowerShell.

PowerShell note: hit F5 in ISE to run a script

If you are going to run these in in a ps1 file, make sure to run Set-ExecutionPolicy RemoteSigned. as admin. Microsoft doesn't seem to like PowerShell at all. It's not the default in Windows 8/10. It's not the default on Windows Server. You can't run scripts by default. Someone needs to be fired. Run the aforementioned command to allow local scripts.

HTTP/S call

We're now ready to be awesome.

To start, let's create a call to ES. In the following code, I'm calling HTTPS with authorization. I'm not giving you the sissy Hello World, this is from production. While you're playing around, you can remove HTTP and authorization. You figure out how. That's part of learning.

$base = 'search.domain.net' 

$call = {
    param($params)

    $uri = "https://$base"

    $headers = @{ 
        'Authorization' = 'Basic fVmBDcxgYWpndYXJj3RpY3NlkZzY3awcmxhcN2Rj'
    }

    $response = $null
    $response = wget -Uri "$uri/$params" -method Get -Headers $headers -ContentType 'application/json'
    $response.Content
}

PowerShell note: prefix your : with ` or else you'll get a headache

So far, simple.

We can call &$call to call an HTTPS service with authorization.

But, let's break this out a bit...

$call = {
    param($verb, $params)

    $uri = "https://$base"

    $headers = @{ 
        'Authorization' = 'Basic fVmBDcxgYWpndYXJj3RpY3NlkZzY3awcmxhcN2Rj'
    }

    $response = wget -Uri "$uri/$params" -method $verb -Headers $headers -ContentType 'application/json'
    $response.Content
}

$get = {
    param($params)
    &$call "Get" $params
}

$delete = {
    param($params)
    &$call "Delete" $params
}

Better. Now we can call various verb functions.

To add PUT and POST, we need to account for the POST body. I'm also going to add some debug output to make life easier.

$call = {
    param($verb, $params, $body)

    $uri = "https://$base"

    $headers = @{ 
        'Authorization' = 'Basic fVmBDcxgYWpndYXJj3RpY3NlkZzY3awcmxhcN2Rj'
    }

    Write-Host "`nCalling [$uri/$params]" -f Green
    if($body) {
        if($body) {
            Write-Host "BODY`n--------------------------------------------`n$body`n--------------------------------------------`n" -f Green
        }
    }

    $response = wget -Uri "$uri/$params" -method $verb -Headers $headers -ContentType 'application/json' -Body $body
    $response.Content
}

$put = {
    param($params,  $body)
    &$call "Put" $params $body
}

$post = {
    param($params,  $body)
    &$call "Post" $params $body
}

In addition to having POST and PUT, we can also see what serialized data we are sending, and where.

ES Catalog Output

Now, let's use $call (or $get, etc) in something with some meaning:

$cat = {
    param($json)

    &$get "_cat/indices?v&pretty"
}

This will get the catalog of indexes.

Elasticsearch note: You can throw pretty anywhere to get formatted JSON.

Running &$cat gives me the following json:

[ {
  "health" : "yellow",
  "status" : "open",
  "index" : "site1!production",
  "pri" : "5",
  "rep" : "1",
  "docs.count" : "170",
  "docs.deleted" : "0",
  "store.size" : "2.4mb",
  "pri.store.size" : "2.4mb"
}, {
  "health" : "yellow",
  "status" : "open",
  "index" : "site2!production",
  "pri" : "5",
  "rep" : "1",
  "docs.count" : "141",
  "docs.deleted" : "0",
  "store.size" : "524.9kb",
  "pri.store.size" : "524.9kb"
} ]

But, we're in PowerShell; we can do better:

ConvertFrom-Json (&$cat) | ft

Output:

 health status index                 pri rep docs.count docs.deleted store.size pri.store.size
------ ------ -----                 --- --- ---------- ------------ ---------- --------------
yellow open   site1!staging         5   1   176        0            2.5mb      2.5mb         
yellow open   site2!staging         5   1   144        0            514.5kb    514.5kb    
yellow open   site1!production      5   1   170        0            2.4mb      2.4mb         
yellow open   site2!production      5   1   141        0            524.9kb    524.9kb     

Example note: !production and !staging have nothing to do with ES. It's something I do in ES, Redis, Mongo, SQL Server, and every other place the data will be stored to separate deployments. Normally I would remove this detail from article samples, but the following examples use this to demonstrate filtering.

PowerShell note: Use F8 to run a selection or single line. It might be worth removing your entire $virtualenv, if you want to play around with this.

