We have some products that heavily rely on azure cognitive search, it is a good product, but gradually we got quite a lot of problems with it, including:
You can't scale up or scale down without deleting the whole instance.
You can't use pipeline to create/update indexes unless you call it using web api. Modifying/delete field in an index is also not straightforward.
No data replication between search instances.
No cross index query even in the same instance.
No case insensitive search
Suggestions for above have been sitting in Microsoft's suggestion site for years and nothing was every done to address it. I really have no idea when or ever Microsoft will bother to provide better service.
As a result, I am starting to look around for alternative products (looking at elastic search at the moment). Is there any product that supports search syntax translation that make the migration easier (so we don't have to break so many things)?
Related
Practical application:
I have several databases that need to be queried from a single search box. Some of them I have direct access to (they're SQL Server / MySQL), others I can only search via an API.
In an ideal world I would inject all of this data into Elasticsearch and use it to determine relevance. Unfortunately I don't have the resources locally to make that run efficiently. Elastic is taking over 400mb of RAM just while idling without adding any actual data or running queries. It looks like most people using Elasticsearch in production are running machines with 32GB - 64GB of RAM. My organization doesn't have access to anything near that powerful available for this project.
So my next idea is to query all the databases and connect to the API's when the user makes a search. Then I need to analyze the results, determine relevance, and return them to the user. I recognize that this is probably a terrible plan in terms of performance. I'm hoping to use memcached to make things more tolerable.
In my research for finding algorithms to determine relevance, I came across tf-idf. I'm looking to apply this to the results I get back from all the databases.
The actual question
My understanding of tf-idf is that after tokenizing every document in the corpus, you perform a term frequency analysis and then multiply it against the inverse document frequency for the words. The inverse document frequency is calculated by dividing the total document count by the the total number of documents with the term.
The problem with this is that if I'm pulling documents from an API, I don't know the true total number of documents in the corpus. I'm only ever pulling a subset, and based on the way those documents are being pulled they're naturally going to all of the terms in them. Can I still apply tf-idf to this by treating the pool of documents returned by these various sources as a single corpus? What's the best way to go about this?
Bonus question
If you have a suggestion for how to accomplish this without hacking together my own search solution or using Elasticsearch I'm all ears...
As you have noticed Elasticsearch is not built to run in memory constrained environments. If you want to use Elasticsearch, but can't set up a dedicated machine, you might consider using a hosted search solution (e.g. AWS Elasticsearch, Elastic Cloud, Algolia, etc.). Those solutions still cost though!
There are two great alternatives that require a bit more work (but not as much as writing your own search solution). Lucene is the actual Search Engine that Elasticsearch is written on top of. It does still load quite a bit of the underlying data structures into memory, so, depending on the size of the underlying data you want to index, it could still run out of memory. But, you should be able to fit quite a bit more data in a single Lucene index than in an entire Elasticsearch instance.
The other alternative, which I know slightly less about, is Sphinx. It is also a Search Engine. And it also allows you to specify how much memory to allocate for it to use. It stores the rest of the data on disk.
Azure Search service maxes out at 300GB of data. As of today, we've exceeded that. Our database table consists mainly of unstructured text from website news articles around the world.
Do we have any options at all here? We like Azure Search and have built our entire back-end infrastructure around it, but now we're dead in the water with being able to add any more documents to it. Does Azure Search allow compression on the documents?
Azure Search offers a variety of SKUs. The biggest one allows you to index up to 2.4 TB per service. You can find more details here.
Note, changing SKUs requires re-indexing the data.
We don't provide data compression. If you'd like to talk to Azure Search program managers about your capacity requirements, feel free to reach out to #liamca.
I am looking into Neo4j as a stripped-down document store. A key aspect of document storage is search, and I know Neo4j includes full text search via legacy indices provided by Lucene.
I would be very interested in hearing the limitations of Neo4j search capabilities in a distributed environment. Does it provide a distributed index? In what ways is it inferior to Solr or ElasticSearch? How far can I take it before I must install Solr?
-- EDIT --
We are trying to integrate two distinct search efforts. The first is standard text content search. For instance, using the Enron emails, we want to search for every email that matches "bananas" or "going to the store" and get those document bodies in response. This is where people often turn to Solr.