Much nicer. Not only that, but we have the actual object we can use to filter on the client side. It's not just text.

(ConvertFrom-Json (&$cat)) `
    | where { $_.index -match '!production' }  `
    | select index, docs.count, store.size |  ft

Output:

index              docs.count store.size
-----              ---------- ----------
site1!production   170        2.4mb     
site2!production   141        532.9kb   

Getting Objects from ES

Let's move forward by adding our search function:

$search = {
    param($index, $json)

    &$get "$index/mydatatype/_search?pretty&source=$json"
}

Calling it...

&$search 'site2!production' '{
    "query": {
        "match_phrase": {
            "content": "struggling serves"
        }
    }
}'

Elasticsearch note match_phrase will match the entire literal phrase "struggling serves"; match would have search for "struggling" or "serves". Results will return with a score, sorted by that score; entries with both words would have a higher score than an entry with only one of them. Also, wildcard will allow stuff like `struggl*.

Meh, I'm not a big fan of this analogy of SELECT * FROM [site2!production]:

&$search 'site2!production' '{
    "query": {
        "match_phrase": {
            "content": "struggling serves"
        }
    },
    "fields": ["selector", "title"]
}'

This will return a bunch of JSON.

{
  "took" : 3,
  "timed_out" : false,
  "_shards" : {
    "total" : 5,
    "successful" : 5,
    "failed" : 0
  },
  "hits" : {
    "total" : 3,
    "max_score" : 0.6106973,
    "hits" : [ {
      "_index" : "site2!staging",
      "_type" : "entry",
      "_id" : "AVGUw_v5EFj4l3MNvkeV",
      "_score" : 0.6106973,
      "fields" : {
        "selector" : [ "sweet/mine" ]
      }
    }, {
      "_index" : "site2!staging",
      "_type" : "entry",
      "_id" : "AVGU3PwoEFj4l3MNvk9D",
      "_score" : 0.4333064,
      "fields" : {
        "selector" : [ "of/ambition" ]
      }
    }, {
      "_index" : "site2!staging",
      "_type" : "entry",
      "_id" : "AVGU3QHeEFj4l3MNvk9G",
      "_score" : 0.4333064,
      "fields" : {
        "selector" : [ "or/if" ]
      }
    } ]
  }
}

We can improve on this.

First, we can convert the input to something nicer:

&$search 'site2!production' (ConvertTo-Json @{
    query = @{
        match_phrase = @{
            content = "struggling serves"
        }
    }
    fields = @('selector', 'title')
})

Here we're just creating a dynamic object and serializing it. No JIL or Newtonsoft converters required.

To make this a a lot cleaner, here's a modified $search:

$search = {
    param($index, $json, $obj)
    if($obj) {
        $json = ConvertTo-Json -Depth 10 $obj
    }

   &$get "$index/mydatatype/_search?pretty&source=$json"
}

You need -Depth <Int32> because the default is 2. Nothing deeper than the default will serialize. It will simply show "System.Collections.Hashtable. In ES, you'll definitely have deep objects.

Now, I can call this with this:

&$search 'site2!production' -obj @{
    query = @{
        match_phrase = @{
            content = "struggling serves"
        }
    }
    fields = @('selector', 'title')
}

This works fine. Not only that, but the following code still work:

&$search 'site2!production' '{
    "query": {
        "match_phrase": {
            "content": "struggling serves"
        }
    }
    "fields" = ["selector", "title"]
}'

Now you don't have to fight with escaping strings; you can also still copy/paste JSON with no problem.

JSON to PowerShell Conversion Notes:

  • : becomes =
  • all ending commas go away
    • newlines denote new properties
  • @ before all new objects (e.g. {})
  • [] becomes @()
    • @() is PowerShell for array
  • " becomes ""
    • PowerShell escaping is double-doublequotes

DO NOT FORGET THE @ BEFORE {. If you do, it will sits there forever as it tries to serialize nothing into nothing. After a few minutes, you'll get hundreds of thousands of JSON entries. Seriously. I tries to serialize every aspect of every .NET property forever. This is why the -Depth defaults to 2.

Next, let's format the output:

(ConvertFrom-Json(&$search 'content!staging' 'entry' -obj @{
    query = @{
        match_phrase = @{
            content = "struggling serves"
        }
    }
    fields = @('selector', 'title')
})).hits.hits.fields | ft

Could probably just wrap this up:

$matchPhrase = {
    param($index, $type, $text, $fieldArray)
    (ConvertFrom-Json(&$search $index $type -obj @{
        query = @{
            match_phrase = @{
                content = $text
            }
        }
        fields = $fieldArray
    })).hits.hits.fields | ft
}

Just for completeness, here's $match. Nothing too shocking.

$match = {
    param($index, $type, $text, $fieldArray)
    (ConvertFrom-Json(&$search $index $type -obj @{
        query = @{
            match = @{
                content = $text
            }
        }
        fields = $fieldArray
    })).hits.hits.fields | ft
}

Finally, we have this:

&$match 'content!staging' 'entry' 'even' @('selector', 'title')

Output:

title                               selector           
-----                               --------           
{There Defeats Cursed Sinews}       {forestalled/knees}
{Foul Up Of World}                  {sweet/mine}       
{And As Rank Down}                  {crown/faults}     
{Inclination Confront Angels Stand} {or/if}            
{Turn For Effects World}            {of/ambition}      
{Repent Justice Is Defeats}         {white/bound}      
{Buys Itself Black I}               {form/prayer}

There we go: phenominal cosmic power in an ity bity living space

Beefier Examples and Objects

Here's an example of a search that's a bit beefier:

&$search  'bible!production' -obj @{
    query = @{
        filtered = @{
            query = @{
                match = @{
                    content = "river Egypt"
                }
            }
            filter = @{
                term = @{
                    "labels.key" = "isaiah"
                }
            }
        }
    }
    fields = @("passage")
    highlight = @{
        pre_tags = @("<span class=""search-result"">")
        post_tags = @("</span>")
        fields = @{
            content = @{
                fragment_size = 150
                number_of_fragments = 3
            }
        }
    }
}

Elasticsearch note: Filters are binary: have it or not? Queries are analog: they have a score. In this example, I'm moving a filter with a query. Here I'm searching the index for content containing "river" or "Egypt" where labels: { "key": 'isaiah' }

Using this I'm able to to filter my documents by label where my labels are complex objects like this:

  "labels": [
    {
      "key": "isaiah",
      "implicit": false,
      "metadata": []
    }
  ]

I'm able to search by labels.key to do a hierarchical filter. This isn't an ES tutorial; rather, this is to explain why "labels.key" was in quotes in my PowerShell, but nothing else is.

Design note: The objects you sent to ES should be something optimized for ES. This nested type example is somewhat contrived to demonstrate nesting. You can definitely just throw your data and ES and it will figure out the schema on the fly, but that just means you're lazy. You're probably the type of person who used the forbidden [KnownType] attribute in WCF because you were too lazy to write DTOs. Horrible. Go away.

This beefier example also shows me using ES highlighting. In short, it allows me to tell ES that I want a content summary of a certain size with some keywords wrapped in some specified HTML tags.

This content will show in addition to the requested fields.

The main reason I mention highlighting here is this:

When you serialize the object, it will look weird:

"pre_tags":  [
   "\u003cspan class=\"search-result\"\u003e"
]

Chill. It's fine. I freaked out at first too. Turns out ES can handle unicode just fine.

Let's run with this highlighting idea a bit by simplifying it, parameterizing it, and deserializing the result (like we've done already):

$result = ConvertFrom-Json(&$search  $index -obj @{
    query = @{
        filtered = @{
            query = @{
                match = @{
                    content = $word
                }
            }
        }
    }
    fields = @("selector", "title")
    highlight = @{
        pre_tags = @("<span class=""search-result"">")
        post_tags = @("</span>")
        fields = @{
            content = @{
                fragment_size = 150
                number_of_fragments = 3
            }
        }
    }
})

Nothing new so far. Same thing, just assigning it to a variable...

The JSON is passed back was this...

{
  "took" : 6,
  "timed_out" : false,
  "_shards" : {
    "total" : 5,
    "successful" : 5,
    "failed" : 0
  },
  "hits" : {
    "total" : 30,
    "max_score" : 0.093797974,
    "hits" : [ {
      "_index" : "content!staging",
      "_type" : "entry",
      "_id" : "AVGhy1octuYZuX6XP7zu",
      "_score" : 0.093797974,
      "fields" : {
        "title" : [ "Pause First Bosom The Oft" ],
        "selector" : [ "hath come/hand of" ]
      },
      "highlight" : {
        "content" : [ " as to fault <span class=\"search-result\">smells</span> ourselves am free wash not tho
se lies enough business eldest sharp first what corrupted blood which state knees wash our cursed oft", " give
 above shall curse be help faults offences this snow me pray shuffling confront ere forgive newborn a engaged 
<span class=\"search-result\">smells</span> rain state angels up form", " ambition eldest i guilt <span class=
\"search-result\">smells</span> forehead true snow thicker rain compelld intent bound my which currents our so
ul of limed angels white snow this buys" ]
      }
    },
    ...    