The second case is more complicated, we have attached a great deal of meta-data to each document. We may have decided that "these" emails were the result of late-night drunk-dialing. Now I want to search for all emails that may have been the result of late-night drunk-dialing. For this kind of meta-data, we believe a graph database is in order.
In a perfect world, I can use one platform to perform both queries. I appreciate that Neo4j (nor OrientDB, Arango, etc) are designed as full text search databases, but I'm trying to understand the limitations thereof.
In terms of volume, we are dealing at a very large scale with batch-style nightly updates. The data is content heavy, with some documents running into hundreds of pages of text, but mostly on the order of a page or two.
I once worked on a health social network where we needed some sort of search and connection search functionalities we first went on neo4j we were very impressed by the cypher query language we could get and express any request however when you throw there billion of nodes you start to pay the price and we started considering another graph db, this time we've made a lot of research, tests and OrientDB was clearly the winner, OrientDB is highly scalable but the thing is that you have to code by yourself, your "search algorithm" if you want to do some advanced things (what is the common point between this two nodes) otherwise you have the SQL like query language (i don't know/remember if he has a name) but you can do some interesting stuff with it
So in conclusion i would definitely go on OrientDB
Neo4j can provide a "distributed index" in the sense that the high availability cluster can make your index available on more than one machine, but I'm pretty sure that's not what you're after. Related to this issue is a different answer I wrote about graph partitioning, and what it takes to distribute a really large number of nodes/relationships across multiple machines. (It's not terribly simple)
Solr and Lucene do two different things (although Solr is built on top of Lucene). I think solr and neo4j are not comparable because they're trying to do completely different things. This site isn't about software recommendations so I can't tell you what you should use other than to say you should read up on solr and neo4j, and figure out which set of functionality you want. As far as I know, this is an exclusive decision as I'm not aware of people integrating solr with neo4j.
Your question is very difficult to answer, I'd recommend expanding on what you are trying to do and what you have tried, you'll probably get better responses.
I'm designing yet another "Find Objects near my location" web site and mobile app.
My requirements are:
Store up to 100k objects;
Query for objects that are close to the point (my location, city, etc). And other search criteria (like object type);
Display results on the Google Maps with smooth performance.
Let user filter objects by object time.
I'm thinking about using Google App Engine for this project.
Could You recommend what would be the best data storage option for this?
And couple of words about dynamic data loading strategy.
I kinda feel overwhelmed with options at the moment and looking for hints where should I continue my research.
Thanks a lot!
I'm going to to assume that you are using the datastore. I'm not familiar with Google Cloud SQL (which I believe aims to offer MySQL-like features in the cloud), so I can't speak if it can do geospatial queries.
I've been looking into the whole "get locations in proximity of a location" problem for a while now. I have some good and bad news for you, unfortunately.
The best way to do the proximity search in the Google Environment is via the Search Service (https://developers.google.com/appengine/docs/python/search/ or find the JAVA link ). Reason being is that it supports a "Geopoint Field", and allows you to query in such a way.
Ok, cool, so there is support, right? However, "A query is complex if its query string includes the name of a geopoint field or at least one OR or NOT boolean operator". The free quota for Complex Search Queries are 100/day. Per 10,000 queries, it costs 60 cents. Depending on your application, this may be an issue.
I'm not too familar with the Google Maps API you might be able to pull off something like this :(https://developers.google.com/maps/articles/phpsqlsearch_v3)
My current project/problem involves moving locations, and not "static" ones (stores, landmarks,etc). I've decided to go with Amazon's Dynamodb and they have a library which supports geospatial indexing : http://aws.amazon.com/about-aws/whats-new/2013/09/05/announcing-amazon-dynamodb-geospatial-indexing/
I'm lost in: Hadoop, Hbase, Lucene, Carrot2, Cloudera, Tika, ZooKeeper, Solr, Katta, Cascading, POI...
When you read about the one you can be often sure that each of the others tools is going to be mentioned.
I don't expect you to explain every tool to me - sure not. If you could help me to narrow this set for my particular scenario it would be great. So far I'm not sure which of the above will fit and it looks like (as always) there are more then one way of doing what's to be done.