The data we want is is hits.hits.fields and hits.hits.highlights.

So, we can get them, and play with the output...

$hits = $result.hits.hits
$formatted = $hits | select `
        @{ Name='selector'; Expression={$_.fields.selector} },
        @{ Name='title'; Expression={$_.fields.title} }, 
        @{ Name='highlight'; Expression={$_.highlight.content} }
$formatted

This is basically the following...

hits.Select(p=>new { selector = p.fields.selector, ...});

Output:

selector                   title                            highlight                                        
--------                   -----                            ---------                                        
hath come/hand of          Pause First Bosom The Oft        { as to fault <span class="search-result">smel...
all fall/well thicker      Enough Nature Heaven Help Smells { buys me sweet queen <span class="search-resu...
with tis/law oft           Me And Even Those Business       { if what form this engaged wretched heavens m...
hand forehead/engaged when Form Confront Prize Oft Defeats  {white begin two-fold faults than that strong ...
newborn will/not or        Were Gilded Help Did Nature      { above we evidence still to me no where law o...
cursed thicker/as free     Cursed Tis Corrupted Guilt Where { justice wicked neglect by <span class="searc...
currents hand/true of      Both Than Serves Were May        { serves engaged down ambition to man it is we...
two-fold can/eldest queen  Then Sweet Intent Help Turn      { heart double than stubborn enough the begin ...
force did/neglect whereto  Compelld There Strings Like Not  { it oft sharp those action enough art rests s...
babe whereto/whereto is    As Currents Prayer That Free     { defeats form stand above up <span class="sea...

This part is important: highlight is an array. Your search terms may show up more than once in a document. That's where the whole number_of_fragments = 3 came in. The highlight size is from that fragment_size = 150. So, for each entry we have, we have a selector (basically an ID), a title, and up to three highlights up to 150-characters each.

Let's abstract this stuff before we go to the final data analysis step:

$loadHighlights = {
    param($index, $word)

    $result = ConvertFrom-Json(&$search  $index -obj @{
        query = @{
            filtered = @{
                query = @{
                    match = @{
                        content = $word
                    }
                }
            }
        }
        fields = @("selector", "title")
        highlight = @{
            pre_tags = @("<span class=""search-result"">")
            post_tags = @("</span>")
            fields = @{
                content = @{
                    fragment_size = 150
                    number_of_fragments = 3
                }
            }
        }
    })

    $hits = $result.hits.hits
    $hits | select `
            @{ Name='selector'; Expression={$_.fields.selector} },
            @{ Name='title'; Expression={$_.fields.title} }, 
            @{ Name='highlight'; Expression={$_.highlight.content} }
}

Now, we can run:

&$loadHighlights 'content!staging' 'smells'

Let's use this and analyze the results:

(&$loadHighlights 'content!staging' 'smells').foreach({
    Write-Host ("Selector: {0}" -f $_.selector)
    Write-Host ("Title: {0}`n" -f $_.title)
    $_.highlight | foreach {$i=0} {
        $i++
        Write-Host "$i $_"
    }
    Write-Host "`n`n"
})

Here's a small sampling of the results:

Selector: two-fold can/eldest queen
Title: Then Sweet Intent Help Turn

1  heart double than stubborn enough the begin and confront as <span class="search-result">smells</span> ourse
lves death stronger crown murder steel pray i though stubborn struggling come by
2  of forehead newborn mine above forgive limed offences bosom yet death come angels angels <span class="searc
h-result">smells</span> i sinews past that murder his bosom being look death
3  <span class="search-result">smells</span> where strong ill action mine foul heavens turn so compelld our to
 struggling pause force stubborn look forgive death then death try corrupted



Selector: force did/neglect whereto
Title: Compelld There Strings Like Not

1  it oft sharp those action enough art rests shove stand cannot rain bosom bosom give tis repentance try upon
t possessd my state itself lies <span class="search-result">smells</span> the
2  brothers blood white shove no stubborn than in ere <span class="search-result">smells</span> newborn art re
pentance as like though newborn will form upont pause oft struggling forehead help
3  shuffling serve lies <span class="search-result">smells</span> stand well queen well visage and free his pr
ayer lies that art ere a there law even by business confront offences may retain

We have the selector, the title, and the highlights with <span class="search-result">...</span> showing us where our terms were found.

Setup

Up to this point, I've assumed that you already have an ES setup. Setup is once, but playing around is continuous. So, I got the playing around out of the way first.

Now, I'll go back and talk about ES setup with PowerShell. This should GREATLY improve your ES development. Well, it's what helps me at least...

ES has a schema. It's not fully Platonic like SQL Server, nor is it fully Aristotelian like MongoDB. You can throw all kinds of things at ES and it will figure them out. This is what ES calls dynamic mapping. If you like the idea of digging through incredibly massive data dumps during debugging or passing impossibly huge datasets back and forth, then this might be the way to go (were you "that guy" who threw [KnownType] on your WCF objects? This is you. You have no self-respect.) On the other hand, if you are into light-weight objects, you're probably passing ES a nice tight JSON object anyway. In any case, want schema to be computed as you go? That's dynamic mapping. Want to define your schema and have ES ignore unknown properties? That's, well, disabling dynamic mapping.

Dynamic mapping ends up being similar to the lowest-common denominator ("LCD") schema like in Azure Table Storage: your schema might end up looking like a combination of all fields in all documents.

ES doesn't so much deal with "schema" in the abstract, but with concrete indexes and types.

No, that doesn't mean it's schemaless. That's insane. It means that index and the types are the schema.

In any case, in ES, you create indexes. These are like your tables. Your indexes will have metadata and various properties much like SQL Server metadata and columns. Properties have types just like SQL Server columns. Unlike SQL Server, there's also a concept of a type. Indexes can have multiple types.

Per the Elasticsearch: Definitive Guide, the type is little more than a "_type" property internally, thus types are almost like property keys in Azure Table Storage. This means that when searching, you're searching across all types, unless you specify the type as well. Again, this maps pretty closely to a property key in Azure Table Storage.

Creating an index

Creating an index with a type is a matter of calling POST / with your mapping JSON object. Our $createIndex function will be really simple:

$createIndex = {
    param($index, $json, $obj)
    if($obj) {
        $json = ConvertTo-Json -Depth 10 $obj
    }
    &$post $index $json
}

Thing don't get interesting until we call it:

&$createIndex 'content!staging' -obj @{
    mappings = @{
        entry = @{
            properties = @{
                selector = @{
                    type = "string"
                }
                title = @{
                    type = "string"
                }
                content = @{
                    type = "string"
                }
                created = @{
                    type = "date"
                    format = "YYYY-MM-DD"  
                }
                modified = @{
                    type = "date"                  
                }
            }
        }
    }
}

This creates an index called content!staging with a type called entry with five properties: selector, title, content, created, and modified.

The created property is there to demonstrate that fact that you can throw formats on properties. Normally, dates are UTC, but here I'm specifying that I don't even care about times when it comes to the create date.

With this created, we can see how ES sees our data. We do this by calling GET //_mapping:

$mapping = {
    param($index)
   &$get "$index/_mapping?pretty"
}

Now to call it...

&$mapping 'content!staging'

Adding Data

Now to throw some data at this index...

Once again, the PowerShell function is simple:

$add = {
    param($index, $type, $json, $obj)
    if($obj) {
        $json = ConvertTo-Json -Depth 10 $obj
    }
    &$post "$index/$type" $json
}

To add data, I'm going to use a trick I wrote about elsewhere

if (!([System.Management.Automation.PSTypeName]'_netfxharmonics.hamlet.Generator').Type) {
    Add-Type -Language CSharp -TypeDefinition '
        namespace _Netfxharmonics.Hamlet {
            public static class Generator {
                private static readonly string[] Words = "o my offence is rank it smells to heaven hath the primal eldest curse upont a brothers murder pray can i not though inclination be as sharp will stronger guilt defeats strong intent and like man double business bound stand in pause where shall first begin both neglect what if this cursed hand were thicker than itself with blood there rain enough sweet heavens wash white snow whereto serves mercy but confront visage of whats prayer two-fold force forestalled ere we come fall or pardond being down then ill look up fault past form serve turn forgive me foul that cannot since am still possessd those effects for which did crown mine own ambition queen may one retain corrupted currents world offences gilded shove by justice oft tis seen wicked prize buys out law so above no shuffling action lies his true nature ourselves compelld even teeth forehead our faults give evidence rests try repentance can yet when repent wretched state bosom black death limed soul struggling free art more engaged help angels make assay bow stubborn knees heart strings steel soft sinews newborn babe all well".Split('' '');
                private static readonly int Length = Words.Length;
                private static readonly System.Random Rand = new System.Random();

                public static string Run(int count, bool subsequent = false) {
                    return Words[Rand.Next(1, Length)] + (count == 1 ? "" : " " + Run(count - 1, true));
                }
            }
        }