The scenario is: 500GB - ~20 TB of documents stored in Hadoop. Text documents in multiple formats: email, doc, pdf, odt. Metadata about those documents stored in SQL db (sender, recipients, date, department etc.) Main source of documents will be ExchangeServer (emails and attachments), but not only. Now to the search: User needs to be able to do complex full-text searches over those documents. Basicaly he'll be presented with some search-config panel (java desktop application, not webapp) - he'll set date range, document types, senders/recipients, keywords etc. - fire the search and get the resulting list of the documents (and for each document info why its included in search results i.e. which keywords are found in document).
Which tools I should take into consideration and which not? The point is to develop such solution with only minimal required "glue"-code. I'm proficient in SQLdbs but quite uncomfortable with Apache-and-related technologies.
Basic workflow looks like this: ExchangeServer/other source -> conversion from doc/pdf/... -> deduplication -> Hadopp + SQL (metadata) -> build/update an index <- search through the docs (and do it fast) -> present search results
Thank you!
Going with solr is a good option. I have used it for similar scenario you described above. You can use solr for real huge data as its a distributed index server.
But to get the meta data about all of these documents formats you should be using some other tool. Basically your workflow will be this.
1) Use hadoop cluster to store data.
2) Extract data in hadoop cluster using map/redcue
3) Do document identification( identify document type)
4) Extract meta data from these document.
5) Index metadata in solr server, store other ingestion information in database
6) Solr server is distributed index server, so for each ingestion you could create a new shard or index.
7) When search is required search on all the indexs.
8) Solr supports all the complex searches , so you don't have to make your own search engine.
9) It also does paging for you as well.
We've done exactly this for some of our clients by using Solr as a "secondary indexer" to HBase. Updates to HBase are sent to Solr, and you can query against it. Typically folks start with HBase, and then graft search on. Sounds like you know from the get go that search is what you want, so you can probably embed the secondary indexing in from your pipeline that feeds HBase.
You may find though that just using Solr does everything you need.
Another project to look at is Lily, http://www.lilyproject.org/lily/index.html, which has already done the work of integrating Solr with a distributed database.
Also, I do not see why you would not want to use a browser for this application. You are describing exactly what faceted search is. While you certainly could set up a desktop app that communicates with the server (parses JSON) and displays the results in a thick client GUI, all of this work is already done for you in the browser. And, Solr comes with a free faceted search system out of the box: just follow along the tutorial.
Going with Solr (http://lucene.apache.org/solr) is a good solution, but be ready to have to deal with some non-obvious things. First is planning your indexes properly. Multiple terabytes of data will almost definitely need multiple shards on Solr for any level of reasonable performance and you'll be in charge of managing those yourself. It does provide distributed search (doing the queries off multiple shards), but that is only half the battle.
ElasticSearch (http://www.elasticsearch.org/) is another popular alternative, but i don't have much experience with it regarding scale. It uses the same Lucene engine so i'd expect the search feature-set to be similar.
Another type of solution is something like SenseiDB - open sourced from LinkedIn - which gives the full-text search functionality (also Lucene-based) as well as proven scale for large amounts of data:
http://senseidb.com
They've definitely done a lot of work on search over there and my casual use of it is pretty promising.
Assuming all your data is already in Hadoop, you could write some custom MR jobs that pull the data in a consistent schema-friendly format into SenseiDB. SenseiDB already provides a Hadoop MR indexer which you can look at.
The only caveat is it is a little more complex to setup, but will save you with the scaling issues many times over - especially around indexing performance and faceting functionality. It also provides clustering support if HA is important to you - which is still in Alpha for Solr (Solr 4.x is alpha atm).
Hope that helps and good luck!
Update:
I asked a friend who is more versed in ElasticSearch than me and it does have the advantage of clustering and rebalancing based on the # of machines and shards you have. This is a definite win over Solr - especially if you're dealing with TBs of data. The only downside is the current state of documentation on ElasticSearch leaves a lot to be desired.
As a side note, you can't say the documents are stored in Hadoop, they are stored in a distributed file system (most probably HDFS since you mentioned Hadoop).
Regarding searching/indexing: Lucene is the tool to use for your scenario. You can use it for both indexing and searching. It's a java library. There is also an associated project (called Solr) which allows you to access the indexing/searching system through WebServices. So you should also take a look at Solr as it allows the handling of different types of documents (Lucene puts the responsability of interpreting the document (PDF, Word, etc) on your shoulders but you, probably, can already do that)