    '
}

n Let's use $gen with $add to load up some data:

$ti = (Get-Culture).TextInfo
(1..30).foreach({
    &$add -index 'content!staging' -type 'entry' -obj @{
        selector = "{0}/{1}" -f ([_netfxharmonics.hamlet.Generator]::Run(1)), ([_netfxharmonics.hamlet.Generator]::Run(1))
        title = $ti.ToTitleCase([_netfxharmonics.hamlet.Generator]::Run(4))
        content = [_netfxharmonics.hamlet.Generator]::Run(400) + '.'
        created = [DateTime]::Now.ToString("yyyy-MM-dd")
        modified = [DateTime]::Now.ToUniversalTime().ToString("o")
    }
})

This runs fast; crank the sucker up to 3000 with a larger content size if you want. Remove "Write-Host" from $call for more speed.

Your output will look something like this

Calling [https://search.domain.net/content!staging/entry]
BODY
--------------------------------------------
{
    "selector":  "death/defeats",
    "title":  "Sinews Look Lies Rank",
    "created":  "2015-11-07",
    "content":  "Engaged shove evidence soul even stronger bosom bound form soul wicked oft compelld steel which turn prize yet stand prize.",
    "modified":  "2015-11-07T05:45:10.6943622Z"
}
--------------------------------------------

When a run one of the earlier searches...

&$match 'content!staging' 'entry' 'even'

...we get the following results:

selector        
--------        
{death/defeats}

Debugging

If stuff doesn't work, you need to figure out how to find out why; not simply find out why. So, brain-mode activated: wrong results? Are you getting them wrong or are they actually wrong? Can't insert? Did the data make it to the server at all? Did it make it there, but couldn't get inserted? Did it make it there, get inserted, but you were simply told that it didn't insert? Figure it out.

As far as simple helps, I'd recommend doing some type of dump:

$dump = {
    param($index, $type)
    &$get "$index/$type/_search?q=*:*&pretty"
}

This is a raw JSON dump. You might want to copy/paste somewhere for analysis, or play around in PowerShell:

(ConvertFrom-Json(&$dump 'content!staging' 'entry')).hits.hits

I'd recommend just using a text editor to look around instead of random faux PowerShell data-mining. Because, JSON and XML both absolutely perfectly human readable, you'll see what you need quick. Even then, there's no reason no to just type the actual link into your own browser:

http://10.1.60.3:9200/content!staging/entry/_search?q=*:*&pretty

I'd recommend the Pretty Beautiful Javascript extension for Chrome.

You can remove &pretty when using this Chrome extension.

Another thing I'd strongly recommend is having a JSON beautifier toggle for input JSON:

$pretty = $False

So you can do something like this:

$serialize = {
    param($obj)
    if(!$pretty) {
        $pretty = $false
    }
    if($pretty) {
        ConvertTo-Json -Depth 10 $obj;
    }
    else {
        ConvertTo-Json -Compress -Depth 10 $obj
    }
}

Instead of calling ConvertTo-Json in your other places, just call &$serialize.

$search = {
    param($index, $type, $json, $obj)
    if($obj) {
        $json = &$serialize $obj
    }
    &$get "$index/$type/_search?pretty&source=$json"
}

Remember, this is for input, not output. This is for the data going to the server.

You want this option because once you disable this, you can do this:

&$match 'content!staging' 'entry' 'struggling' @('selector', 'title')

To get this...

Calling [http://10.1.60.3:9200/content!staging/entry/_search?pretty&source={"fields":["selector","title"],"query":{"match":{"content":"struggling"}}}]

title                            selector         
-----                            --------         
{Those Knees Look State}         {own/heavens}    

Now you have a URL you can dump into your web browser. Also, you have a link to share with others.

Regarding logs, on Linux systems you can view error messages at the following location:

/var/log/elasticsearch/CLUSTERNAME.log

I like to keep a SSH connection open while watching the log closely:

tail -f /var/log/elasticsearch/david-content.log 

Your cluster name will be configured in the YAML config somewhere around /etc/elasticsearch/elasticsearch.yml.

Updating (and Redis integration)

Updating Elasticsearch objects ("documents") is interesting for two reasons, a good one and a weird one:

Good reason: documents are immutable. Updates involve marking the existing item as deleted and inserting a new document. This is exactly how SQL Server 2014 IMOLTP works. It's one secret of extreme efficiency. It's an excellent practice to follow.

Weird reason: you have to update to know the integer ID to update a document. It's highly efficient, which makes it, at worst, "weird"; not "bad". It once allowed updates based on custom fields, you'd have a potential perf hit. Key lookups are the fastest.

Prior to Elasticsearch 2.x, you could add something like { "_id": { "path": "selector" } } to tell ES that you want to use your "selector" field as your ID. This was deprecared in version 1.5 and removed in 2.x (yes, they are two separate things). Today, _id is immutable. So, when you see docs saying you can do this, check the version. It will probably be something like version 1.4. Compare the docs for _id in version 1.4 with version 2.1 to see what I mean.

When you make a call like the following example, an cryptic ID is generated:

POST http://10.1.60.3:9200/content!staging/entry

But, you can specify the integer:

POST http://10.1.60.3:9200/content!staging/entry/5

This is great, but nobody anywhere cares about integer IDs. These surrogate keys have absolutely no meaning to your document. How in the world could you possibly know how to update something? You have to know the ID.

If you have your own useful identifier, then good for you, just use the following:

POST http://10.1.60.3:9200/content!staging/entry/tacoburger

Yet, you can't use any type of slash. Soooo.... forget it. Since I usually use ES to store content linked from URLs, this is isn't going to fly. Besides, nobody wants to have to keep track of all various encodings you have to do to make your data clean. So, we need to add some normalization to our IDs both to make ES happy and to keep our usage simple.

So, OK, fine... have to save some type of surrogate key to key map somewhere. Where could I possibly save them? Elasticsearh IS. MY. DATABASE. I need something insanely efficient for key / value lookups, but that persists to disk. I need something easy to use on all platforms. It should also be non-experimental. It should be a time-tested system. Oh... right: Redis.

The marketing says that Redis is a "cache". Whatever that means. It's the job of marketing to either lie about products to trick people into buying stuff or to downplay stuff for the sake of a niche market. In reality, Redis is a key/value database. It's highly efficiently and works everywhere. It's perfect. Let's start making the awesome...

I'l all about doing things based on first-principles (people who can't do this laugh at people who can do this and accuse them of "not invented here syndrome"; jealous expresses it in many ways), but I'm here I'm going to use the Stackoverflow.Redis package. It seems to be pretty standard and it works pretty well. I'm running it in a few places. Create some VS2015 (or whatever) project and add the NuGet package. Or, go find it and download it. But... meh... that sounds like work. Use NuGet.. Now we're going to reference that DLL..

$setupTracking = {
    Add-Type -Path 'E:\_GIT\awesomeness\packages\StackExchange.Redis.1.0.488\lib\net45\StackExchange.Redis.dll'
    $cs = '10.1.60.2'
    $config = [StackExchange.Redis.ConfigurationOptions]::Parse($cs)
    $connection = [StackExchange.Redis.ConnectionMultiplexer]::Connect($config)
    $connection.GetDatabase()
}

Here I'm adding the assembly, creating my connection string, creating a connection, then getting the database.

Let's call this and set some type of relative global:

$redis = &$setupTracking

We need to go over a few things in Redis first:

Redis communicates over TCP. You sends commands to it and you get stuff back. The commands are assembler-looking codes like:

  • HGET
  • FLUSHALL
  • KEYS
  • GET
  • SET
  • INCR

When you use INCR, you are incrementing a counter. So...

INCR taco

That sets taco to 1.

INCR taco

Now it's 2.

We can get the value...

GET taco

The return value will be 2.

By the way, this is how you setup realtime counters on your website. You don't have to choose between database locking and eventual consistency. Use Redis.

Then there's the idea of a hash. You know, a dictionary-looking thingy.

So,

HSET chicken elephant "dude"

This sets elephant on the chicken hash to "dude".

HGET chicken elephant

This gets "dude". Shocking... I know.

HGETALL chicken

This dumps the entire chicken hash.

Weird names demonstrate that the name has nothing to do with the system and it forces you to think, thus remembering it better long-term.

To get all the values, do something like this:

KEYS *

When I say "all", I mean "all". Both the values that INCR and the values from HSET will show. This is a typical wildcard. You can do stuff like KEYS *name* just fine.

Naming note: Do whatever you want, but it's commmon to use names like "MASTERSCOPE:SCOPE#VARIABLE". My system already has a well defined internal naming system of Area!Environment, so in what follows we'll use "content!staging#counter" and "content!staging#Hlookup"

OK, that's enough to get started. Here's the plan: Because the integer IDs mean absolutely nothing to me, I'm going to treat them as an implemenation detail; more technically, as a surrogate key. My key is selector. I want to update via selector not some internal ID that means nothing to me.

To do this, I'll basically just emulate what Elasticsearch 1.4 did: specify what property I want as my key.

To this end, I need to add a new $lookupId function, plus update both $add and $delete:

$lookupId = {
    param($index, $selector)

    if($redis) {
        $id = [int]$redis.HashGet("$index#Hlookup", $selector)
    }
    if(!$id) {
        $id = 0
    }
    $id
}

$add = {
    param($index, $type, $json, $obj, $key)
    if($obj) {
        $json = &$serialize $obj
        if($key) {
            $keyValue = $obj[$key]
        }
    }
    
    if($redis -and $keyValue) {
        $id = &$lookupId $index $keyValue
        Write-Host "`$id is $id"
        if($id -eq 0) {
            $id = [int]$redis.StringIncrement("$index#counter")
            if($verbose) {
                Write-Host "Linking $keyValue to $id"
            }
            &$post "$index/$type/$id" $json
            [void]$redis.HashSet("$index#Hlookup", $keyValue, $id)
        }
        else {
            &$put "$index/$type/$id" $json
        }

    }
    else {
        &$post "$index/$type" $json
    }
}

$delete = {
    param($index)
    &$call "Delete" $index

    if($redis) {
        [void]$redis.KeyDelete("$index#counter")
        [void]$redis.KeyDelete("$index#Hlookup")
    }
}

When stuff doens't exist, you get some type of blank entity. I've never seen a null while using the Stackoverflow.Redis package, so that's something to celebrate. The values that Stackoverflow.Redis methods work with are RedisKey and RedisValue. There's not much to learn there though, since there are operators for many different conversions. You can work with strings just fine without needing to know about RedisKey and RedisValue.

So, if I'm sending it a key, key the key value from the object I sent in. If there is a key value and Redis is enabled and active, see if that key value is the ID of an existing item. That's a Redis lookup. Not there? OK, must be new, use Redis to generate a new ID and send that to Elasticsearch (POST $index/$type/$id). The ID was already there? That means the selector was already assigned a unique, sequential ID by Redis, use that for the update.

For now, POST works fine for an Elasticsearch update as well. Regardless, I'd recommend using PUT for update even though POST works. You never know when they'll enforce it.

Let's run a quick test:

$selectorArray = &$generate 'content!staging' 'entry' 2 -key 'selector'

($selectorArray).foreach({
    $selector = $_
    Write-Host ("ID for $selector is {0}" -f (&$lookupId 'content!staging' $selector))
})

Output:

ID for visage/is is 4
ID for if/blood is 5

I'm going to hope over to Chrome to see how my data looks:

http://10.1.60.3:9200/content!staging/_search?q=*:*

It's there...

{
    "_index": "content!staging",
    "_type": "entry",
    "_id": "4",
    "_score": 1,
    "_source": {
        "selector": "visage/is",
        "title": "Bound Fault Pray Or",
        "created": "2015-11-07",
        "content": "very long content omitted",
        "modified": "2015-11-07T22:24:23.0283870Z"
    }
}

Cool, ID is 4.

What about updating?

Let's try it...

$obj = @{
    selector = "visage/is"
    title = 'new title, same document'
    content = 'smaller content'
    created = [DateTime]::Now.ToString("yyyy-MM-dd")
    modified = [DateTime]::Now.ToUniversalTime().ToString("o")
}
&$add -index 'content!staging' -type 'entry' -obj $obj -key 'selector' > $null

Output:

{
    "_index": "content!staging",
    "_type": "entry",
    "_id": "4",
    "_score": 1,
    "_source": {
        "selector": "visage/is",
        "title": "new title, same document",
        "created": "2015-11-07",
        "content": "smaller content",
        "modified": "2015-11-07T23:11:58.4607963Z"
    }
}

Sweet. Now I can update via my own key (selector) and not have to ever touch Elasticsearch surrogate keys (_id).